• Backward-looking quality metrics measure the old route, not the next one. High margins, stable earnings, and strong returns can persist long after a profit pool has started migrating.

• The key question in disruption is not whether a moat is real, but whether it travels. Teece’s insight matters: value accrues to whoever owns the scarce complementary assets needed in the new structure.

• AI will not kill all incumbents or save all challengers; it will create dispersion. The winners will be companies whose trust, workflow ownership, data, distribution, or ecosystem become more valuable as the route changes.

For centuries, Venice was the highest-quality asset in Europe.

It had the location, the fleet, the merchants, the institutions, the financing networks, and the Arsenal, perhaps the most advanced industrial operation of its time. More importantly, it had the route. Goods moving between East and West passed through the Adriatic, and Venice sat at the tollbooth. Its wealth was not simply a function of what it produced; it was a function of where it stood.

A fifteenth-century quality screen would have loved Venice. Durable competitive position, recurring trade flows, pricing power, institutional depth, high returns on capital, and centuries of proof. This was not a speculative asset. It was not a turnaround. It was the definition of proven quality.

Then the Portuguese rounded Africa.

Venice did not suddenly become incompetent. Its merchants did not forget how to trade. Its ships did not stop floating. Its institutions did not collapse overnight. The problem was simpler and more devastating: the route moved.

Once that happened, the very things that made Venice great became less relevant. The Arsenal could keep building ships. The bankers could keep financing commerce. The politicians could keep managing the republic. But the flow of value no longer needed to pass through the Adriatic. Venice’s quality had not disappeared. It had been stranded.

This is the central problem with quality investing during disruption. The spreadsheet measures the city. It does not tell you where the route is going.

The Missing Half of the Quality Trap

In my earlier piece, The Quality Trap, I argued that traditional quality screens are often museum catalogs. They identify the winners of the last era with remarkable precision, then mistake that precision for foresight.

The argument was straightforward. Companies with high returns on equity, stable earnings, low leverage, strong margins, and pricing power often look safest at precisely the moment their profit pools have begun to migrate. Newspapers in 2000 were wonderful businesses on paper. IBM in 2010 looked far cleaner than Microsoft. Legacy consumer brands had historical margins, distribution advantages, and brand equity that had compounded for decades. The numbers were not wrong. They were simply describing an economic structure that was already changing.

I still think that argument is right. But it was incomplete.

If every incumbent with great historical metrics is suspect when the route changes, then the conclusion is too blunt. Walmart survived Amazon. The New York Times survived digital media. Microsoft survived the transition from desktop software to cloud computing. In fact, Microsoft is the complication hiding inside my own argument: in 2010 it was itself a legacy incumbent, apparently past its prime, and yet it became one of the defining winners of the next era.

That means the real question cannot be “incumbent or challenger?” It cannot even be “quality or disruption?” Those are lazy oppositions. The harder question is this: when the route changes, which advantages travel?

Some moats become islands. They remain real, but the traffic leaves. Other moats become bridges. They allow a company to cross from one profit pool to the next.

The investor’s job is not to admire the moat. It is to determine whether anyone still needs to cross it.

Teece and the Theory of the Route

The cleanest way to understand this distinction was laid out by David Teece in his 1986 paper, Profiting from Technological Innovation.

Teece’s insight was simple, powerful, and still underappreciated: innovation and value capture are not the same thing. The company that invents a technology does not necessarily earn the profits from that technology. Those profits often accrue to whoever owns the complementary assets required to commercialize it.

The classic example is EMI and the CT scanner. EMI helped invent one of the most important medical technologies of the twentieth century. It had the breakthrough. It had the intellectual property. It had the technological lead.

What it lacked was the surrounding system. Hospitals did not just need a machine. They needed sales support, training, maintenance, procurement confidence, servicing infrastructure, regulatory credibility, and trust. General Electric had those assets. EMI had the invention; GE had the complements. The technology mattered, but the complement determined who got paid.

This is the intellectual foundation for thinking about quality during disruption. When a new technology appears, the exciting thing is rarely the only valuable thing. Often it is not even the most valuable thing. The durable economics accrue to the bottleneck around the technology: the asset that remains scarce after the innovation spreads.

This is why “follow the profit pool” is necessary but not sufficient. You also have to ask who owns the scarce complement needed to capture that profit pool.

Sometimes the complement belongs to the innovator. Sometimes it belongs to the incumbent. Sometimes it belongs to neither. The answer has little to do with who is old, who is new, who invented the technology, or who has the cleanest historical metrics. It has everything to do with who controls the bottleneck in the new structure.

This explains why Walmart survived and Borders did not. Both faced e-commerce. Borders had stores, inventory expertise, and a retail brand. Walmart had procurement scale, logistics infrastructure, supplier leverage, local presence, and physical nodes that could be integrated into an omnichannel model. In one case, the assets were stranded. In the other, they could travel.

It also explains why The New York Times diverged from most newspapers. The internet destroyed the local newspaper bundle, but it increased the value of trusted national brands. In a world of infinite content, scarcity moved from distribution to trust. The Times owned an asset that became more valuable after the route changed.

This is the refinement to the quality trap. The question is not whether the old moat was real. It usually was. The question is whether that moat controls a scarce complement in the new structure.

Why Traditional Metrics Invert

The Sparkline piece on AI disruption is valuable because it gives empirical scaffolding to what otherwise sounds like a collection of anecdotes. (https://www.sparklinecapital.com/post/ai-disruption)

Its core finding is simple and important: traditional value metrics work reasonably well in insulated industries, but struggle badly in industries exposed to technological disruption. That should not be surprising, but it is still underappreciated. In stable industries, low multiples often reflect mispricing. In disrupted industries, low multiples often reflect the market recognizing that current earnings are not durable.

The mechanism is brutal. Price moves first. Fundamentals move later.

A company facing disruption can look cheap for years. Revenue does not collapse immediately. Customers take time to leave. Contracts roll off slowly. Management cuts costs. Buybacks support earnings per share. Margins can hold up even as the strategic position deteriorates. The multiple compresses against earnings that still look real. That is the value trap.

But the same logic applies to quality.

A quality trap happens when the metrics remain beautiful because the old route has not fully decayed. Return on equity, operating margin, free cash flow, and earnings stability are all lagging indicators. They are evidence of what the company has been able to extract from the previous structure. They are not, by themselves, evidence that the next structure will require the same company.

This is why backward-looking metrics do not merely fail during disruption; they can invert. They select for the companies that look best under the old regime. But in a route change, the best-looking old-regime companies may be the most exposed. They are optimized for a path customers are abandoning.

That does not mean traditional quality is useless. It means its usefulness is conditional. In a stable structure, quality metrics are evidence. In a changing structure, they are questions.

Why are margins high? Because the company has a durable advantage, or because it is underinvesting? Why is free cash flow strong? Because the business compounds internally, or because management has no better use for capital? Why is leverage low? Because the company is disciplined, or because it is harvesting? Why are returns on capital high? Because the moat is widening, or because the denominator is being starved?

The same number can mean opposite things depending on whether the route is stable or moving.

Sparkline’s “intangible value” framework is interesting for precisely this reason. I would not treat it as a magic formula; any proprietary backtest deserves skepticism. But the direction is right. It attempts to measure the assets that Teece would have told us to look for: intellectual property, brand, human capital, and network effects. In other words, it tries to identify the complements that traditional accounting either misses, expenses, or misclassifies.

That is the real lesson. The answer to disruption is not to abandon valuation or quality. It is to redefine what must be valued.

Static, Stranded, and Traveling Quality

There are three kinds of quality during disruption.

Static quality is the easiest to identify. The business has great historical metrics. It screens well. It has high margins, stable earnings, low leverage, and a long record of profitability. Static quality is not fake; it is simply incomplete. It tells you the company was advantaged in the structure that produced the data.

Stranded quality is more dangerous. This is the business whose moat remains real but is attached to the old route. The company may still have brand, scale, distribution, assets, and customer relationships. The problem is that the customer’s route to value is changing in a way that makes those assets less central. The moat protects something, but not the bottleneck.

Traveling quality is what investors should be looking for. These are companies whose advantages become useful in the new structure. They may be incumbents or challengers. They may screen well or badly. The defining feature is that their scarce assets migrate with the profit pool.

This framework is more useful than the standard debate between quality and growth. Growth investors can overpay for companies that have the technology but not the complement. Quality investors can overpay for companies that have the complement to an old world. Value investors can buy optically cheap companies whose earnings are about to be structurally impaired.

The right question is more specific: what remains scarce after the technology spreads?

If the answer is “nothing the company owns,” the moat is stranded. If the answer is “the company’s trust, data, distribution, workflow, regulatory position, ecosystem, or implementation capability,” the moat may travel.

Software Is the Test

The software selloff is the cleanest current test of this framework.

The market’s fear is understandable. If AI makes software dramatically cheaper to build, then the economics of software must be re-examined. A product that once required large engineering teams may be built by a smaller team with AI assistance. Features that took quarters to develop may take weeks. Internal tools that companies previously bought may become easier to generate. The scarcity of code is falling.

If code was the moat, this is a problem.

This is why treating beaten-down software as a single bargain basket is dangerous. Some software companies are cheap because the market is panicking. Others are cheap because their business model depended on a scarcity that AI is dissolving.

The distinction is not obvious from the multiple. A software company trading at a low forward earnings multiple may be a bargain or a trap. The question is what the customer was really paying for.

Some software is vulnerable because it is a feature bundle. It automates a narrow task, has shallow workflow integration, limited proprietary data, weak switching costs, and pricing linked to work AI can now compress. These products may still be useful, but usefulness is not the same as pricing power. If AI makes alternatives easier to build, the product becomes less scarce.

Other software is different. It is a system of record. It sits inside a company’s daily operations. It carries permissions, history, approvals, audit trails, integrations, reporting, compliance, customer data, and employee training. The software is not merely a tool; it is a place where work happens.

For these companies, AI may be disruptive, but not necessarily destructive. If the vendor owns the workflow, AI can make the workflow more valuable. If the vendor owns the data context, AI can make that context more useful. If the vendor has trust in a regulated or mission-critical environment, AI can increase the premium on reliability. If the product is deeply integrated, AI features from the incumbent may be adopted faster than standalone alternatives.

This is the Teece question applied to software: if AI makes code abundant, what is the scarce complement?

For some companies, there may not be one. For others, the answer may be workflow ownership, proprietary data, compliance, trust, implementation capability, ecosystem depth, or customer context.

This leads to a practical test: if the cost of creating software fell by 90%, would the customer still choose this vendor?

If the answer is no, the moat was probably software scarcity. If the answer is yes, the moat is somewhere else.

That is why AI may create enormous dispersion within software. It will compress some companies and strengthen others. It will reveal which products were merely expensive because building software used to be hard, and which products were expensive because they occupied scarce positions inside customer operations.

The market’s first reaction is to sell the category. The investor’s job is to sort the category.

Adoption Versus Exposure

One of the more interesting points in Sparkline’s work is that software is not merely exposed to AI. It is also adopting AI aggressively.

That distinction matters. Exposure tells you where the threat is. Adoption tells you who is responding.

A company facing disruption has three choices. It can deny the shift, harvest the old model, or cannibalize itself. The third path is the only one that produces traveling quality, but it is also the hardest to recognize in traditional financial statements. It often depresses margins, increases expenses, and creates uncertainty before it creates proof.

This is why capital allocation matters more than commentary. Every management team says it is embracing AI. The question is whether the business is being rebuilt around it. Where is the company hiring? What is being integrated into the core product? Is AI a feature, a pricing lever, or a restructuring of the workflow? Is the company willing to weaken a legacy revenue pool to strengthen its position in the new one?

Microsoft’s cloud transition worked because it was not a press release. It was capital allocation. Azure absorbed investment before it produced the financial profile investors later celebrated. The old company funded the new route.

That is the pattern to look for today. The winners in software will not be the companies that mention AI the most. They will be the companies whose existing complements, data, workflow, trust, distribution, developer ecosystems, customer relationships, make their AI adoption more valuable than a competitor’s demo.

The losers will not necessarily be the companies that lack AI. Some will have AI features. Some will have impressive demos. The issue will be that their complements are weak. They will have product but not position.

Where the Route Goes

Venice’s tragedy was not incompetence. That is what makes the story useful. The city did not fail because it forgot how to be Venice. It failed because being Venice mattered less once the route changed.

That is the risk in every quality portfolio today. The businesses may be well managed. The margins may be real. The returns may be high. The balance sheets may be clean. The moats may even be deep.

But if the route is moving, the key question is not how good the city looks. The question is whether the route still runs through it.

That is the refinement to the quality trap. Backward-looking metrics fail because they measure the quality of the old route. The investor’s job is to determine whether the company’s moat is stranded on that route or whether it owns the scarce complement required on the new one.

Some companies will be Venice: excellent, historic, and bypassed. Some will be Walmart: attacked by the new route but able to bend its assets toward it. Some will be Microsoft: written off as a legacy incumbent while quietly owning the relationships, talent, and infrastructure needed for the next platform shift. Some will be EMI: inventors of the future without the complements needed to profit from it.

AI will create all four.

It will create companies with breathtaking technology and weak value capture. It will create incumbents with beautiful financials and stranded moats. It will create apparent bargains whose earnings have not yet caught down to reality. It will also create survivors whose old advantages become more valuable when carried into the new structure.

The old quality question was: is this a great business?

The better question is: will the things that made it great still matter where the profit pool is going?

That is the difference between static quality and traveling quality. Static quality tells you the city was rich. Traveling quality tells you the route still passes through it.

The spreadsheets will be the last to know.

Subscribe now

• The proof may not come as a software revenue line: We were waiting for Johansson to disclose Saab’s software/integration economics, but Q1 suggests the platform is first becoming visible in product architecture, especially C-UAS, where sensors, command-and-control, effectors, and upgrades are sold as one integrated system.

• Saab’s real flywheel is sovereign architecture: The bull case is not simply that Saab sells more defence equipment; it is that each radar, C2 layer, Gripen system, naval combat system, or counter-drone deployment makes the next Saab product easier to adopt by deepening training, doctrine, interoperability, maintenance, and political switching costs.

• Q1 strengthened the base case, not the victory lap: Organic growth, EBIT leverage, Surveillance strength, and operational cash flow all support the thesis, but free cash flow was still negative, Aeronautics remains a drag, and software disclosure is still missing, so Saab remains an “own and verify” stock, not a thesis fully proven.

I ended the last Saab piece waiting for a number.

Not a revenue number. Not an EBIT number. Not another backlog number. Saab already had those. I was waiting for the number that would either vindicate or embarrass the thesis we had been building for a year: how much of Saab is really software, systems integration, electronic warfare, data links, combat management, AI, and upgrade economics hiding inside a company still reported like a traditional defence contractor?

Q1 did not give us that number.

That should have been a problem. Instead, it made the story more interesting.

Our prior Saab arc was straightforward. Saab was no longer just a Swedish defence contractor. It was becoming Europe’s “sovereign stack”: a company whose value came not merely from selling fighters, radars, missiles, or submarines, but from connecting those systems into a NATO-compatible architecture that countries could operate and upgrade with greater independence. The missing proof point was disclosure. At some point, we expected Micael Johansson to point to a slide and say: here is the software layer, here are the margins, here is why Saab should not be valued like a normal defence contractor.

He still has not said it.

There are two ways to respond to that. The easy way is to say the thesis remains unproven. The better way is to ask whether we were looking for proof in the wrong place.

What If Software Is Not a Segment?

The mistake may be importing a software-company disclosure expectation into a defence architecture business.

Microsoft can report Azure because Azure is a product line. Adobe can report subscription revenue because the business model makes the software line visible. Saab is different. Its software is not consumed separately by a procurement officer with a credit card and a dashboard. It sits inside radars, fighters, electronic-warfare systems, combat-management systems, counter-drone architectures, naval platforms, training routines, command structures, and national doctrine.

The customer is not buying “software.” The customer is buying the ability to operate a defence system that can sense, decide, adapt, and remain under sovereign control.

That is why the software number still matters, but may not be the first place the truth appears. If Saab is right, the economics will show up first in how customers buy: not isolated equipment, but architectures. Not a radar, then a separate command system, then a separate effector, but an integrated chain from detection to decision to action. Not one national project, but a series of adjacent procurements that become easier because the country is already inside the Saab architecture.

The question after Q1, then, is not “where is the software disclosure?” It is: are customers behaving as if Saab’s integration layer is becoming the product?

Q1’s answer was not definitive. But it was suggestive.

The SEK 2.6 Billion Order That Matters More Than Its Size

The most important order in Q1 was not large enough to move the model.

Sweden’s Counter-UAS order was worth SEK 2.6 billion, with deliveries planned for 2027-2028. In a company with a SEK 274 billion backlog, that number is not the point. The architecture is the point.

Saab described the system as an integrated Counter-UAS solution, from “sensor to shooter,” modular, adaptable to evolving drone threats, built with Saab and selected third-party technologies, and including further development and integration.

That is the sovereign stack in miniature.

A drone appears. A sensor detects it. A command system classifies it. An effector is selected. The architecture adapts as the threat evolves. The customer does not simply buy a radar, a cannon, a missile, or a vehicle. The customer buys the system that makes all of those pieces work together.

On the call, Johansson described the system as flexible and effector-agnostic — able to work with different effectors, from Bofors rounds to Trackfire and eventually Nimbrix. That is the right kind of agnostic. Saab does not need to own every fist if it owns the brain that decides where the fist lands.

This is where Saab differs from the usual European rearmament story. Rheinmetall’s question is capacity: can it produce enough shells, vehicles, and land systems into a shortage? Saab’s question is architecture: can it become the integration layer for countries that want NATO-compatible capability without total dependence on any one superpower or prime contractor?

Capacity still matters. But for Saab, capacity is the cost of the thesis, not the thesis itself.

The Flywheel Is Boring, Which Is Why It Works

Defence switching costs are not created by delight. They are created by doctrine.

A country buys a Saab radar. That radar connects into Saab command-and-control. The next counter-drone system is easier to buy from Saab because the sensor layer already exists. The next naval system is easier if the combat-management layer is trusted. The next aircraft, electronic-warfare package, or upgrade becomes less like a fresh procurement and more like an extension of the architecture already in place.

This is not a social network effect. No one is posting selfies on Giraffe 1X. It is slower, more political, more bureaucratic, and much harder to see. But in defence, those features can make it more durable.

Training compounds. Maintenance routines compound. Data links compound. Local industrial participation compounds. Software upgrade paths compound. Operational trust compounds. The switching costs are not just financial; they are institutional.

That is why Q1’s Giraffe 1X commentary mattered. Johansson said Saab had manufactured some Giraffe 1X systems partly “on speculation,” rather than only waiting for signed contracts, which enabled fast deliveries to customers facing urgent missile and drone threats. In a bespoke defence world, that is an unusual posture. It suggests some Saab systems are becoming repeatable nodes in an architecture, not one-off national projects that begin from zero every time.

Naval points in the same direction. Saab’s new Naval business area, formed after the quarter, combines Kockums with Naval Combat Systems. That is not just reorganization. It is strategy made visible: platforms and combat systems belong together if the value is integration.

The Quarter After the Fireworks

Only now do the Q1 numbers matter.

Q4 was the fireworks quarter: Colombia Gripen, France GlobalEye, Poland submarines selected, Ukraine letter of intent, record orders, and a backlog that made the sovereign-stack thesis feel almost too neat. Q1 was quieter. Order bookings fell 5% to SEK 18.2 billion, and book-to-bill was below 1x. A sell-side recap would start there.

If this were merely a defence-cycle story, that would matter more. If it is an architecture story, the more interesting fact is that the system kept moving without another megadeal.

Sales rose 21% to SEK 19.2 billion, organic growth was 23.6%, EBIT rose 32%, and operational cash flow improved to SEK 1.0 billion. All business areas and Combitech delivered double-digit sales growth. Surveillance grew sales 32% and EBIT 52%, helped by Giraffe 1X deliveries and C-UAS. Dynamics remained the profit engine, with a 17.5% margin.

That is what we needed to see. Not perfection. Repeatability.

Saab did not need a giant new order to grow faster than 20%. It did not need a software disclosure slide to show operating leverage. It did not need a fighter headline for Surveillance to look increasingly like the place where sensors, software, command-and-control, and threat adaptation collapse into one business.

This is what I mean by the architecture becoming observable.

The Part That Still Bothers Me

This is where I do not want to let the thesis get too elegant.

Free cash flow was still negative SEK 301 million. Inventories rose. Tangible fixed assets rose. Saab is investing heavily in capacity, inventory, suppliers, and facilities because architecture still has to be manufactured.

Some of that is exactly what we want. If speed wins defence orders, and if customers need systems now, then carrying inventory can be strategic. The Giraffe 1X example suggests Saab is willing to take working-capital risk where it has demand visibility.

But there is a line between strategic inventory and bad inventory, and the income statement will not tell you where that line is. Cash will.

Saab’s medium-term target is cumulative cash conversion above 60%. Q1 moved in the right direction, but one quarter does not answer the funding question. If Saab’s architecture story requires too much capital to scale, the platform multiple will have a ceiling.

Aeronautics is the second constraint. Sales grew 15%, but EBIT fell 11%, and the margin compressed to 6.3%, hurt by R&D amortization, currency, and T-7 start-up costs. Gripen is central to the sovereignty narrative, but aircraft programs remain slow, complex, and capital-intensive. The base case can tolerate that. The bull case cannot ignore it.

What Changed in My View

Before Q1, I thought software disclosure was the next decisive gate. After Q1, I still think it matters, but I no longer think it is the only gate.

The cleanest proof would be a number. The earlier proof may be customer behaviour.

If C-UAS exports follow Giraffe 1X installations, if Naval shows better economics after combining platforms with combat systems, if Surveillance sustains margins above 10%, if countries that buy one Saab node return for adjacent nodes, then the architecture thesis will be proving itself before accounting catches up.

That is the update. We were waiting for Saab to separate software from hardware. Q1 suggests customers may increasingly be buying the place where software and hardware become inseparable.

Variant Perception

The consensus view is reasonable: Saab is an excellent European defence company, growing rapidly, but priced for a lot of success. Order intake was down, free cash flow was negative, Aeronautics disappointed, and the software number still does not exist.

The variant perception is not to dismiss those issues. It is to say they are not the central issue.

The central issue is whether Saab is becoming an architecture layer. If it is, every system sold increases the probability that the next system is also Saab. A radar creates pull-through for C2. C2 creates pull-through for counter-drone. Counter-drone creates pull-through for effectors. Gripen creates decades of training, maintenance, upgrades, weapons integration, and electronic-warfare dependence. Naval combat systems can do the same at sea.

The bull case is not that Saab sells more equipment. The bull case is that each piece of equipment makes the next piece easier to sell.

Three-Year Scenarios

Using a reference price around SEK 514, the setup is attractive but not obvious.

I would not raise the bull probability yet. Q1 did not give us software disclosure, a major new platform award, or clean free cash flow. What it did give us was a stronger base case and a lower probability that Q4 was merely order-book euphoria.

This remains an “own and verify” stock. The expected value is attractive, but the proof points still matter.

What To Watch

The first proof point is C-UAS exports. Sweden proves the concept; export orders prove the architecture.

The second is Surveillance margin. Above 10% sustained suggests integration economics are scaling. Below 8.5% suggests Q1 was mix.

The third is Naval margin from Q2 onward. The reorganization only matters if platforms plus combat systems produce better economics than platforms alone.

The fourth is Aeronautics margin. Above 8% would show Gripen economics surfacing. Below 6% would mean T-7 and program drag are still eating the story.

The fifth is cash conversion. Above 60% keeps the thesis self-funded. Below 50% turns the platform story into a capital-intensity problem.

And yes, the sixth is still software disclosure. It is no longer the only proof. It remains the cleanest one.

The Number Can Wait. The Architecture Can’t.

Q1 did not answer the question we asked. It suggested a better question.

We wanted Saab to prove the software layer by disclosing it. Instead, Q1 showed the software layer becoming visible in the architecture customers are buying: Giraffe 1X into C-UAS, C-UAS into command-and-control, Naval into combat systems, Surveillance into the integration layer.

That is not yet disclosed software economics. It is not durable free cash flow. It is not permission to ignore valuation.

But it is the right kind of evidence.

Saab is no longer trying to prove that countries want sovereignty. Q4 did that. Q1 asked whether sovereignty compounds when the headlines go quiet.

The answer is not final. But it is getting harder to dismiss.

Subscribe now

• The thesis moved from funding to throughput. Q4 proved sovereign customers would pre-fund Rheinmetall’s capacity; Q1 asks whether that funded capacity can turn into accepted deliveries, cash flow, and delivery credibility fast enough.

• The moat is intact, but the governor is visible. Demand and funding are not the issue; the bottleneck is the conversion layer, customer acceptance, lot testing, procurement timing, inventory-to-revenue mechanics, and management bandwidth across Naval, Air Defence, Digital, autonomy, and deep strike.

• At ~€1,207, the price finally makes the debate investable. The stock is no longer pricing most of the bull case, but Q2 is the proof point: if inventory converts and Q2 revenue validates management’s >50% growth claim, the flywheel works; if not, the “sovereign infrastructure” thesis starts looking like industrial overreach.

Rheinmetall reported Q1 sales of €1.94 billion, operating result of €224 million, an 11.6% operating margin, operating free cash flow of -€285 million, Rheinmetall Nomination of €4.87 billion, and backlog of €73.0 billion. That is the earnings recap. It is also the least useful way to understand the quarter.

The right place to start is with our own evolution. In Q3, we worried that Rheinmetall’s story was running into industrial physics: backlog was enormous, but cash burn and ammunition margin pressure suggested that being indispensable did not automatically mean being cash-generative. In Q4, we changed our view. The important discovery was not that free cash flow improved; it was how it improved. Sovereign customers were prepaying because they needed Rheinmetall’s capacity more than Rheinmetall needed incremental orders. We upgraded the moat from regulatory entrenchment to sovereign dependency: governments were financing the industrial capacity they could not replace.

Q1 does not reverse that conclusion. It makes it more precise. The funding debate is no longer the center of the thesis. The throughput debate is.

Q4 proved Rheinmetall’s customers would fund the moat. Q1 asks whether Rheinmetall can operate it.

The Factory Outran the Bureaucracy

The sell-side version of Q1 is easy: sales growth disappointed, cash flow was negative, and the company promised a stronger Q2. That is true, but too shallow.

Management’s explanation was unusually specific. Around €200 million of trucks were ready but awaiting German customer delivery timing; another €100 million of powder-related revenue slipped because lot acceptance testing pushed delivery into Q2. Papperger therefore guided to Q2 growth of more than 50%.

That is the key point: not the miss, but the specificity of the test. A vague “second-half recovery” is management asking for faith. “More than 50% growth in Q2” is management putting a date on credibility.

The quarter’s core tension is that the factory appears to be scaling faster than the conversion layer around it: customer delivery windows, acceptance tests, procurement procedures, and inventory-to-revenue mechanics. CFO Klaus Neumann said roughly €0.5 billion of inventory went into finished and pre-finished goods ready for Q2 shipment; almost half of the €1 billion inventory increase was tied to goods prepared for Q2.

If that inventory ships, Q1 was the balance-sheet shadow of growth. If it does not, Q1 was the first warning that the capacity flywheel has a governor.

The Mechanism Is Intact, Not Yet Confirmed

The fundamental question is whether Rheinmetall can turn customer-funded capacity into a durable sovereign infrastructure role. Europe has moved from a just-in-time defense model to a just-in-case security model. In the old world, budgets were the bottleneck. In the new world, budgets are increasingly visible; the bottleneck is deliverable industrial capacity.

Rheinmetall’s core capability is not “making tanks” or “making shells.” It is capacity credibility: powder, ammunition, propulsion, vehicles, air defense, digital systems, autonomy, and now naval, assembled into something governments can actually procure, deploy, and maintain.

That mechanism compounds if backlog and prepayments fund capacity, capacity improves delivery credibility, credibility wins larger integrated orders, and those orders deepen Rheinmetall’s role as sovereign defense infrastructure. It breaks if inventory does not convert, if acceptance delays become structural, if Weapon & Ammunition margins normalize faster than volume scales, or if Naval and Digital absorb capital without proving platform economics.

Q1 gave us evidence that the mechanism is still working. It did not prove it.

Operating profit grew 17% on 8% sales growth. Vehicle Systems grew only 3%, but margin improved to 9.6%. Weapon & Ammunition was flat on sales, but held 19.4% margin. Digital Systems grew 16%, with margin rising to 5.2%. Air Defence was the standout: more than 30% growth, with margin expanding from 12.5% to 15.6%. Naval contributed its first month at €77 million of sales and roughly 10% profitability.

That is not a broken quarter. It is an unfinished quarter.

The Integration Layer

The most important strategic development in Q1 was not in the income statement. It was in the widening of the architecture.

Naval closed in February. Management said the first month was in line with expectations, at about €80 million of sales and 10% EBIT, with profitability expected to move over time toward 15%. F125, F126, and Romania are the first tests of whether this becomes a real Sea domain or merely an expensive shipyard acquisition.

The same logic applies to autonomy and deep strike. Rheinmetall is not trying to invent everything internally. It is using partnerships as time arbitrage: Boeing for Ghost Bat, Lockheed for CCA cooperation, Indra for Spain, Kraken for unmanned surface vessels, Destinus for cruise missiles and ballistic rocket artillery. Papperger’s point was explicit: partnerships are faster than development programs, and Germany does not have time to wait for organic development of every new capability.

Viewed individually, each initiative can be justified. Viewed collectively, they reveal both the bull case and the risk. Rheinmetall is trying to become the industrial integration layer across vehicles, ammunition, air defense, ships, drones, missiles, sensors, and command systems. That is a better business than a product supplier. It is also a harder business to execute.

The company is becoming more important because it is becoming broader. It is becoming riskier for the same reason.

Variant Perception: Rearmament Is Not the Edge

The market already understands that European defense spending is going up. That is not the variant perception.

The better question is whether Rheinmetall is becoming the sovereign infrastructure layer that converts budgets into capability. The loop is:

backlog → prepayments and inventory → shipments → cash → capacity → delivery credibility → larger integrated orders → more backlog.

Q1 matters because it tests the middle of the loop. Demand is visible. Funding was validated in Q4. Now comes the harder-to-model part: physical throughput.

At €1,542 after Q4, we argued that the business had improved but the entry had not. The ceiling rose, but much of the upgraded thesis was priced. At roughly €1,207 now, the setup is different. The stock is no longer asking investors to pay for most of the bull case. It is closer to paying for the base case while preserving upside if Q2 proves the governor is bureaucratic rather than structural.

That does not mean the stock is obvious. It means the debate is finally worth having.

Three Years, Three Outcomes

By 2028, this should be clear. Capacity will either translate into output or it will not. Digital will either show platform economics or remain project revenue. Naval will either integrate or absorb management time. Arminius, ammunition, Air Defence, and Middle East demand will either convert or stay in procurement fog.

Assuming a current price around €1,207:

The probability-weighted value lands roughly in the high-€1,700s to low-€1,800s before dividends. That is a materially better setup than at €1,542. But it is still not a sleepy value stock. You are being compensated for execution risk, not spared from it.

What Has to Happen Next

The thesis is now unusually measurable.

First, Q2 revenue needs to validate management’s timing explanation. Above €3.2 billion would be a strong signal; below €2.8 billion would suggest the governor is more structural than bureaucratic. Second, operating free cash flow must improve in Q2 and H2; inventory has to be a bridge, not a trap. Third, Weapon & Ammunition margins need a clean baseline under the new structure; 28%+ would restore confidence, while below 25% would reopen the Q3 concern. Fourth, Arminius language matters: any move from Q2/Q3 to 2027 raises the bear probability. Fifth, Digital Systems needs at least one platform KPI by Q3, software or services mix, TaWAN attach rate, D-LBO milestone, lifecycle revenue, because without it the bull multiple is narrative, not evidence. Sixth, Naval margins above 8% through the year would suggest the acquisition works; below 5% would suggest shipyards are doing what shipyards often do.

What Changed

Our view did not reverse after Q1. It became more precise.

Q3 warned that backlog could run into industrial physics. Q4 showed sovereign customers would finance the capacity they could not replace. Q1 now asks the next and harder question: can Rheinmetall turn funded capacity into delivered capability fast enough to make the dependency permanent?

Rheinmetall’s problem is no longer finding demand. It is no longer, at least for now, finding funding. The problem is throughput.

Throughput is where rhetoric ends. Trucks must be accepted. Powder must be qualified. Naval programs must be integrated. Digital systems must prove recurring economics. Air defense conversations must become contracts. Inventory must become cash.

At €1,207, the price finally acknowledges some of that risk. Whether the business validates the price arrives in Q2. Everything else is commentary.

Subscribe now

• Baidu has crossed the first threshold: AI is no longer a side story. Core AI-powered revenue is now the majority of Baidu General Business, confirming our prior view that Baidu is not defending search so much as rebuilding itself around AI.

• The market may be using the wrong scoreboard: Baidu is not the largest Chinese cloud player: Alibaba has scale, Tencent has economics, Huawei has infrastructure, but Baidu may be much more relevant in the narrower AI-native compute layer where its full-stack approach matters.

• The second threshold is economics: AI Cloud proves demand, but applications, margins, and free cash flow will decide the multiple. The bull case requires Baidu to turn infrastructure growth into application-layer lock-in, not merely replace high-margin search ads with lower-margin compute revenue.

When we wrote “Baidu: The Elephant’s Dance,” the central argument was not that Baidu’s next quarter would be good. It was that the market was using the wrong mental model. Investors were judging Baidu as a declining search company; we argued that Baidu was doing something more deliberate and more dangerous: using search, distribution, data, and cash flow to build a vertically integrated AI company. The IBM analogy mattered because IBM’s transformation under Lou Gerstner was not an optimization exercise. It was an identity change. Baidu, we argued, was attempting the same: not defending search, but dismantling it to build something else.

1Q26 is the first quarter where that argument no longer lives only in metaphor. The elephant is not merely dancing in management commentary; it has crossed an accounting line. Baidu’s Core AI-powered Business reached RMB13.6 billion, up 49% year-over-year, and for the first time exceeded half of Baidu General Business revenue. AI Cloud Infra revenue rose 79% year-over-year to RMB8.8 billion, while GPU Cloud revenue grew 184%.

That is not a normal quarterly datapoint. It is the moment when a transformation moves from “management story” to “income statement reality.” But it also makes the debate harder, not easier. Before this quarter, the market could ask whether Baidu’s AI business was real. After this quarter, that question is less interesting. The better question is whether the AI business Baidu is building is economically superior to the search business it is replacing.

That is the second threshold.

The question is not growth. It is compounding.

The structural mechanism that could make Baidu more valuable is not simply “AI revenue growth.” The mechanism is a loop: intent data creates demand for AI answers and actions; those answers and actions create inference and cloud workloads; those workloads improve infrastructure utilization and model/application performance; lower cost and better performance attract more enterprise and consumer usage; and that usage produces more intent and workload data.

This loop strengthens at scale if Baidu’s stack — Kunlun chips, PaddlePaddle, ERNIE, AI Cloud, search distribution, agents, Wenku, Drive, and Apollo — makes each incremental AI interaction cheaper, faster, and more useful. It breaks if Baidu is merely replacing high-margin search ads with lower-margin compute revenue.

This distinction matters because there are two very different ways to read 1Q26. The first is simple: AI revenue is now big. The second is more important: Baidu may be trying to turn intent into infrastructure, and infrastructure into applications. The former is a revenue story. The latter is a platform story.

The metric is the message.

The most revealing part of Baidu’s release is not the headline revenue beat. It is the metric Baidu wants investors to focus on: Core AI-powered Business. Companies do not introduce new metrics by accident. They do it when old metrics no longer capture the story they want the market to understand — or when old metrics reveal too much pain.

That makes Baidu’s new frame both self-serving and useful. The old Baidu was online marketing plus iQIYI plus optionality. The new Baidu wants to be AI Cloud Infra, AI Applications, AI-native Marketing, and Apollo Go, with legacy search slowly receding into the background.

Source: Baidu 1Q26 results.
This is the quarter’s essential message: Baidu General Business barely changed in size, but it changed dramatically in composition. The company did not beat because the old business recovered. It beat because the new business became large enough to cover more of the old business’s decline.

Crossing 50% does not prove Baidu is an AI winner. It proves Baidu can no longer be analyzed as merely a declining search company.

The search business is not being saved.

The prior article argued that Baidu was not trying to preserve search in its old form. 1Q26 supports that view, but it also forces a more disciplined update: intentional cannibalization is not automatically value creation.

Online marketing revenue fell 22% year-over-year in 1Q26 and dropped to 48% of Baidu General Business revenue, down from 63% a year earlier. That decline should not be treated as an unfortunate side issue. It is the cost of the strategy.

The old Baidu monetized intent through links. The new Baidu wants to monetize intent through answers, agents, workflows, cloud workloads, and eventually transactions. A traditional search query ended with a click. An AI search interaction may end with a completed booking, a healthcare consultation, a shopping decision, a document generated in Wenku, a file workflow in Drive, or a business process executed by an enterprise agent.

That is a better strategic ambition than defending ten blue links. But it is also a harder business model to prove. Clicks were simple to monetize. Actions are more valuable, but also more complex: they require trust, orchestration, partner integrations, attribution, and new pricing models.

So yes, the first part of the thesis is working: Baidu is dismantling the old search model. The unresolved question is whether the replacement model captures more value than the one being dismantled.

The wrong scoreboard.

The biggest missing piece in a simple Baidu earnings analysis is the competitive scoreboard. Baidu is not transforming in a vacuum. It is fighting Alibaba, Tencent, Huawei, ByteDance, and others for position in China’s AI stack. The mistake is to treat that as one race.

On the broad cloud scoreboard, Baidu is not the winner. Alibaba has the scale. Its Cloud Intelligence Group generated RMB41.6 billion of revenue in the March quarter, up 38% year-over-year; external cloud revenue rose 40%, and Cloud Intelligence adjusted EBITA rose 57% to RMB3.8 billion. Alibaba is building AI on top of an existing cloud empire. That is a formidable starting point. (Morningstar, Inc.)

Tencent starts from a different fortress. It does not need AI to become a different company; it can insert AI into WeChat, games, advertising, fintech, productivity tools, content, and cloud. Tencent’s 1Q26 revenue was RMB196.5 billion, gross margin was 56.6%, non-IFRS operating profit was RMB75.6 billion, and free cash flow was RMB56.7 billion, even as capex reached RMB31.9 billion. More revealingly, Tencent disclosed non-IFRS operating profit excluding new AI products of RMB84.4 billion, which effectively shows a very profitable core business funding the AI build-out. (Tencent)

Huawei is harder to underwrite because the public numbers are opaque, but strategically it matters because it sits where sovereignty, chips, devices, telecom infrastructure, government relationships, and enterprise trust intersect. ByteDance has the consumer distribution, recommendation engine, and product velocity, but its advantage is more naturally consumer-AI than enterprise infrastructure.

Baidu’s position is narrower than Alibaba’s and less comfortable than Tencent’s. It is not the biggest hyperscaler. It does not have Tencent’s cash-flow fortress. What Baidu has is urgency and coherence: AI is not a feature being added to the business; AI is becoming the replacement operating system for the business.

The more interesting scoreboard is therefore not broad cloud. It is AI-native compute. Third-party summaries of Frost & Sullivan’s self-developed GPU cloud market report put Baidu Intelligent Cloud first with 40.4% share, ahead of Huawei at 30.1%. This is a narrower market than overall cloud, and the definition matters: self-developed GPU cloud is not the same thing as all public cloud. But that is precisely the point. Baidu looks small on the old scoreboard and much more relevant on the AI-specific one. (C114Pro)

That distinction reframes the quarter. Alibaba is building AI on top of cloud. Tencent is using AI to reinforce a cash-flow machine. Baidu is using AI to become a different company.

Compute is proof. Applications are control.

AI Cloud Infra is where Baidu’s full-stack thesis first shows up. Management has been explicit about this. On the 4Q25 call, Baidu described AI accelerator infrastructure as the primary growth driver of AI Cloud, emphasized its full-stack end-to-end architecture, and pointed to heterogeneous computing, unified scheduling, and Kunlun chips as sources of performance and cost efficiency. Robin Li also made the more important strategic point: “applications matter more than models.”

This is the correct answer to the model race. In a world of DeepSeek, Qwen, Doubao, Kimi, and open-source releases, simply having a model is not enough. Models alone are products with uncertain durability. The value is in deployment: how cheaply inference can be served, how reliably enterprise workloads can run, how quickly applications can be built, and how deeply those applications become embedded in customer workflows.

Baidu’s 1Q26 AI Cloud Infra number is therefore significant. It suggests customers are buying the infrastructure. But infrastructure is only the first proof. The final form is applications.

The distinction is crucial. Infrastructure revenue proves customers need Baidu’s compute. Application revenue proves customers are locked into Baidu’s platform.

That is where Baidu has more to prove. AI Applications revenue was RMB2.5 billion, approximately flat year-over-year and down 10% quarter-over-quarter. AI-native Marketing Services grew 36% year-over-year, but fell 15% sequentially. These are not fatal numbers, but they are the missing link in the bull case.

AI Cloud can make Baidu relevant again. AI applications are what can make Baidu valuable again.

The bill comes due.

The uncomfortable part of the quarter is that AI revenue is not free. The old Baidu was a high-margin intent monetization business. The new Baidu, at least in its current phase, looks more like infrastructure: servers, chips, depreciation, bandwidth, financing, and capex.

Baidu’s cost of revenue rose 7% quarter-over-quarter, primarily due to costs related to AI Cloud. Operating income recovered sequentially, and operating cash flow was positive at RMB2.7 billion, but the quality debate remains open. Bloomberg Intelligence’s post-result comment was blunt: better-than-expected earnings “masked” a 450 bp gross-margin miss, illustrating the dilutive impact of low-margin AI sales.

This does not invalidate the thesis. It clarifies it.

The absolute capex number matters less than the burden it carries. For Tencent, AI capex enhances the fortress. For Alibaba, it defends cloud scale. For Baidu, it funds the escape route from search. That makes the same kind of spending feel different. Tencent can spend into AI while the core throws off enormous free cash flow. Alibaba can spend because cloud scale is already a strategic pillar. Baidu has to spend because the old profit pool is being cannibalized.

That is the second threshold in financial form. The first debate was whether AI could become material. The second is whether material AI revenue can become high-return AI revenue.

What changed in our view.

1Q26 makes us more confident that the prior strategic diagnosis was right. Baidu is not pretending to be an AI company. The revenue mix has genuinely changed, and AI Cloud Infra is growing fast enough to matter. The market’s old question — “can AI offset search revenue?” — is increasingly stale.

But the quarter also makes us less willing to jump straight to the bull case. The prior article leaned into the elegance of the full-stack thesis: intent, model, framework, chip, infrastructure, application. That architecture still matters. It may matter more after 1Q26. But full-stack integration only creates value if it shows up in margins, cash flow, and application-layer monetization.

So the updated thesis is this: Baidu has crossed the first threshold. AI is now large enough to redefine the company. The second threshold is harder. AI must become profitable enough to re-rate it.

The variant perception is therefore not “Baidu AI is underappreciated.” That is too simple. The sharper version is: Baidu’s AI revenue is no longer the controversy. The controversy is whether that revenue compounds like a platform or depreciates like infrastructure.

Three futures.

At roughly HK$135–136, the stock still embeds skepticism. That skepticism is not irrational; it is the correct price for an unresolved transformation. The next three years depend less on whether Baidu can grow AI revenue and more on what kind of revenue AI turns out to be.

This keeps the spirit of the prior scenario framework, but raises the bar. The bull case is more visible after 1Q26, but it is not yet earned.

The scoreboard from here.

The key markers are not complicated. AI-powered revenue must remain above 50% of Baidu General Business; otherwise the new identity looks premature. AI Cloud Infra needs to remain above 50% year-over-year through 2026 to prove the Q1 step-up was not pull-forward. Gross margin needs to recover above 40%; below 38% keeps the utility bear case alive. Free cash flow needs to turn sustainably positive by 2H26. Online marketing decline needs to narrow to better than -15% year-over-year, or else cannibalization remains ahead of monetization.

Most importantly, AI Applications need to return to growth. That is the real tell. If applications stay flat while AI Cloud grows, Baidu may become a respectable infrastructure vendor. If applications reaccelerate, Baidu may become a platform.

That difference is the investment case.

Baidu has crossed the first threshold: AI is now large enough to replace the story of search. The second threshold is harder: AI must become profitable enough to replace the economics of search. The wrong scoreboard matters because Alibaba has scale, Tencent has economics, Huawei has national infrastructure, and ByteDance has attention. Baidu’s edge is narrower but potentially powerful: the conversion of intent into AI workloads, and eventually into actions.

The elephant is no longer standing still. The question is whether it is walking toward a better business, or merely away from the old one.

Subscribe now

• Right on the business, early on the stock: NU has compounded revenue, ARPAC, customers, Mexico, and AI; but the stock is flat because the market still prices it as a credit-sensitive bank.

• Q1 made the thesis testable: 15–90 NPL hit 5.0% and RA-NIM fell to 9.5%; management’s explanation is credible but now needs proof.

• Re-rating requires decision quality: NU gets a platform multiple only if AI/data-driven underwriting shows up in RA-NIM recovery, NPL normalization, and Mexico earnings.

A year ago, the Nubank thesis was elegant. Trust would compound into engagement; engagement would compound into data; data would improve underwriting; better underwriting would expand monetization; and higher monetization, layered on a low-cost digital architecture, would create one of the rare financial platforms that could grow like a technology company while earning returns like a great bank.

A year later, the business has largely followed that script. The stock has not.

That is the uncomfortable starting point for any honest Q1’26 update. Nubank’s quarterly revenue has moved from roughly $3.4 billion in Q1’25 to $5.3 billion in Q1’26. ARPAC has risen from $11.6 to $15.9. Customers have increased from 118.6 million to 135.2 million. Mexico has gone from promising expansion market to break-even business. AI has moved from a strategic claim to a production tool inside credit decisioning. Net income is up sharply year-over-year.

And yet the stock remains near the same low-teens range that framed much of our earlier work.

The easy answer is that the market does not understand Nubank. The better answer is more uncomfortable: the market may understand the upside, but it is not yet willing to underwrite the downside.

That is the tension Q1 brought back to the surface. We have been broadly right about the operating trajectory. We were too early — perhaps too optimistic — about when the market would pay for it.

What We Thought We Owned

Our view of Nubank has evolved over the past year.

In The Purple Way, the thesis was that Nubank was not simply a digital bank. It was a new financial institution built around customer trust, low-cost distribution, and a culture of product simplicity in markets where incumbents had trained customers to expect friction. The purple card was not just a product; it was a wedge into a banking system that had grown fat on fees, branches, and inertia.

By Q2’25, Mexico became the key proof point. The question was whether Nubank’s international growth was being bought with promotional deposit rates or earned through product utility. When Mexico customers largely stayed after yields were cut, the conclusion was powerful: trust had become a funding advantage. The cheapest capital was not a teaser rate; it was habitual use inside an app.

By Q3, the thesis moved from distribution to decision quality. NuFormer made the argument more ambitious: Nubank was not only acquiring customers more cheaply; it was potentially making better financial decisions because it had proprietary data, a unified technology stack, and machine-learning models trained on behavioral signals incumbents could not replicate.

Q4 sharpened the debate. The market sold higher provisions. We argued that the provision increase was largely mechanical: NuFormer-driven credit limit expansion front-loaded IFRS 9 expected losses before the associated revenue appeared. But we also said the important question was whether that limit expansion would convert into NII without credit deterioration.

Q1 did not ask a new question. It asked the same question again, with less room for elegance.

The Question Q1 Forces

The fundamental question is simple:

Is Nubank’s scale making its underwriting engine smarter, or merely making its credit book larger?

This is the entire investment debate.

If Nubank’s scale improves underwriting, then every new customer makes the system more valuable. More customers create more data; more data improves credit decisions, personalization, fraud detection, collections, and cross-sell; better decisions raise ARPAC and reduce losses; higher profitability funds more product development and market expansion. That is the compounding loop.

If, however, scale simply allows Nubank to extend more unsecured credit faster, then the story is less special. The company may still be an excellent digital bank, but it deserves to be valued as a bank: profitable, growing, but exposed to credit cycles and capital constraints.

This is why AI matters. Not because it makes customer service cheaper, although it may. Not because it makes the app better, although it may. AI matters because the core product of banking is judgment under uncertainty. Every credit limit, every loan approval, every collection intervention, every deposit rate, every product offer is a decision.

The old unit of competition in banking was the branch. Then it was the app. Nubank is arguing that the new unit of competition is the decision.

Q1 made that argument measurable.

Why the Market Sold First

The initial selloff after the Q1 print was not irrational. Revenue was strong, but the first-pass market reaction was not about revenue. It was about risk-adjusted profitability.

Risk-adjusted NIM fell to 9.5% from 10.5% in Q4. The 15–90 NPL ratio rose to 5.0% from 4.1%. Credit Loss Allowance rose to $1.79 billion. Adjusted net income was $937 million, down slightly from Q4 despite record revenue.

Those numbers trigger bank-investor muscle memory. The concern is straightforward: perhaps Nubank is not using AI to price risk better; perhaps it is using AI to expand risk faster.

Management’s response was credible. The company argued that the 15–90 NPL move was mostly seasonal, that growth and mix drove provisioning, and that 90+ NPLs actually improved to 6.5%. On the call, management said risk-adjusted NIM should move back toward second-half 2025 levels as Q1 dynamics normalize.

That explanation matters. It is also not enough.

The market is not simply pricing this quarter’s outcome. It is pricing the width of the distribution. Nubank’s AI underwriting may be superior. Cohort economics may be healthy. Coverage may be strong. But the market has not yet seen this model through a genuine adverse credit cycle.

That is the discount.

Why the Stock Came Back

The recovery from the lows should not be interpreted as the market accepting the platform thesis. It was more modest than that. Investors rejected the worst-case interpretation.

They had reasons. The credit bridge was detailed. Late-stage delinquencies improved. The seasonal pattern in early delinquencies had precedent. Mexico reached break-even. AI disclosures were more tangible than in prior quarters. And the broader market backdrop was supportive enough that buyers were willing to underwrite an explanation rather than extrapolate the first headline.

Still, the stock finished down. That matters too.

Q1 did not resolve the debate. It made the debate measurable.

What Changed in Our View

The long-term thesis is not broken. It is more conditional.

Before Q1, it was reasonable to argue that the market was over-penalizing provision optics. After Q1, that may still be true, but it now requires proof. The burden of proof has shifted from demonstrating the flywheel to narrowing the left tail.

That is a subtle but important change.

Revenue growth is no longer enough. ARPAC growth is no longer enough. Mexico reaching break-even is not enough. AI adoption metrics are not enough.

The next phase of the thesis depends on whether decision quality shows up in risk-adjusted returns. The market does not need Nubank to eliminate credit risk; that would be impossible. It needs evidence that Nubank’s data, underwriting, and cost structure make credit outcomes more predictable than they would be at a traditional lender.

That is also where our own framework needs humility. We have been right about the business direction. We were early on the re-rating. The market has not refused to see Nubank’s progress; it has refused to capitalize that progress at a platform multiple before the credit engine is proven through stress.

The variant perception, therefore, is not that Nubank deserves no credit discount. It is that the discount may be too large if the company can demonstrate that its data loop narrows the left tail.

Mexico Is the Counterweight

The reason Q1 is not simply a credit-risk story is Mexico.

Brazil proved that Nubank’s model can scale. Mexico is testing whether it can travel. Q1 was a real milestone: Mexico crossed 15 million customers, became the third-largest financial institution in the market, and reached break-even.

That does not yet force a re-rating. Break-even is validation, not materiality. But it changes the probability distribution. A year ago, Mexico was mostly evidence of transferability. Now it is beginning to look like the second earnings engine.

That matters because the bull case for Nubank has never been only Brazil. It has been that the company’s core capability — trust, data, underwriting, low-cost product manufacturing — can travel across markets. Mexico is the first serious test of that claim. The U.S. is still an option. Colombia is still early. Mexico is the bridge between ambition and evidence.

What Would Close the Gap

There are only three credible ways for the valuation gap to close.

First, Nubank needs a visible credit test. Not seasonal Q1 noise, but an actual period of macro deterioration where industry NPLs rise and Nubank’s outcomes prove structurally better. That would be the definitive evidence, but it cannot be manufactured.

Second, Mexico must become material. Break-even validates the playbook. A business generating $100–200 million of quarterly earnings would force investors to model a second Brazil.

Third, the company needs a pattern: three or four consecutive quarters of rising earnings, stable or improving credit metrics, and risk-adjusted NIM back above 10%. One quarter can be explained. A pattern gets capitalized.

This means the re-rating is probably not immediate. Nubank may be mispriced, but the catalyst is evidence accumulation, not one earnings call.

Scenarios

The scenarios are not just financial. They are classification outcomes.

The bear case is not that Nubank fails. It is that Nubank remains a bank. The bull case is not simply that revenue grows. It is that the market changes the category.

What We Track Now

The next few quarters need fewer metrics, not more.

The 15–90 NPL ratio needs to fall below 4.7% in Q2; if it stays near or above 5.0%, the seasonality argument weakens. Risk-adjusted NIM needs to move back above 10%; if it stays below 9.7%, growth quality remains in question. 90+ NPLs need to stay below 6.7%; a move above 6.8% would challenge the underwriting story.

ARPAC should hold above $15.5 and move toward $17 over time. Mexico needs to remain profitable. The efficiency ratio can normalize toward 20%, but core efficiency should continue improving. And AI needs to show up in cohort behavior, not just adoption metrics.

Early, Not Exonerated

The best theses are not the ones that never get challenged. They are the ones where the challenge clarifies what must be true.

Nubank’s Q1 did not break the thesis. It made it more honest. The company has executed like a platform, but the market is still valuing the unknowns like a bank. That is frustrating, but it is not irrational.

I still think we are early, not wrong. But early no longer means waiting for investors to notice revenue growth, ARPAC, Mexico, or AI. It means waiting for enough credit evidence to narrow the left tail.

If that evidence arrives, the platform multiple becomes available. If it does not, the market will have been right to keep Nubank in the bank bucket all along.

Subscribe now

• The thesis worked. All four signposts cleared, and AMAT raised CY26 semi equipment growth from >20% to >30%; the debate is now what is priced in at $468.

• Complexity now has a volume multiplier. AMAT is not just getting more process steps per wafer from GAA/HBM/packaging; agentic AI may also drive more CPU, DRAM, and NAND wafers.

• Duration is the variant; cash flow is the test. The bull case is AMAT becoming a manufacturing-complexity compounder, but Q2 FCF of $210M means Q3 cash conversion has to normalize.

In February, after Applied Materials reported its fiscal Q1, I set four signposts. They were specific, measurable, and designed to test whether the complexity tax thesis, the idea that 3D chip architectures were permanently shifting semiconductor value toward materials engineering, was structural or just another cycle dressed up as a paradigm shift.

Here’s what happened:

All four cleared. I should be pleased, and I am, but I also need to be honest about something. In January, at $318, I wrote that “risk/reward at $362 is balanced, not compelling. Better entry at $300-310.” The stock is now $468. The thesis was completely right. The entry call was too conservative. The stock re-rated and never looked back.

That tension, correct thesis, missed entry, is the honest frame for thinking about what comes next. Because the question is no longer whether the complexity tax is real. It’s whether $468 already pays fair value for it.

Three Things That Weren’t in the Model

I want to resist the temptation to recap the quarter. The numbers were excellent, record revenue of $7.91 billion, record non-GAAP EPS of $2.86, gross margin crossing 50% for the first time in 25 years. You can read those anywhere. What matters is the three things that are genuinely new and weren’t in my framework three months ago.

First, management upgraded CY2026 semiconductor equipment growth from “>20%” to “>30%.” That’s a 50% increase in the growth outlook in 90 days. This was not a routine beat-and-raise. It was a demand acceleration that management itself didn’t see coming, driven by customers finding incremental clean room capacity and placing equipment orders that weren’t in the forecast as recently as February.

Second, AMAT’s largest customers are now providing rolling eight-quarter demand forecasts. This is, I believe, the single most important disclosure for the re-rating debate. The semi-cap industry has always traded at a cyclical discount because visibility was poor. Two years of customer-provided demand signals directly attacks the foundation of that discount. It doesn’t eliminate cyclicality. But it changes the planning horizon enough to matter for how the business should be valued.

Third, CEO Gary Dickerson introduced agentic AI as an incremental demand driver:

“Agentic AI models do more than respond to queries. They plan, reason, and execute tasks autonomously. They therefore require a computing architecture that is more CPU-intensive while also increasing demand for DRAM and NAND.”

He framed this as additive to existing training and inference workloads. That distinction matters, and it’s the piece that forces me to update my thesis.

Complexity × Volume

My prior articles framed Applied’s advantage as a complexity tax. GAA transistors add 30% more process steps. HBM requires 3-4x the wafer capacity per bit. Backside power delivery doubles certain equipment categories. Each architecture transition increases the number of materials engineering steps per wafer, and Applied has the broadest portfolio to capture them. The tax doesn’t cycle, it accumulates.

That framework was correct but incomplete. It captured the supply side, more steps per wafer, while treating the demand side, number of wafers, as cyclical. This quarter suggests volume is compounding too.

Agentic AI creates incremental CPU demand (more logic wafer starts), incremental DRAM demand (more memory wafer starts), and incremental NAND demand (more storage wafer starts), all on top of the training and inference workloads the market is already modeling. Dickerson noted that global token generation has tripled in three months. Customers added 10+ fab projects to AMAT’s tracking list in a single quarter, bringing the total above 100.

The updated mechanism is simple: complexity compounds AND volume compounds, at least while AI workload diversification continues. That’s why the >20% to >30% upgrade happened so fast. It’s not that management was being conservative in February, it’s that agentic demand emerged as a material driver faster than anyone anticipated.

The gross margin data confirms this isn’t just scarcity pricing. When I asked in February whether 54.5% Semi Systems margins were structural or cyclical, the answer was ambiguous. It’s less ambiguous now. CFO Brice Hill was explicit on the call: pricing generally works through two-to-three-year project contracts and moves slowly. The margin lift comes from the portfolio becoming more valuable as AMAT launches tools that solve harder problems, Trillium ALD for GAA gate stacks, Precision PECVD for isolation trenches, Sym3 Z for conductor etch. These are first-of-kind tools with limited competition. Hill guided Q3 gross margin to 50.1%, continued expansion on higher volume.

Four years of steady expansion. Scarcity pricing spikes and collapses. Portfolio enrichment grinds higher. This looks like the latter.

What Consensus Gets Wrong

Consensus sees AMAT as a well-positioned AI beneficiary in a strong WFE cycle. Consensus is right on the direction and wrong on two things.

Duration. Sell-side models show growth accelerating through FY2027, then flattening. This is the traditional WFE cycle template applied to a non-traditional demand environment. GAA is not a one-time upgrade; it’s the new baseline for every leading-edge fab going forward. HBM stacks are going from 12 to 16 to 20+ dies. Backside power is arriving. Panel-level packaging is next. These transitions don’t replace each other, they stack. And now agentic AI is adding a volume multiplier on top of the complexity tax. The assumption that growth mean-reverts in FY2028 requires believing the complexity tax stops accumulating. I see no evidence for that.

AGS. Management upgraded AGS growth from “low double digits” to “mid-teens” sustainably, with “potentially higher this year.” AGS revenue hit $1.665B with 29.2% operating margins. Over 35,000 chambers connected to AIx servers. Renewal rates above 90%. This is a $7B+ run-rate recurring business growing mid-teens with near-30% margins, and the market still treats it as a cyclical cushion. I valued it at $80B standalone in January. Nothing this quarter weakened that view; if anything, the growth rate upgrade strengthened it.

The non-consensus call: AMAT is transitioning from cyclical WFE company to manufacturing-complexity compounder. The market is paying a cyclical-plus multiple (~28x FY2027) for what may deserve a compounder multiple (30-35x). The gap closes if growth sustains through FY2028 and cash generation normalizes.

The Cash Flow Question

I should flag the one thing that didn’t look good, because a thesis that ignores counter-evidence isn’t a thesis, it’s a sales pitch.

Free cash flow was $210 million. On $2.29 billion of non-GAAP net income, that’s a 9% conversion rate. The headline is bad. The detail is worse: accounts receivable grew 23% quarter-over-quarter while revenue grew 13%, implying DSOs jumped from roughly 63 to 72 days. Inventory hit $6.3 billion, an all-time high.

Management’s explanation, building capacity and inventory ahead of a >30% growth year, EPIC Center construction, legal and tax payments, is plausible. The 8-quarter customer visibility provides the demand backstop that makes the inventory build rational rather than reckless. But the receivables specifically need watching. If DSOs normalize below 65 days in Q3, it was timing. If they stay above 70, something structural changed in customer payment terms and the quality-of-earnings debate starts in earnest.

Physical scaling is expensive. Software companies compound with near-zero marginal cost. AMAT compounds with factories, inventory, suppliers, and capex. The P&L says the portfolio is becoming more valuable. The cash flow statement says proving that value requires capital. The bull case needs both to be true.

Three Worlds at $468

In January, at $318, I laid out three scenarios through FY2028. “Physics Wins” implied $480 at 30x $16 EPS. The stock is already at $468. Time to recalibrate for a 3-year horizon through FY2029.

Source: Author estimates based on AMAT guidance, management commentary, and historical WFE cycle analysis.

The probability-weighted expected value is roughly $490-530, 5-15% above current price. The asymmetry is modest, which is itself the key insight. At $318, the stock was priced for “The Inventory Reckoning” and you only needed “The Higher Floor” to work. At $468, the stock is priced for “The Higher Floor” and you need “The Compounding Era” to materialize for significant upside.

That’s not bearish. But it’s a fundamentally different risk/reward than six months ago. The thesis has been de-risked by execution, and the stock has re-priced accordingly.

What would shift probabilities: “The Compounding Era” becomes more likely if Q3-Q4 revenues grow linearly (per Hill’s explicit guidance), if the EPIC Center unveiling in October attracts additional partners, and if FY2027 guidance when it comes implies sustained >15% growth. “The Inventory Reckoning” becomes more likely if DSOs stay elevated, DRAM spot pricing falls more than 10% in a quarter, or TSMC cuts FY2027 capex below $50B.

What I’m Watching

Six signposts for the Q3 earnings print in August:

1. Q3 revenue ≥$9.1B, confirms acceleration beyond the guide midpoint

2. Semi Systems gross margin ≥54.5%, pricing power intact at higher volume

3. DSOs back below 65 days, working capital normalizing, not structural

4. FCF conversion above 35%, earnings quality validated after Q2 weakness

5. “2027 another strong record year” language maintained, duration signal

6. CY2026 semi equipment growth held at ≥30%, no downward revision

In January I wrote that the complexity tax doesn’t cycle, it accumulates. In February I set four signposts to test that claim. All four cleared. The thesis has accelerated faster than I expected, and honestly, faster than the stock could have been bought at a comfortable price.

At $468, the market is no longer paying you to identify the shift from lithography to materials engineering. It’s asking you to size the duration. I think the duration is longer than consensus models, complexity compounds, volume is now compounding alongside it, and the visibility mechanism changes the risk profile of the business. But the margin of safety is thinner than it was six months ago, and the cash flow statement is a reminder that physical compounding has costs that software compounding does not.

The complexity tax is real. The question is whether the stock already collects enough of it.

Subscribe now

• Alibaba’s Q4 validated the strategy but not the economics. Cloud external revenue grew 40%, AI revenue is now meaningful, and Qwen/MaaS monetization is becoming measurable, but adjusted net income was effectively zero, FCF was deeply negative, and Cloud margins remain around 9%.

• The key debate is no longer whether Alibaba can grow AI/cloud revenue. It can. The real question is whether Alibaba can earn platform returns on that revenue, or whether China AI cloud becomes essential infrastructure with utility-like economics.

• Our view is downgraded but not broken. The “Second Trust” thesis still survives: Qwen could become a trusted transaction layer across Taobao, Alipay, Amap, DingTalk, and Alibaba Cloud. But until Cloud margins expand, All Other losses narrow, and FCF turns, Alibaba is not China’s AWS; it is a very important company still trying to prove that importance belongs to shareholders.

Alibaba’s Q4 FY2026 was the kind of quarter that makes a stock screen useless.

Revenue missed. Adjusted EBITDA missed badly. Adjusted net income was 86 million yuan, effectively zero, against consensus of 15 billion. Free cash flow was negative 17.3 billion yuan, bringing the full year to negative 46 billion. On any conventional earnings framework, the stock should have gone down and stayed down.

It went up 7.5%.

The rally was not purely about earnings. It coincided with reports that Jensen Huang and other US CEOs were joining Trump for meetings with Xi in Beijing, a broader China AI and détente trade that lifted the KraneShares China Internet ETF alongside Alibaba. The marginal buyer on May 13 may have been underwriting geopolitical thaw, not AI monetization milestones. That distinction matters for how durable the move is.

But the deeper question is not whether Alibaba had a good quarter. It did not. The question is whether the quarter made Alibaba’s long-term architecture more believable.

Our answer: yes, on demand, no on economics.

What We Said

Our prior Alibaba thesis had a simple starting point: Alibaba’s original superpower was trust infrastructure. Taobao did not win merely because it was a better website. It won because Alipay solved the prisoner’s dilemma between buyers and sellers. The “Second Trust” thesis argued that AI creates a new version of that problem, will users trust software to execute transactions, not just suggest them?, and that Alibaba, with 930 million Taobao users and 1.4 billion Alipay accounts, does not need to manufacture that trust from scratch.

The investment case rested on three systems. The Compute Spine was Alibaba Cloud and Qwen: the infrastructure layer for China’s AI adoption. The Frequency Layer was quick commerce: expensive, but necessary to defend daily habit against Meituan. The Monetization Engine was CMR, Quanzhantui, 88VIP, and the software service fee: the mechanism by which user activity turned into platform revenue.

We said the thesis came down to one question: would the next few bills get smaller?

Q4 answered in the least satisfying way possible. The systems are more real than before. The bills are not smaller. They changed shape, AI capex replaced merchant subsidies, model development replaced user acquisition, but the total burden stayed the same.

Here is the scorecard:

The Compute Spine exceeded expectations. We set the bar at 25% cloud growth. Got 40% external growth. AI-related product revenue posted triple-digit growth for the eleventh consecutive quarter, now representing 30% of external cloud revenue. Management gave specific forward targets: MaaS ARR above 10 billion yuan in the June quarter, 30 billion yuan by year-end. The demand side validated more strongly than our bull case assumed.

But Cloud EBITDA margin compressed to 9.1%. We specified “stable or improving margins” as part of the criteria. We got the growth without the margins.

The Monetization Engine produced an ambiguous result. We set CMR growth above 8% as the structural validation threshold. Reported CMR grew 1%. Management then introduced an accounting reclassification, moving merchant subsidies from S&M expense to contra-revenue, and claimed 8% adjusted growth. When the company needs an accounting change to hit your validation number in the same quarter the headline metric decelerates to 1%, the appropriate response is not to accept the adjusted figure at face value. The Monetization Engine may be intact. It may be impaired. The fact that we cannot tell is itself a downgrade.

The Frequency Layer delivered exactly what we expected: strategic validation and financial destruction simultaneously. Quick commerce revenue surged 57%. Order volume was 2.7x. China E-Commerce EBITDA fell 40%. Management committed to positive unit economics by end of FY2027, the first specific timeline. Whether they hit it is the next test.

Free cash flow and buybacks confirmed our November warning. FCF was negative 46 billion yuan for the year. Buybacks did not resume. The $1.05/ADS dividend signals maintenance, not confidence. “Peak investment” was management narrative, not financial reality.

The strongest sentence I can write about Q4 is this: the strategic thesis validated; the economic thesis did not.

The Metric Alibaba Wants You to See

The most important disclosure was not the headline miss. It was the introduction of AI annualized recurring revenue as a primary metric.

Management disclosed quarterly AI revenue of 9 billion yuan, an ARR of roughly 35.8 billion yuan ($5.3 billion), and specific forward targets for MaaS monetization. Companies introduce new metrics when old ones tell a bad story. The old metrics, EBITA, net income, free cash flow, are catastrophic. The new metric is spectacular. That is not inherently dishonest. Amazon introduced the AWS disclosure the same way. But it should trigger analytical caution, not relief.

The critical detail most coverage missed: most MaaS revenue comes from Alibaba’s own proprietary models, Qwen, T-MOR, voice and video generation, not third-party resale. That means Alibaba captures the full margin stack: model IP, hosting, inference. Customers building on Qwen-specific APIs face real switching costs. If that pattern holds for multiple quarters, it changes what Cloud is worth, because it is platform economics rather than commodity infrastructure.

The Bloomberg screen still showed an overwhelmingly bullish analyst stack, 47 buys, 3 holds, and 1 sell, for a company that just printed near-zero adjusted net income. That is not proof the analysts are wrong. But it is a reason to haircut consensus estimates, which embed a near-quintuple in EBITDA over the next twelve months.

The Comparison That Reframes the Question

The biggest change to our view is not about Alibaba’s technology. It is about the returns investors should expect from that technology.

The seductive narrative is “AWS in China.” China skipped enterprise cloud adoption; AI forces that migration; Alibaba owns the leading domestic cloud; the stock re-rates. There is truth in that story. But the economics tell a different one.

Google Cloud is growing faster than Alibaba Cloud at 3.5x the revenue with margins nearly four times higher. AWS generates more quarterly revenue than Alibaba Cloud generates in a year. The entire Chinese cloud market is smaller than AWS alone.

The structural explanation is market fragmentation: seven-plus competitors for a market one-seventh the size of the US, with state-owned telcos pricing below cost. Server deployment costs have doubled year-over-year despite T-Head chip production. Google Cloud went from operating losses to 33% margins in roughly three years. Alibaba Cloud has been operating for fifteen years at sub-10% EBITDA margins.

This leads to the variant perception that I think matters most:

The market is debating whether Alibaba can grow AI/cloud revenue. The better question is whether Alibaba can earn platform returns on that revenue.

State support can guarantee demand, but it can also cap returns. If AI cloud is treated as national infrastructure, Alibaba may be allowed, even encouraged, to grow. But the same political economy that protects the market may also prevent Alibaba from extracting Western hyperscaler economics from it. In that world, Alibaba becomes essential without becoming exceptional.

The Part AWS Cannot Copy

The only credible counter to the margin comparison is not better cloud infrastructure. It is the commerce integration that no Western hyperscaler can replicate.

AWS has better cloud economics. Google Cloud has better margin momentum. Microsoft has better enterprise distribution. Alibaba’s advantage is the possibility that Qwen becomes a trusted transaction interface inside Taobao, Alipay, Amap, DingTalk, and the merchant stack. If that happens, Alibaba is not merely selling compute. It is converting consumer intent into commerce, commerce into data, data into models, and models into recurring inference demand that runs on Alibaba Cloud.

That is the Second Trust thesis: trust scaled through AI agents that can decide, recommend, execute, and transact. It is also the only version of the bull case where Alibaba earns something better than utility-grade returns on its cloud investment.

Q4 did not prove it. Q4 only proved the cost of trying. The “All Other” segment, where Qwen model development and consumer AI sit, posted a 21.2 billion yuan EBITDA loss in a single quarter. The commerce-AI integration may work. But the financial evidence for it showing up in CMR or conversion rates is not yet visible.

Three Futures

Scenarios are useful only if they describe what the company becomes.

The Kingdom (15-20% probability → HKD 280-330)

Alibaba becomes China’s AI-commerce platform. Qwen becomes a transaction interface. Cloud growth sustains above 30%. AI becomes a majority of external cloud revenue. MaaS mix lifts Cloud margins toward 15-18%. Quick commerce breaks even. CMR grows 10%+ as AI tools improve merchant ROI. Revenue compounds at 12-14% through FY2029 with group net margins reaching 14-16%. FCF turns strongly positive by FY2028. Multiple re-rates to 18-20x.

This requires Chinese cloud margins to break their fifteen-year pattern AND the commerce-AI loop to produce measurable economic uplift. Possible but demanding.

The Grid (50% probability → HKD 170-220)

Alibaba becomes important infrastructure but not a platform monopoly. Cloud grows 20-25%. Margins improve modestly to 10-13%, better than today, but nowhere near Western levels. Quick commerce stops bleeding but never becomes a profit engine. CMR grows mid-to-high single digits. Revenue compounds at 8-10%. Net margins reach 11-13%. FCF turns positive but below prior expectations. Multiple stays at 14-16x.

This is what management is actually describing when Wu says ROI will be “extremely clear in 3-to-5 years.” Not a re-rating story. A steady compounder where returns come from earnings growth, not multiple expansion.

The Furnace (20-25% probability → HKD 95-120)

Alibaba keeps burning capital to defend relevance. Cloud grows but capex grows with it. Quick commerce remains a permanent tax. Qwen investment sits in All Other without visible payback. CMR slows. Revenue grows 4-6%. Margins stay flat at 7-9%. FCF remains weak. Multiple compresses to 10-12x.

This is not collapse. It is capital consumption, strategic importance without shareholder returns. Railroads were essential. Utilities are essential. Essential infrastructure does not automatically mean good equity.

A separate 10-15% probability covers geopolitical disruption, delisting, severe chip sanctions, Taiwan escalation, where fundamentals become irrelevant. We do not model it, but we size positions to survive it.

The Dashboard

The single most important signpost is Cloud margin. AI revenue growth will be easy to celebrate. Margin will tell us whether Alibaba is building a platform or selling scarce compute at cost.

The Bottom Line

Alibaba’s first trust machine was escrow. It made strangers comfortable transacting online, and that was enough to build one of China’s most important companies. The second trust machine is AI, Qwen deciding, recommending, executing, transacting across the largest consumer platform in China. If it works, Alibaba is not just a cloud infrastructure provider. It is the transaction and compute layer for China’s AI economy.

Q4 made that possibility more believable. It also made the cost impossible to ignore.

The old Alibaba compounded because trust scaled almost for free. The new Alibaba is trying to compound trust through data centers, chips, delivery riders, models, and AI agents. That can still work. But it is no longer enough to say the architecture is elegant.

Until Cloud margins expand, All Other losses narrow, and free cash flow turns, Alibaba is not China’s AWS. It is a very important company still trying to prove that importance belongs to shareholders. Fifteen years of margin data say the proof will not come. One quarter’s AI ARR disclosure says maybe it will.

I know which one I want to believe. I am waiting for the data to tell me which one is true.

Subscribe now

• Toast is no longer just trying to be the system restaurants use; it is trying to be the system that does the work restaurants used to outsource.

• The quarter strengthened the long-term thesis; higher monetization, strong ARR growth, Toast IQ usage, and early Toast IQ Grow ROI, but weakened near-term confidence because guidance did not flow through cleanly.

• The next test is not revenue growth; it is whether Toast can monetize outcomes: higher take rate, paid AI attach, stable GPV/location, contained credit losses, and a credible direct-demand loop through Toast Local.

The Thesis Held; the Stock Didn’t

In November, we wrote about Toast at roughly $37 as the restaurant integration layer. In February, we wrote again around $26, arguing that the market was confusing chosen reinvestment with structural margin pressure. Today the stock is closer to $24, and before asking readers to follow the thesis a third time, the honest place to start is with the gap.

The business did many of the things we wanted. The stock did not.

That deserves an accounting.

Toast grew ARR 26% to $2.2 billion, added roughly 7,000 net locations, grew GPV 22% to $51.3 billion, delivered $179 million of adjusted EBITDA, generated $115 million of free cash flow, and raised full-year adjusted EBITDA guidance to $790–810 million. The headline operating metrics were not weak.

And yet the stock sold off because the quarter did not answer the question investors wanted answered: will Toast let upside fall to the bottom line? Q1 adjusted EBITDA beat consensus, but Q2 guidance came in below Street expectations, and the full-year raise largely reflected the Q1 beat rather than a meaningfully higher second-half trajectory. The market was not reacting to what Toast had done; it was reacting to what Toast refused to promise.

That was the same tension as last quarter, but with a more important twist. The prior debate was whether Toast’s margin pause reflected investment or deterioration. This quarter showed something more interesting: Toast is no longer merely reinvesting the profit sanctuary into new markets. It is trying to use AI to turn its restaurant operating system into something closer to restaurant labor.

The register was the wedge. The operating system was the moat. The invisible manager is now the bet.

The Quarter Had Two Stories, and the Market Priced the Worse One

From Software You Use to Work You No Longer Do

The fundamental question for Toast is no longer whether restaurants will adopt cloud POS. They have. Nor is it whether Toast can be profitable. It is. The question is whether Toast can convert workflow ownership into outcome ownership.

Toast’s management said this directly. Aman Narang described Toast’s evolution from point-of-sale into a system of record, then laid out the next step: Toast has “always provided the software,” but with AI it can now “provide the service that can actually do the work” for operators across marketing, bookkeeping, payroll, tax, and other non-core functions. The reason this matters is that the data required for those functions, guest orders, visit frequency, labor spend, inventory spend, business performance, already lives inside Toast and becomes more valuable with every new location and transaction.

That is the bull half of the quarter. Toast IQ now has 40,000 weekly active locations. Toast IQ Grow, the first agentic product, builds and optimizes campaigns from a restaurant’s past performance data, sales forecast, and eventually weather and local events. Management said pilot customers saw an average 8% sales increase versus similar Toast restaurants.

This is not just an AI feature. A standalone marketing tool can write an email. Toast can decide which guest should receive it, which menu item should be promoted, what margin that item carries, whether the restaurant can staff the incremental demand, and whether similar restaurants saw the same pattern. The intelligence is not in the model alone; it is in the workflow context.

DoorDash Is Not a POS Problem; It Is a Demand Problem

The bear half of the quarter arrived from the other direction. During Q&A, Rothschild’s Dominic Ball asked directly about DoorDash POS activity in San Francisco, Phoenix, and New York, and whether DoorDash was moving from partner to peer. That is the right strategic question. Toast may own the operating layer, but DoorDash and Uber own large parts of the demand layer.

These are not the same games. Toast’s integration layer is hard to displace in full-service restaurants, hotels, fine dining, complex multi-location enterprise, and operationally dense retailers. Replacing the terminal is easy; replacing payments, payroll, inventory, kitchen workflows, scheduling, lending, marketing, and now an agent layer is not. But in delivery-heavy QSR, where operational complexity is lower and demand dependency on aggregators is higher, the moat is thinner. DoorDash does not need to replace Toast everywhere to matter. It only needs to create enough narrative damage in enough high-volume categories to compress the multiple.

That is what happened this quarter: the agent thesis got more real, and the aggregator threat got more visible. The market priced the second faster than the first.

Toast Local Is the Countermove, Not the Answer Yet

Toast Local is the piece that both bulls and bears need to take more seriously.

Management framed Toast Local as a low- or no-commission channel for restaurants to generate demand, with no-fee ordering, loyalty rewards, targeted offers, and reservations through Resy and Toast Tables at more than 20,000 locations. Weekly app downloads more than doubled last quarter.

This is not a DoorDash killer. That is the wrong test. DoorDash is a consumer habit; Toast Local is still a strategic option. The better question is whether Toast can create enough demand-side surface area to prevent the operating system from becoming a back-office utility while aggregators own the guest relationship.

The old Toast thesis was internal: own the workflows. The new Toast thesis is both internal and external: own enough of the workflow and enough of the guest loop that the data compounds in both directions. A reservation can inform staffing. A loyalty offer can inform inventory. A payment credential can make checkout invisible. A guest history can make table service more personal. A slow Tuesday can trigger an agent-created campaign.

That is the point. Toast Local does not need to become DoorDash. It needs to connect demand to operations in a way DoorDash cannot, because DoorDash does not run the restaurant.

The Old Variant Perception Was Payments; the New One Is Management

Our prior view was that Toast was misclassified. The market saw payments-heavy revenue and valued the company like a payments processor with software attached. We argued the opposite: Toast is a restaurant operating system with payments as the monetization rail. That view still holds. The prior articles emphasized the vertical SaaS + payments flywheel, the integration substrate, the 40%-margin core profit sanctuary, and the decision to reinvest rather than harvest.

What changed is the nature of the upside.

The old variant perception was: Toast is not payments; it is restaurant infrastructure.

The updated variant perception is: Toast is not just infrastructure; it is trying to become the invisible manager.

That is a more ambitious thesis, but also a more falsifiable one. AI usage is no longer enough. Toast has to show that restaurants will pay for work performed, not merely software provided. If Toast IQ Grow drives an 8% sales lift but Toast cannot capture meaningful economics, the product may be great for retention but not transformative for valuation. If it can charge for outcome-oriented work across marketing, payroll, bookkeeping, inventory, scheduling, and demand generation, then total monetization as a percentage of GPV should rise for years.

That metric is already moving. Total monetization across SaaS and fintech crossed 1% of GPV for the first time, reaching 103 bps, up 5 bps year over year. SaaS ARR grew 27%, subscription gross margin exceeded 80%, and non-payments fintech contributed $51 million of gross profit and 10 bps of take rate.

If Toast were merely a processor, GPV and gross profit growth would eventually converge. The platform thesis requires the opposite: the same dollar of restaurant volume becomes more valuable as Toast attaches more software, fintech, marketing, capital, and automated labor to it.

Why the Stock Fell Anyway

The market’s frustration is not irrational.

GPV per location declined 1% year over year. Hardware and professional services gross profit was negative 13% of recurring gross profit streams as Toast leaned into customer acquisition and absorbed tariff costs. Bad debt and credit-related expenses were $28 million. Management also said free cash flow conversion will be slightly lower than 2025 because it is strategically buying memory chips and holding more inventory near term.

None of this breaks the thesis. It does raise the proof burden.

Toast is asking investors to underwrite several things at once: continued core share gains, enterprise execution, international and retail expansion, hardware drag normalization, Toast Capital credit discipline, and now AI monetization. That is a lot to ask in a market that has turned hostile toward software stories without near-term flow-through.

The right critique is not “Toast missed the quarter.” It did not. The right critique is: Toast needs to prove that reinvestment creates a higher-quality revenue layer, not just more revenue.

Three Futures for the Invisible Manager

The scenario math is less important than the strategic world each scenario implies, but discipline still matters. Starting from roughly $24, and using the Bloomberg path you shared, FY26 revenue around $7.4 billion, FY28 revenue around $10.2 billion, FY28 EBITDA around $1.35 billion, the three-year outcomes look like this.

The Aggregator’s Revenge. DoorDash and other demand aggregators pressure delivery-heavy QSR, Toast IQ Grow sees usage but weak paid attach, GPV per location stays soft, credit losses rise, and hardware drag lingers. Toast remains a good vertical POS/payments company, but not a re-rated platform. Revenue compounds roughly 12–15%, EBITDA margins stall around 10–12%, and the stock trades at 10–12x EBITDA. That yields roughly $20–28 three years out. The buyback and balance sheet help, but the multiple stays capped.

The Quiet Compounder. DoorDash proves to be a narrow threat, not a platform breaker. Core restaurants keep adopting Toast, enterprise and drive-thru scale steadily, international and retail contribute but do not transform the story, and AI improves retention and attach without becoming a huge standalone profit pool. Revenue compounds 17–19%, EBITDA margins move toward 14–16%, and the stock trades at 14–17x EBITDA. That yields roughly $42–55. This remains the base case: Toast is not yet the invisible manager, but it is a much better business than the current multiple implies.

The Operating System for Local Commerce. Toast IQ Grow monetization is clean, the agent stack extends into payroll, bookkeeping, scheduling, inventory, and demand generation, Toast Local creates a credible direct-demand loop, and total monetization moves from 103 bps toward 130–150 bps over time. Revenue compounds 22–25%, EBITDA margins approach 18–21%, and the market reclassifies Toast as a high-quality vertical platform at 18–22x EBITDA. That gets the stock to $75–90.

The important point is not that the bull case is certain. It is not. The important point is that at $24, the market is not paying for it. The stock is priced closer to The Quiet Compounder with a DoorDash overhang than to The Operating System for Local Commerce.

The Next Few Quarters Are About Proof, Not Story

The next few quarters need to turn the story from plausible to measurable.

The first metric is total monetization. Above 105 bps of GPV would suggest the platform is becoming more valuable per dollar of restaurant volume. Below 100 bps would make the processor critique harder to dismiss.

The second is the spread between subscription ARR growth and GPV growth. A 3–5 point positive spread supports the software layer. Convergence suggests commoditization.

The third is Toast IQ Grow monetization. Usage, pilots, and anecdotes are not enough anymore. The market needs attach rate, ARPU uplift, revenue contribution, or explicit evidence that AI work is being paid for.

The fourth is GPV per location. Positive growth would ease the macro concern. Worse than -2% would suggest restaurant demand pressure is becoming more than noise.

The fifth is hardware and professional services losses. Better than -10% of recurring gross profit would suggest the subsidy is contained. Worse than -15% would make the acquisition-cost problem harder to dismiss.

The sixth is Toast Capital. Credit costs in the high-$20 million to low-$30 million quarterly range are manageable. A move into the mid- or high-$30 million range would force a harder look at whether lending is adding more cyclical risk than strategic value.

The seventh is new TAM mix. Enterprise, international, and retail approaching 30% of net adds would validate the expansion thesis. If they remain small, the core is still doing most of the work.

The Bet Is Cheaper, but the Bar Is Higher

The business evidence improved. The stock evidence did not.

That distinction is uncomfortable, but it is the entire setup. Toast is not broken because Q2 guidance disappointed. It is also not vindicated because Toast IQ usage is high. The company has moved from a familiar thesis, vertical software plus payments, into a more ambitious one: software that becomes labor.

The market is focused on the right problems: guide shape, DoorDash, GPV per location, hardware drag, credit risk, and AI monetization. But it may be using the wrong time horizon. The question is no longer whether Toast can run the restaurant. The question is whether Toast can increasingly manage it.

At $37, that was an expensive possibility. At $26, it was a more interesting one. At $24, the option is cheaper, but the proof burden is higher.

The register was the wedge. The operating system was the moat. The invisible manager is now the bet.

Subscribe now

• The original question was whether difficulty would become economic power or just higher capital intensity. This quarter moved the answer toward economic power: Electronics margins expanded 500bp in two years, and the FY2026 profit upgrade was driven 14:1 by ASP/product mix over volume.

• Ibiden’s real asset is qualified complexity, not capacity. AI packages are getting larger, denser, more multilayered, and harder to yield. That concentrates volume at the few suppliers that can qualify each new generation, and Ibiden is beginning to get paid for it.

• The thesis has advanced, but the next proof is returns. Customer advances, Ono utilization, Gama ramp, and ROIC will determine whether Ibiden becomes structural AI infrastructure or simply an important supplier doing harder work for similar economics.

In April, I wrote that the core question about Ibiden was not whether AI substrates were getting harder. They obviously were. The question was whether that difficulty would compound into Ibiden’s economics, better margins, deeper customer lock-in, a widening yield gap, or merely force the company to spend ever more capital to stay relevant.

I framed three scenarios. I identified the customer advance payment balance as the leading indicator. I said that if advances kept rising, the market had underestimated what Ibiden was becoming, and if they stalled, the old cyclical classification was right.

On May 11, Ibiden reported results that moved the answer, not definitively, but clearly enough that the burden of proof has shifted.

The 14:1 Ratio

The most important data point in the earnings was not the guidance beat. It was a decomposition.

When Ibiden revised its FY2026 Electronics operating profit from the prior ¥57 billion plan to ¥75 billion, it published a bridge: ASP and product mix contributed +¥20.5 billion. Volume contributed +¥1.5 billion. FX added +¥5 billion. Costs absorbed -¥9 billion.

That 14:1 ratio, value to volume, is what “difficulty compounding into economics” looks like in the financials. Ibiden is not shipping dramatically more substrates. It is shipping dramatically more complex substrates, and the complexity commands higher prices because fewer suppliers can manufacture them at yield.

The technology roadmap quantifies why. Die sizes are headed from 3.5 reticle to above 9 by CY2030. Substrate bodies from 80×80mm to above 130×130mm. SAP layers from 9-X-9 to 14-X-14. Each step makes the manufacturing harder, pushes cumulative yields lower, and narrows the qualified supplier base. The substrate is no longer a board underneath the chip. It is part of the system architecture, connecting chiplets, routing signals, delivering power, managing warpage at tolerances that compound difficulty with every additional layer.

Electronics OPM moved from 13.6% in FY2024 to 18.6% in FY2025, guided to 22.7% in FY2026 and roughly 29% by FY2027. That is not cyclical recovery. That is value migration, and the profit bridge proves it is driven by what Ibiden ships, not how much.

The mid-term plan revision confirmed the trajectory. FY2027 operating profit was raised from ¥90 billion to ¥150 billion, a 67% increase to the anchor year. Management does not revise an anchor year by that magnitude because the cycle improved. It revises because it has concluded the structural position is more valuable than previously communicated. The long-term target: over ¥1 trillion in revenue and ¥300 billion in operating profit by FY2030, at 30% margins.

In April I wrote that the yield-learning loop, harder packages narrow the qualified field, concentrated volume accelerates learning, faster learning wins the next generation, was the mechanism to watch. This quarter, the profit bridge quantified it. Difficulty is entering the P&L.

What the Advance Payments Are Actually Saying

Which makes the one piece of contradictory evidence worth confronting honestly.

In April, I identified customer advance payments as the leading indicator. I said rising toward ¥130 billion was thesis confirmation.

They declined. From ¥92.1 billion to ¥81.0 billion.

The most likely explanation is mechanical, Ibiden delivered against previously prepaid capacity while the next round, tied to Gama Cell 6 and Ono Cell 8 mass production beginning in FY2027, has not yet been contracted in full. The ¥81 billion still represents 16% of forward revenue. Customers are simultaneously asking for capacity commitments through CY2029.

But I set the signpost. The direction was wrong. That deserves weight.

What complicates the picture is a signal I did not have in April. When Goldman asked about capacity beyond the current ¥500 billion plan, management said the binding constraint is headcount and engineering talent, not capital. Customers want more than Ibiden can build, and the bottleneck is the engineers in Ogaki City who solve next-generation yield problems, not the money to build factories.

This changes the framework in a way that is, on balance, more favorable than the advance payment decline is unfavorable. A capital constraint can be solved by any well-funded competitor. A talent constraint concentrated in a 50-kilometer radius compounds in favor of the incumbent. It is a human-capital version of the yield loop, the engineers who solved the current generation’s problems are the same ones who will solve the next generation’s, and they cannot be hired away because the institutional knowledge is embedded in the factory, not in any individual.

I expect advances to inflect upward in the next two to three quarters as Gama and Ono contracts formalize. If they fall below ¥70 billion instead, I will reassess.

The Pricing Choice

Citi’s Naito asked the natural question: with capacity tight and your position strong, why not raise prices more aggressively?

CEO Kawashima’s answer: the company does not want to increase prices “beyond the standard zone” and prefers “a reasonable price level based on the discussion with our customer.”

In April I wrote that the resolution of Ibiden’s pricing tension would come through mix rather than rate, each generation’s substrate is larger, more complex, and higher-priced, even if pricing on any given product stays disciplined. The profit bridge confirmed exactly that.

But the restraint is also strategic. Ibiden’s lead at the technology frontier lasts roughly six to twelve months before Taiwanese competitors qualify. If Ibiden extracted maximum rent during each window, customers would accelerate second-sourcing. By pricing reasonably, Ibiden reduces that urgency. This is the TSMC playbook, not the NVIDIA playbook. TSMC can price with restraint because restrained pricing keeps Intel and Samsung from investing as aggressively in catching up. Ibiden’s logic is identical: near-term margin for long-term incumbency. The published OPM targets represent what Ibiden chooses to earn, not what it could.

Three Classifications

The question is not what the stock is worth under different assumptions. It is what kind of company Ibiden actually is. The classification determines the economics.

The Cyclical is Morgan Stanley’s view: Ibiden matters but the difficulty premium gets competed away. Competitors add qualified capacity, pricing normalizes, and the ¥500 billion capex absorbs the margin expansion through depreciation. The work gets harder; the returns do not scale.

The Compounder is the path Ibiden is on now. Each generation lifts mix, margins hold above 22%, ASIC diversification broadens the customer base, but competition at maturity of each generation keeps this from becoming a toll road. Difficulty pays, but doesn’t compound into monopoly economics.

The Infrastructure is the reclassification case. Complexity accelerates faster than competitors can qualify. The talent constraint deepens. A second capex decision in 2027 raises FY2030 targets. Glass core evolves as a hybrid architecture, glass beneath, ABF build-up layers on top, preserving rather than disrupting Ibiden’s process advantage. The market stops treating this as a cyclical Japanese industrial.

In April my view was Compounder with optionality toward Infrastructure. After this quarter, the evidence has shifted further, the 14:1 ratio, the mid-term plan magnitude, the talent constraint, but the stock has moved past my prior bull case, which complicates the entry even as it confirms the thesis.

What Resolves It

Advances received. Above ¥90B by H1 FY2026 confirms the co-funding cycle is intact. Below ¥70B means the structure was scarcity-driven, not relationship-driven. Still the single most important metric.

Ono utilization. Above 90% by Q3 FY2026 confirms execution and validates the H2 margin step-up. Below 85% reopens the execution question.

Electronics OPM. Sustaining above 22% confirms difficulty is compounding faster than depreciation absorbs it. Below 18% favors the Cyclical classification.

ROIC. Inflecting above 12% by FY2028 is the ultimate verdict, it separates “strategically important but capital-intensive” from “structural compounder.” Everything else is a leading indicator.

Ajinomoto allocation. Whether ABF film supply gates Ibiden’s upside scenarios remains unaddressed. If it surfaces as a binding constraint, the market will not have priced it.

In April I asked whether difficulty compounds faster than capital consumes it. The profit bridge says yes. The margin trajectory says yes. The talent constraint says the loop may be stronger than I initially framed.

The advance payments say: not yet confirmed, check back in two quarters.

The constraint beneath AI has entered the P&L. Now it must enter returns.

Subscribe now

• CEG’s product has evolved: from clean “pressure” to time-to-power, power delivered where and when AI customers need it.

• Calpine turns the reservoir into a routing layer: nuclear is the anchor; gas, retail, powered land, and development make it usable.

• The key test: can CEG convert its clean-firm optionality into contracted base earnings through PJM, Freestone-like deals, and sustained fleet reliability?

The mistake in analyzing Constellation is to treat Q1 as an earnings print. It was not. It was the quarter where the company’s real product became visible.

We used to say Constellation sold pressure, not power. That was right. Nuclear plants were the reservoir, contracts were the pipes, the production tax credit was the levee, and capacity markets were the tolls paid for readiness. In a grid increasingly fed by intermittent inflows, the company that could provide steady, carbon-free pressure would become more valuable. That was the core of our original thesis.

Then Calpine forced the next refinement. Constellation was no longer just a reservoir company. It was becoming a power-orchestration platform: nuclear baseload, gas flexibility, batteries, demand response, customer relationships, and structured contracts packaged into solutions that no pure-play generator could easily replicate.

Q1 sharpens the point again. The customer is not simply buying clean power. The customer is not even simply buying firm clean power. The customer is buying a date: power at a specific site, by a specific quarter, with interconnection, reliability, regulatory approval, carbon attributes, and contract structure already solved.

In an AI infrastructure cycle where the bottleneck has moved from chips to energized sites, time-to-power is the product. Constellation is becoming the routing layer that sells it.

The Constraint Has Migrated

Constellation reported $4.49 of GAAP EPS and $2.74 of adjusted operating EPS in Q1, up from $0.38 and $2.14 a year ago. It reaffirmed full-year adjusted operating EPS guidance of $11–$12, commissioned the 105 MW Pastoria Solar Project, brought the 460 MW Pin Oak Creek Energy Center into commercial operation, and received PUCT approval for the Freestone / CyrusOne data-center net-metering application.

Those facts establish that the quarter was fine. They do not explain why the quarter matters.

The reason it matters is that AI infrastructure is now running into a physical-world constraint. Hyperscalers can fund capex. They can buy chips. They can build campuses. What they increasingly cannot do on their preferred timeline is get power delivered, approved, firmed, attributed, and contracted at scale.

That is the key strategic shift. A clean MWh is useful. A firm MWh is more useful. But a clean, firm MWh that arrives when the customer’s construction schedule needs it is a different product altogether. That product requires generation, interconnection, land, permitting, regulatory credibility, retail structuring, and balance sheet capacity. Constellation is assembling all of those pieces.

The reservoir is still the anchor. But the value is now in routing the pressure to the customer’s deadline.

Freestone Shows the Product

The Freestone / CyrusOne agreement should be read as a time-to-power deal, not just a power deal.

Constellation signed a 380 MW agreement to connect and serve a CyrusOne data center adjacent to the Freestone Energy Center, with exclusivity for another 380 MW Phase 2. The PUCT approved the net-metering application, subject to conditions. On the call, management said construction is underway on the substation that will enable power delivery, expected in the fourth quarter.

That is the product in miniature. The customer gets a site, a power source, a regulatory pathway, and a timeline. Constellation gets to monetize not just energy, but compression of uncertainty.

This is why Freestone is more important as a template than as a single project. The premium is not only for the megawatt-hour. It is for taking a multi-year grid-access problem and turning it into something a customer can plan around. That is what AI infrastructure buyers actually need.

The next version of the CEG thesis should therefore be more precise than “AI needs clean power.” AI needs power on time. Constellation’s advantage is that it can increasingly sell the schedule.

Calpine Added the Routing Layer

Calpine’s immediate contribution showed up in earnings. Management said higher Q1 earnings were mostly attributable to Calpine, and that full-year guidance includes roughly $2/share of Calpine accretion. But the strategic value of Calpine is not the accretion. It is the architecture.

Before Calpine, Constellation’s story was mostly nuclear scarcity. After Calpine, Constellation can package nuclear baseload with dispatchable gas, geothermal, retail relationships, powered land, development capability, and storage options. Dominguez made that point directly, connecting Calpine’s natural gas, solar, battery storage, and natural-gas data-center capabilities to Constellation’s ability to unlock the value of the combined nuclear and gas fleet.

This is the difference between owning an asset and owning a system. Nuclear is the reservoir. Gas is the pressure valve. Retail is the customer interface. Development is the execution muscle. The balance sheet funds the next pipe. The routing layer is the integration of those parts.

The new operating metrics reinforce the point. Constellation now treats Equivalent Forced Outage Factor as a key metric for the expanded gas, oil, and pumped-storage hydro fleet after Calpine; Q1 EFOF was 4.5%. Nuclear capacity factor was 92.3%, with no non-refueling outage days.

Companies reveal their strategic direction through what they start measuring. The old Constellation was judged primarily on nuclear reliability. The new Constellation wants investors to evaluate clean baseload and dispatchable reliability together. That is what a time-to-power platform requires.

Optionality Gets Attention; Base Earnings Get a Multiple

The most important thing management is doing is not reporting earnings. It is trying to change the valuation category.

That is what the base-versus-enhanced earnings framework is about. Base earnings are the visible components: PTC-supported nuclear, long-term contracts, customer margins, contracted growth, and share repurchases. Enhanced earnings are the variable components: power prices above assumptions, commercial margins above long-term averages, gas utilization, and volatility capture. Management reiterated a long-term outlook of 20%+ base EPS growth through 2029, anchored by visible drivers including the nuclear PTC, long-term contracts, and durable customer margins.

This is management fighting its old multiple.

The 2024 AI-power narrative made Constellation exciting because its uncontracted nuclear fleet looked like a call option on hyperscaler demand. Every available MWh represented a future deal. That narrative can drive a stock up very quickly. It can also reverse when investors ask why the next deal has not arrived.

Management’s new message is better: the option still matters, but the goal is to convert the option into base. A scarcity option trades on anticipation. A contracted platform compounds on execution. Optionality gets attention; base earnings get a multiple.

This is not a retreat from the AI thesis. It is the maturation of it. Constellation is trying to make the AI story investable by moving earnings from “possible” to “contracted.” The route from one to the other is the routing layer.

PJM Controls the Endpoint

The demand side is not the problem. Management said demand for additional compute, and therefore additional power, has not slowed; projected 2026 hyperscaler spending is nearly 75% higher than last year and still being revised upward.

The constraint is permission.

Dominguez said PJM has put forward a market-based solution to address incremental capacity needs from large-load growth, with a proposed timeline to submit a framework to FERC in June. He called the timeline “faster than we had hoped.” Constellation has also submitted about 5,000 MW of new capacity resources into PJM’s interconnection queue, including nuclear uprates, new gas generation, and battery storage.

This is the scale version of Freestone. ERCOT showed one template. PJM determines whether the template can become a much larger business.

The customer wants to turn on the tap. Constellation has the reservoir, pumps, customer relationships, and capital. PJM controls how quickly the pipe gets connected.

That is not a reason to dismiss the thesis. It is the thesis. In a regulated physical infrastructure market, the company that can navigate rules, queues, cost allocation, and customer urgency is the company that turns scarcity into cash flow.

What Can Break the Thesis

There are three ways this can go wrong.

First, Freestone could remain a one-off. If Constellation signs more nuclear-only PPAs but fails to replicate powered-land or bundled reliability transactions, the routing-layer thesis is still directionally right but less valuable than we think. The product would be clean-firm power, not time-to-power.

Second, PJM could slow the route. Customer demand is real, but contracts need rules. If the June filing slips, if bilateral contracting is delayed, or if political pressure changes cost-allocation economics, Constellation’s monetization cadence slows. That would not make the reservoir less valuable. It would delay the moment when reservoir value becomes base earnings.

Third, operational excellence must hold. The reservoir has to work for the pipes to matter. Q1 nuclear capacity factor of 92.3% was acceptable given planned refueling, and there were no non-refueling outage days, but this remains the metric that underwrites the entire premium.

These are real risks, but they are risks to timing and valuation more than to the core structural shift. Power delivery has become the bottleneck. Constellation is one of the few public companies built to solve it.

The Math, Subordinated

The valuation debate is not about next quarter’s EPS. It is about whether optionality becomes base earnings.

The middle case is no longer “merchant nuclear with upside.” It is a delayed platform: valuable, compounding, and underappreciated, but not yet fully reflected in estimates.

The entry point matters. Below $285, the risk/reward is supported by management’s own behavior: Constellation repurchased about 1.2 million shares at roughly that price, deploying $335 million, and Dominguez called it an intentional statement that the stock was a compelling use of cash.

The Scoreboard

The thesis now has six practical markers.

Contracted clean MWh should move toward 70M+ MWh for 2030 delivery by year-end 2026. The enhanced-to-base mix should fall below 35% by 2027. Nuclear capacity factor should sustain 93–94%+ outside planned outage noise. PJM should file on time and allow the bilateral process to begin. Buybacks below $300 should be meaningful, not symbolic. And most importantly, powered-land cadence should accelerate: more than 1 GW signed under the Freestone template in 2026 would confirm that time-to-power is repeatable.

That is the scoreboard. Not quarterly EPS. Not rumor flow. The scoreboard is whether pressure becomes base earnings.

The Bet

The reservoir thesis is intact. Reliable clean power is more valuable than ever. Calpine made the reservoir more flexible. Freestone showed the product. PJM will determine the pace. Free cash flow gives management time: Constellation now expects $8.4 billion of free cash flow before growth across 2026–2027 and $11.5–$13 billion across 2028–2029.

Our view has evolved, but the direction is consistent. In 2025, Constellation sold pressure. In April, it became a power platform. After Q1, the sharper formulation is that Constellation sells the date the pressure arrives.

That is more valuable than power because time is now the constraint. It is also why Constellation’s advantage is widening. The company is not waiting for the AI power market to arrive. It is building the routing layer through which that demand will flow.

The reservoir is full. The pipes are no longer theoretical. The question now is how quickly the customer flow begins.

Subscribe now

• DoorDash’s 10%+ post-earnings rally suggests investors are no longer scoring the company mainly on revenue or take rate, but on GOV growth, order scale, EBITDA, and evidence that CPG advertising is beginning to work.

• The core debate is whether grocery can reach break-even at scale; if it can, fulfillment becomes the cost of admission to a much higher-margin advertising business.

• By 2029, DoorDash is either reclassified as a media company with logistics attached, or remains a well-run delivery business priced too richly for its underlying economics.

A number can be true and still lie.

DoorDash reported first-quarter revenue of $4.04 billion, up 33% YoY. The number missed expectations. Net Revenue Margin fell to 12.8%, the lowest in five quarters. Free cash flow was $420 million, down from $494 million a year earlier. Adjusted EBITDA as a percentage of GOV fell to 2.4%, down from 2.6% in both the year-ago quarter and the sequential quarter.

Then the stock went up 10%+.

A stock does not rise sharply after missing revenue unless investors have decided revenue is no longer the number that matters most. The market chose a different set of facts: 933 million orders, $31.6 billion of GOV up 37%, adjusted EBITDA of $754 million, and a Q2 GOV guide of $32.4–33.4 billion against a Street estimate of $32.2 billion. Buried in the Q&A, Tony Xu offered the sentence that almost nobody flagged as the most important thing said all afternoon: DoorDash had, in his words, “cracked the code on CPG advertisers.” Almost no analyst follow-up. No mention in the press release. No quantification.

The market was not saying the quarter was perfect. It was saying the old scoreboard was incomplete. This piece is about what the new scoreboard is actually measuring, and what it implies for where DoorDash is worth in 2029.

In prior pieces we argued that DoorDash was running the Zomato playbook, not the Meituan one; that the Digital Shelf, grocery-enabled CPG advertising, was the single variable determining whether DoorDash was a $150 stock or a $300 stock; and that the Tudor reliability framework explained both the moat and the threat. Q1 didn’t change that thesis. It advanced it.

Two Layers, Two Economies

Local commerce delivery contains two distinct economic layers that are routinely conflated. Distinguishing them is the analytical move that makes everything else clear.

The fulfillment layer, moving items from merchants to consumers, is logistics. Marginal costs are real and irreducible. Labor scales linearly with volume. Margins are bounded by the physics of moving atoms through space. The end-state of mature fulfillment markets is duopoly economics, not winner-take-all.

The intent layer, capturing what consumers want before they transact and monetizing that intent through merchant-paid placement, is media. Marginal costs approach zero. Scale economics are non-linear. The structure rewards whoever owns the largest pool of high-intent consumer attention in a given category.

These two layers have different cost curves, different competitive dynamics, and should command different multiples. The strategic question for any local commerce company is not “what are your delivery margins” but “which layer are you actually building, and how far along are you?”

Why Restaurants Capped the First Business, and Why Grocery Doesn’t

DoorDash spent its first decade dominating restaurant fulfillment. What’s been less examined is why restaurants, despite enormous order volume and rich intent data, couldn’t anchor a transformative media business.

The restaurant advertising market is structurally small. A typical independent restaurant operates with $30,000 to $50,000 of annual marketing budget. Across DoorDash’s merchant base, the addressable advertising opportunity caps at $1 to $1.5 billion. Meaningful, but not transformative. The deeper problem is the demand side: the advertisers paying restaurants to be visible are local SMBs, not large branded buyers running national campaigns. Restaurants gave DoorDash the capability to build a media business but not the demand side to make it large.

Grocery changes the demand side on every dimension that matters. Coca-Cola, Procter & Gamble, Unilever, Kraft Heinz spend 10 to 15 percent of revenue on marketing. Global CPG advertising exceeds $200 billion annually and is migrating toward retail media, where attribution to a purchase is direct rather than inferred. The catalog is different, SKUs map to brands with ad budgets attached. The intent signal is different, grocery purchases are routine, predictable, high-frequency.

The reference case is Amazon. Amazon retail runs at thin margins; Amazon advertising now generates over $50 billion annually at near-pure profit. Retail is the cost of admission to operate the media business at the necessary scale.

But grocery introduces a problem restaurants didn’t have. Restaurants make the order after it arrives; grocery promises a shelf the platform can’t see accurately. The app says strawberries are available; the shopper arrives to find them gone. This is the Tudor problem, Frederic Tudor didn’t get rich because he sold ice; he got rich because he made ice reliable, and reliability required controlling the entire chain. DoorDash cannot monetize grocery intent if it cannot reliably deliver what was promised on the digital shelf. DashMart is the response. Whether DoorDash can solve grocery reliability at marketplace scale is the operational question that determines whether the second business unlocks.

The Break-Even Threshold

Here is the structural claim that defines the entire investment debate: grocery delivery doesn’t need to be profitable. It needs to break even at scale.

Once grocery delivery is unit-economic neutral, every additional order is a free option on the larger advertising business. The fulfillment layer becomes the cost of admission; the media layer becomes the unlocked profit pool. Below break-even, grocery is a margin drag and the media business doesn’t unlock. Above break-even, grocery becomes the substrate, the advertiser pool grows, the data improves, and the company gets revalued.

The competitive race in local commerce is not about who delivers groceries most efficiently. It’s about who reaches grocery break-even first, with sufficient consumer scale to attract CPG advertising at meaningful volume.

Q1 advanced this variable materially. Management disclosed for the first time that new verticals are expected to turn gross profit positive in the second half of 2026. The CPG advertising comment in Q&A was the qualitative companion. Together they form the most consequential disclosure on the call. They are also why the market repriced.

What We Got Right, What We Underweighted

We argued the Digital Shelf was the only variable that mattered for terminal valuation. Q1 confirmed this, the market reacted not to revenue or NRM but to evidence the second business was advancing. We argued the Tudor frame would explain both moat and constraint; the blind storefront problem remains DoorDash’s central operational challenge, and DashMart’s role as a reliability solution is now visible. We argued organic growth at DoorDash’s scale was more durable than the saturation thesis implied. U.S. restaurant GOV growth above the 16-quarter average at 67% market share validates this.

What we underweighted was the speed at which the market would shift scoreboards. Our base case treated Q1 as a modestly positive print where investors began testing whether the framework had changed. Instead, the +16% reaction signals investors are already pricing the optionality of the second business, not waiting for confirmation, not requiring quantification, accepting directional commentary as sufficient evidence to begin repricing. That is a meaningful signal about how this market reads the company.

The implication is uncomfortable for the patient bull. The asymmetry is no longer “wait for proof, then add.” The market will price most of the optionality before the proof arrives. What’s left for the patient investor is the gap between optionality pricing and full reclassification pricing, the spread between a market that believes DoorDash might become a media company and a market that has accepted DoorDash is one. That spread closes only when advertising disclosure becomes specific enough to defend in numbers rather than words. Until then, the stock will trade on incremental directional evidence, with each quarter’s tone potentially worth more than each quarter’s numbers. This is unfamiliar terrain for investors trained to wait for fundamentals.

Three Worlds, 2029

The consensus model treats DoorDash as a logistics business with an advertising option. The variant perception is that this is the wrong terminal frame. By 2029, DoorDash is either a media company with a logistics cost of goods sold or a well-run logistics business priced as something it isn’t. The middle outcomes are unstable.

• The Reclassification Completes (25%). Grocery reaches break-even by mid-2027. Advertising scales to 12-15% of revenue by 2029, growing 30%+ annually. Deliveroo delivers $250M+ EBITDA by 2027. CPG advertiser concentration becomes large enough that DoorDash discloses it publicly. Revenue CAGR 2025-2029 of 22%, reaching $30B by 2029. EBITDA margin expanding to 22%. 2029 EBITDA of $6.6B at 22x, DoorDash gets repriced as retail media. Implied price: $310.

• The Market Settles (50%). Grocery reaches break-even by H2 2027. Advertising reaches 8-10% of revenue by 2029. Deliveroo hits the $200M target. New verticals reach contribution breakeven but don’t drive structural margin expansion. Revenue CAGR of 17%, reaching $26B. EBITDA margin at 19%. 2029 EBITDA of $4.9B at 18x, the market accepts both businesses but doesn’t fully reclassify. Implied price: $210.

• The Refrigerator Arrives (25%). Grocery doesn’t reach break-even at marketplace scale; the reliability tax persists. CPG advertising flows to first-party retailers (Walmart, Amazon, Instacart) rather than third-party delivery platforms. AI agents emerge as the primary surface for local commerce intent, capturing the value DoorDash hoped to monetize. Revenue CAGR of 12%, reaching $22B. EBITDA margin compresses to 15%. 2029 EBITDA of $3.3B at 13x, logistics multiple. Implied price: $120.

Probability-weighted price: ≈$215. The asymmetry sits almost entirely in the multiple, not the operating numbers. The 2029 EBITDA range across the three scenarios is $3.3B to $6.6B, a 2x spread. The multiple range is 13x to 22x, a 1.7x spread. But the price range is $120 to $310, a 2.6x spread, because the multiple and EBITDA compound rather than offset. What this means is that the scenarios diverge primarily on what kind of company the market decides DoorDash is, not what kind of company it operationally becomes. The fundamental question over the next three years is not “how fast does DoorDash grow” but “what does the market decide it has been growing into.” At ~$200, the stock prices roughly the base case with modest weight to reclassification.

What to Watch

Five signposts will determine which world materializes. Advertising disclosure is the most important, the moment DoorDash breaks out advertising as a percentage of revenue is the moment the second business is real enough to defend in numbers rather than words. New verticals gross profit positive in H2 2026 must be confirmed in Q3 commentary; deferral materially weakens the thesis. NRM stabilization at 12.5-13.0% while EBITDA/GOV improves means lower take rates are buying something; below 12.5% with no offsetting margin expansion means structural pricing pressure. Deliveroo hitting the $200M FY EBITDA target shifts the international narrative; missing reopens the capital allocation debate. AI agent positioning, whether DoorDash announces proprietary agentic features at scale, and whether any major LLM platform announces direct delivery integration that bypasses DASH, answers the refrigerator question.

The Question That Matters

The Q1 print didn’t answer the structural question. It advanced the timeline on which the answer becomes knowable. By 2029, DoorDash is either operating two businesses on top of each other or one business priced as if it were two. There is no comfortable middle outcome. The market has begun pricing the optionality of the second business; full revaluation requires the disclosure and execution that hasn’t happened yet.

The ice houses are still useful. The question over the next three years is which of these worlds we’re actually in.

Subscribe now

• The market’s Uber framework changed: despite a revenue miss and weak GAAP EPS, the stock rose because investors focused on bookings growth, non-GAAP earnings, and stronger-than-expected guidance.

• Uber One completes the AV thesis: 50 million members now drive half of gross bookings, suggesting Uber’s demand layer is becoming recurring, bundled, and harder to bypass.

• The next leg of repricing depends on proof: Uber needs to show AV take-rate economics, continued Uber One penetration, Delivery margin expansion, and insurance-driven Mobility acceleration.

Uber reported first-quarter revenue of $13.2 billion, missing consensus by ninety million dollars. GAAP earnings per share was thirteen cents against expectations of seventy. And the stock went up ten percent.

That reaction is more interesting than anything in the financial statements. A stock does not rise ten percent on a revenue miss and an earnings collapse unless investors have decided they were measuring the wrong thing. What happened on May 6th was not a response to a good quarter. It was a framework change. The market looked past revenue, looked past GAAP earnings, looked past the $1.5 billion non-cash hit from equity revaluations in Didi and Grab, and decided that the things that actually matter, gross bookings growth of twenty-one percent constant currency for the third consecutive quarter, non-GAAP earnings per share up forty-four percent, and a Q2 guide that cleared every estimate on the Street, were the numbers worth paying for.

I want to sit with that for a moment, because it changes the posture of everything I have been writing about this company.

For eighteen months I have argued that the market was looking at Uber through the wrong lens. The prevailing fear was autonomous vehicles: if cars drive themselves, who needs a platform connecting riders to drivers? Uber becomes the middleman that technology eliminates. My view was the opposite. I argued that the scarce resource in an autonomous world would not be the car itself but utilization, keeping expensive fixed-cost assets productively busy. Human-driven Uber supply is variable cost: drivers bring their own car, absorb the depreciation, and turn on the app when they feel like earning. Robotaxis are fixed cost: someone owns a hundred-thousand-dollar asset that depreciates whether it is moving or not. At those price points, idle time is not a missed opportunity. It is a financial disaster. The company that solves utilization wins the autonomous era, regardless of who manufactures the vehicle.

That was the through-line of every prior article, Ford’s system versus Ford’s car, Microsoft’s Windows versus the PC manufacturers, Insull’s demand programming versus his generators, the lighthouse that enables the fleet rather than profiting from individual ships. Uber’s real asset was not the driver or the vehicle. It was the coordination layer that made capacity useful.

Every quarter I made this case, the market disagreed. The stock traded between sixty-eight and eighty dollars. Analysts maintained hundred-dollar-plus price targets that the market refused to validate. Q4 was the sharpest disconnect: Uber printed $9.8 billion in annual free cash flow and the stock fell seven percent.

Then came Q1. And for the first time, the market agreed.

That creates a new problem for someone who has been writing against consensus for five articles. If the market is beginning to see what you see, the question is no longer “what is the market missing?” It is “how far does the repricing go, and what determines the distance?”

The answer, I think, is that the repricing has only started. Because Q1 did not just confirm the old thesis. It completed it.

The Thesis Had a Hole

The utilization argument was always a supply-side thesis. It explained why autonomous vehicle companies might need Uber: because expensive fleets need demand aggregation to achieve acceptable returns on capital. Waymo can deploy a thousand cars in San Francisco, but if those cars sit idle forty percent of the day because consumers have not built the habit of opening the Waymo app, the return on invested capital is catastrophic. Uber, with two hundred million monthly consumers, solves that problem on day one.

This logic holds up. Uber said this quarter that it now has more than thirty autonomous partners, that AV Mobility trips grew more than ten times year over year, and that the company remains on track to facilitate AV rides in up to fifteen cities by year-end. It launched Uber Autonomous Solutions, fleet management, insurance through Marsh and Apollo, financing through Santander, charging infrastructure, which is Uber explicitly productizing the coordination layer for AV operators. The message to every robotaxi company is the same: you build the driver; we will build everything else.

But the supply-side argument always left a demand-side question unresolved. If Uber’s value depends on owning the consumer relationship, how strong is that relationship really? Is it habitual and locked in, or is it transactional and easily redirected? A consumer who opens the Uber app because it happens to be the most convenient option today could just as easily open a Waymo app or a Tesla app tomorrow if the price or experience is better. Convenience is not a moat. It is a temporary advantage that erodes the moment someone offers a more convenient alternative.

That was the hole in the thesis. And Q1 filled it.

Fifty Million Reasons

The most important number in the quarter was not revenue, not earnings, not bookings. It was this, from Dara Khosrowshahi’s opening remarks: Uber One has reached fifty million members, and those members now drive half of all gross bookings across Mobility and Delivery.

Half. From roughly one-quarter of the monthly active user base.

This changes how the demand layer should be understood. A marketplace with two hundred million monthly consumers has demand. A marketplace where fifty million subscribed members generate half of the economic activity has something qualitatively different: a recurring relationship with its most valuable customers. These members spend three times more than non-members. They retain at higher rates. They use multiple products at one and a half times the rate of single-product consumers. And the program is growing at fifty percent year over year, which means Uber added twenty million members in twelve months.

The behavioral mechanism is what matters. Once a consumer pays ten dollars a month for ride credits, delivery fee waivers, and hotel discounts, the marginal decision changes. The question is no longer “should I use Uber for this?” It becomes “why wouldn’t I use Uber, since I already pay for it?” Every new vertical Uber adds, hotels, grocery, parking, travel, becomes another reason to justify the membership, which increases engagement across all verticals. The bundle is the moat, not any individual product.

This is why the hotel launch through Expedia is more interesting than it looks. Khosrowshahi explained on the call that Uber Reserve, the pre-booking product originally built for airport trips, proved the company could move consumers from on-demand behavior to planned behavior. He then connected this to the travel opportunity: airports represent roughly fifteen percent of Mobility bookings, forty percent of U.S. riders take trips outside their home city, and Uber saw 1.5 billion trips outside users’ home cities last year. Hotels are not a random adjacency. They are an attempt to capture movement intent earlier in the journey and wrap it in Uber One economics.

The pattern is the same across every expansion: use one behavior to seed the next. Rides seed delivery. Delivery seeds grocery. Reserve seeds travel. Travel seeds hotels. Uber One ties them together. Each new surface increases frequency, which deepens the membership value proposition, which makes the next expansion easier to justify.

This is the missing half of the utilization thesis. The first version explained why supply might need Uber. The second version, the one Q1 revealed, explains why demand is becoming harder to redirect away from Uber. A consumer locked into a membership bundle that spans rides, food, grocery, and hotels is not switching to Waymo’s standalone app to save two dollars on a ride. The switching cost is the bundle, not the individual transaction.

And that matters for everything. It matters for delivery margins. It matters for advertising. It matters for hotel conversion. It matters for the economics of autonomous vehicles. Predictable, recurring demand is exactly what expensive supply needs, whether that supply is a human driver, a courier, a restaurant, or a robotaxi.

The first version of the Uber thesis was about keeping supply busy. The completed version is about keeping supply busy with demand that is locked in.

The Question That Narrowed

The AV bear case has not disappeared. But it has become more specific, which is progress.

The old question was existential: will autonomous vehicles destroy Uber? The new question is economic: what take rate and margin does Uber earn when AV supply plugs into its network?

Q1 moved the evidence in Uber’s favor. Management said category position in San Francisco and Los Angeles, the two U.S. cities with the most autonomous vehicle exposure, expanded over the past six months. In Austin and Atlanta, where Waymo operates exclusively through Uber, driver earnings are up and more drivers are joining. Khosrowshahi said explicitly: “At this point, we don’t see any effect of the Waymo launches on our overall business.”

If AVs were disintermediating Uber, these metrics would be moving the other direction. They are not. The evidence is consistent with the category expansion thesis, autonomous vehicles grow the total ride-hailing market rather than cannibalizing Uber’s share of it.

But I want to be honest about what remains unproven.

Uber still has not disclosed the single number I called the most important metric in my February article: the take rate on autonomous rides. Management gave us “ten times year-over-year growth” in AV trips, a relative metric without a base. They launched an entire autonomous services suite. They signed financing and insurance partnerships. But they did not tell us what percentage of an autonomous ride’s economics Uber captures.

The charitable interpretation is that the sample size is too small and the commercial model is still being established. Management does not want to anchor a number that will evolve as AV deployment scales. The less charitable interpretation is that the number is not yet impressive enough to share. I do not know which is true. But I do know that a thesis about Uber becoming the coordination layer for autonomous mobility requires, at some point, evidence that the coordination layer earns an attractive toll. Structural arguments about utilization and demand aggregation are necessary. Disclosed unit economics are what makes them investable.

If Uber reports Q2 without disclosing AV take rates, I will interpret that as a signal the economics are not yet where management wants them. If take rates are disclosed at twenty percent or higher, the thesis strengthens materially. This is the single most important number Uber is not sharing.

The Hidden Flywheel

The most underappreciated part of the quarter was not about AVs or membership. It was about insurance.

Uber’s CFO said the company expects hundreds of millions of dollars in insurance savings in 2026, driven by policy work, technology improvements, favorable renewals, and increased risk transfer to third-party carriers. Then he said the more interesting part: Uber is passing some of those savings back to consumers through lower ride prices, and the lower prices are driving measurable trip acceleration, particularly in Los Angeles and across California.

This is a compounding mechanism hiding in the cost structure. More trips generate more actuarial data. Better data reduces insurance cost per trip. Lower insurance costs enable lower consumer prices. Lower prices drive more trips. The cycle repeats.

The reason this matters for the broader thesis is that it is a scale advantage no competitor can replicate without Uber’s data. Uber’s captive insurance subsidiary processes claims from billions of historical trips across thousands of cities. A standalone AV fleet with a thousand vehicles has essentially no actuarial history. The insurance cost gap between Uber and a new entrant could be fifteen to twenty-five percent, and that gap widens with every additional trip Uber processes.

In my Q3 article I used Samuel Insull’s electrification of Chicago as an analogy for how Uber programs demand to fill idle hours. The insurance flywheel is a different expression of the same underlying dynamic. Insull drove down unit costs by improving utilization. Uber is driving down unit costs through data-driven risk reduction. Both are mechanisms where scale advantage compounds rather than dissipates.

The CFO said this would be the first year since COVID where Uber expects good leverage on the insurance cost line for U.S. Mobility. If Los Angeles is the leading indicator, where trip growth is running significantly above the California and national averages, then the insurance flywheel could add two to three percentage points of Mobility trip growth over the next twelve months. That is not in consensus estimates. And it is the kind of quiet, operational compounding that drives durable outperformance rather than quarterly beats.

The Honest Reckoning

Being early on a thesis is not the same as being right about all of it. Five articles in, with the market beginning to agree, this is the right moment to confront the tension I find most uncomfortable.

Uber One’s success creates a concentration question that the bull case tends to gloss over. Fifty million members driving fifty percent of bookings is a powerful flywheel. It is also a dependency. One-quarter of the user base generating half the economics means Uber’s growth is increasingly driven by deepening engagement with existing high-value members rather than expanding the audience.

Monthly active consumers actually declined sequentially from 202 million in Q4 to 199 million in Q1. Management did not mention this on the call. They highlighted year-over-year growth, the Uber One milestone, and trip growth instead. The sequential decline is almost certainly seasonal, Q4 holiday demand brings in occasional users who do not return in January. But the pattern it suggests deserves attention: fewer users doing more trips, spending more per trip, locked in through membership. The platform becomes more resilient to competitive threats but more vulnerable to anything that disrupts the high-frequency member cohort specifically.

No such disruptor exists today. But the shape of the risk is worth naming. If a competitor, Waymo with a superior autonomous experience, or a hypothetical DoorDash membership bundle that undercuts Uber One on price, figured out how to poach the top quartile of Uber’s member base, the impact on bookings would be disproportionate to the number of users lost. The membership concentration that makes Uber’s demand layer so powerful is the same concentration that makes it fragile at the margin.

I do not think this is an imminent risk. But I think it is the right risk to monitor, because the bulls, including me, have spent more time celebrating Uber One’s growth than stress-testing what happens when growth slows.

Where the Remaining Distance Lives

The market moved from pricing Uber as a disruption victim at sixty-eight to seventy-three dollars to pricing it as a platform with a plausible thesis at eighty. That is meaningful. It is also not finished.

At eighty dollars and twenty-four times forward earnings, the stock reflects acceptance that Uber is not being displaced in the near term, that bookings growth is durable, and that earnings are compounding at twice the rate of revenue. What it does not yet reflect is Uber being valued as infrastructure, the demand layer that fragmented supply must flow through, with a membership base that makes that demand recurring rather than transactional.

The distance between eighty dollars and what I believe the business is worth, somewhere around a hundred and twenty to a hundred and thirty in the base case, depends on three things over the next twelve months. None of them are about AV trip counts.

The first is Uber One proving it is infrastructure rather than a loyalty program. If membership reaches sixty-five to seventy-five million and drives fifty-five to sixty percent of bookings with retention rates holding, the market will have to start valuing the membership base the way it values recurring revenue in other businesses. A platform where the majority of economic activity comes from pre-committed members is a different animal than a marketplace where consumers choose transactionally each time. It commands a different multiple.

The second is Delivery margins reaching four percent of bookings and proving that demand routing generates real profit. Delivery is the working proof that the coordination thesis produces financial results, not just strategic narratives. Fifteen billion dollars of Delivery run-rate bookings already originate from the Mobility app at zero incremental customer acquisition cost. If the margin on that routed demand continues expanding, it validates the same mechanism that will eventually make AV coordination profitable.

The third is the insurance flywheel showing up in sustained U.S. Mobility acceleration. If California trip growth continues outpacing the national average through Q2 and Q3, this moves from hidden advantage to consensus growth driver, and from there into models that currently do not include it.

The AV thesis remains the long-term option value. But the near-term repricing depends on the platform thesis. And the platform thesis, as of Q1 2026, is no longer speculative. It is accumulating evidence.

The View From Here

For five articles I argued that the market was focused on the wrong thing. That autonomous vehicles were not the threat the stock price implied. That Uber’s coordination layer, its ability to keep supply busy, to route demand efficiently, to program idle hours into productive ones, was being ignored while the market fixated on a future where robots replace drivers.

Q1 was the quarter where the market started listening. Not because the numbers were perfect, revenue missed, GAAP earnings were ugly, free cash flow barely grew. Because the numbers that actually measure the business pointed in a single direction: the platform is compounding, demand is locking in, and the coordination layer is getting stronger.

The thesis I started with was about busy cars. Utilization as the scarce resource. Uber as the system that keeps expensive assets productive.

That thesis was incomplete.

What Q1 revealed is the other half: the demand that makes utilization valuable is becoming bundled, habitual, and harder to bypass. Fifty million members driving half of bookings. Cross-platform consumers growing faster than single-product users. Insurance savings cycling back into demand acceleration. Delivery margins expanding because routed demand converts at higher margins than acquired demand.

The first version of the Uber thesis was about supply needing a coordinator. The completed version is about demand becoming the thing that supply cannot afford to ignore.

That is a more powerful thesis. It is also a more testable one. The evidence is arriving quarter by quarter. The market has started to reprice it. And at eighty dollars, I believe there is considerable distance left to travel.

Subscribe now

• The Ontology was valuable before agents. Palantir’s core asset is its ability to map institutional reality: objects, relationships, permissions, workflows, and actions, across fragmented systems; agents simply multiply the traffic moving through that map.

• Q1 showed the architecture monetizing. Revenue grew 85%, U.S. revenue grew 104%, adjusted operating margin hit 60%, net dollar retention reached 150%, and FY26 guidance moved to 71% growth, numbers that look less like normal software and more like infrastructure compounding.

• The variant view is that complexity is Palantir’s demand generator. The market sees an expensive AI software winner; the more interesting possibility is that falling inference costs create more agents, more actions, and more governance needs, making the Ontology more valuable as AI spreads.

In the age of steam, factories were built around a central drive shaft. A single engine powered it, and belts and pulleys distributed force to every machine on every floor. The architecture of the building, where equipment sat, how work flowed, where bottlenecks formed, was dictated not by the logic of production but by proximity to the shaft.

When electricity arrived, factory owners did the obvious thing: they replaced the steam engine with an electric motor and kept the shaft. Productivity improved modestly. Then someone realized the real insight. If each machine has its own motor, the shaft is unnecessary. And without the shaft, the entire architectural logic of the factory dissolves. Machines can be arranged by workflow instead of proximity to power. Henry Ford’s Highland Park plant in 1913 was designed for electricity rather than adapted to it. Productivity didn’t improve five percent. It improved tenfold. The infrastructure didn’t just power the old machines. It made the architecture that housed them obsolete.

That is the right analogy for Palantir after Q1.

Most enterprises are in the first phase of AI adoption. They are bolting models onto the old line shaft of enterprise software, a copilot in Salesforce, an assistant in ServiceNow, a summarizer in Office. Useful, yes. Transformative, not yet. Palantir is arguing the shaft can come out. If the Ontology becomes the operating map of the enterprise, where objects, relationships, permissions, workflows, and actions are defined, then many applications stop being systems of truth and become interfaces. Some will remain important. Others will have to justify why they exist at all.

What We Said, What Happened

In October, I argued that Palantir had built the modern equivalent of MS-DOS, a coordination layer for incompatible enterprise systems, valuable not for any feature but for the coordination it enabled across an otherwise chaotic ecosystem. In February, after Q4 delivered 70% growth and 57% margins, I updated the thesis using the Toyota Production System: the value lived in the connections between machines, not the machines themselves. I set five conditions and scored four of five as met or exceeded. The old bull case became the new base case.

Then Q1 exceeded the updated bull case across every dimension that matters.

The mechanism we identified, coordination value appreciates with institutional complexity, now has a name from management. Shyam Sankar stood on the earnings call and articulated what is functionally the Jevons Paradox applied to enterprise AI: as inference costs collapse, token consumption explodes; as consumption explodes, the complexity of governing agent workflows grows exponentially; and as complexity grows, the Ontology becomes more essential, not less. We predicted the dynamic. The company’s CTO described it back to us using different words.

But Q1 also revealed something the framework didn’t fully anticipate. In February, I wrote that the Ontology coordinated between applications, the wiring between machines. What Q1 showed is that the Ontology has moved past coordination into absorption. That is the line shaft coming out. And it changes the thesis.

From Coordination to Absorption

The key sentence on the call was not Karp’s “N of 1” flourish. It was Sankar’s claim that AIP “replaces static workflows not by replicating the playbook, but by eliminating the need for one.” The examples that followed were specific. Thomas Cavanagh Construction, where 97% of employees use Foundry daily and every other piece of software must justify its existence. Palantir itself, which replaced its old CRM with an AI-first solution built on AIP in a few months. The SAP migration example, where AIP agents achieved 99% validation accuracy in two weeks with 70% timeline and cost reduction.

Palantir is not saying every application dies. It is saying many applications lose their privileged position as systems of operational truth. In a world organized around applications, the workflow lives inside the app. In a world organized around the Ontology, the workflow lives in the institutional model, and the app becomes replaceable. That is a subtle distinction, and it is everything.

The old view of Palantir was that it connected applications, SAP over here, Salesforce over there, manufacturing systems somewhere else, Excel everywhere, and Palantir as the integration layer that makes the mess navigable. That was already valuable. But once the Ontology encodes institutional reality, customers, claims, machines, contracts, permissions, policies, and actions, the application is no longer necessarily the owner of the process. It is one possible interface to the process. The coordination layer is becoming the operating layer. The wiring is becoming the building.

Agents Are the Accelerant

The temptation is to say agents make Palantir valuable. That is backwards. The Ontology was already valuable. Agents make it more valuable, and the mechanism explains why deceleration models keep breaking.

Sankar’s Jevons argument is the missing piece. AI inference costs have fallen roughly a thousand-fold in three years. The result is not less AI usage but an explosion of agent workflows, tool calls, and autonomous actions inside enterprises. Every one of those actions requires authorization, cost attribution, and auditability. The force that commoditizes models, falling token costs, is the same force that strengthens the governance layer. Cheap AI commoditizes intelligence but increases the value of governed action.

The flywheel is straightforward:

That is why net dollar retention exploded to 150%, up from roughly 118% twelve months ago. Existing customers are not buying more seats for analysts. They are feeding a growing population of non-human actors operating inside the platform. The number of agents scales with token consumption. Token consumption scales with falling costs. The addressable market is no longer “enterprises that need coordination software.” It is every AI agent action inside a governed enterprise. That is a categorically different number.

The acceleration is now eleven quarters deep and showing no sign of exhaustion:

One quality signal the market has not absorbed. For the first time in Palantir’s history, GAAP earnings per share ($0.34) exceeded adjusted earnings per share ($0.33). Stock-based compensation fell to 12.4% of revenue from 17.6% a year ago, and the tax adjustment in the non-GAAP reconciliation now exceeds the SBC add-back. The profitability story is no longer a non-GAAP construct. GAAP net income was $871 million at a 53% margin. The earnings quality critique, the last structural bear argument standing, is empirically dead.

What the Market Is Getting Wrong

The consensus view treats Palantir as an expensive AI software stock that will eventually decelerate. I think the consensus is making three errors that compound.

The first is a category error. Palantir is being valued against SaaS companies where 15-25x revenue is expensive. But the economic signature of Q1, 85% growth, 60% margins, 150% NRR, $1.5 million revenue per employee, 70 salespeople, does not match application software. It matches infrastructure monopolies during secular demand expansions. The correct comp set is not Salesforce. It is NVIDIA during the CUDA lock-in, or Microsoft during Windows standardization.

The second error is directional. The market treats AI commoditization as a risk to Palantir. Q1 says the opposite. Growth accelerated because models commoditized, not despite it. The Jevons mechanism means every decline in token cost expands the frontier of automation, which expands the surface area requiring governance. The market is modeling a headwind that is structurally a tailwind.

The third error is about TAM. Analysts size Palantir’s opportunity against the enterprise software market. The actual opportunity is governing machine-to-machine complexity inside every institution that uses AI, a market that did not exist three years ago and grows at the rate of inference cost decline. Wrong category, wrong direction of impact, wrong TAM. Through those three lenses simultaneously, the stock looks very different than it does on a conventional screen.

Three Years From Now

Only fourteen public companies have ever sustained 40%+ revenue CAGR for three years after crossing $10 billion in annual revenue. The list includes PDD, Meta, Alibaba, Tencent, Tesla, NVIDIA, Amazon, and Alphabet. Every one is a consumer platform or semiconductor company. Zero enterprise software companies have done it. Palantir will cross $10 billion around mid-2027. If the structural thesis holds, it becomes the first.

Here is how I would frame the next three years, using the pre-print price of approximately $144 as the reference point rather than pretending precision where none exists:

The bear case is not that Palantir is fake. That debate is settled. The bear case is that Palantir is excellent but bounded, indispensable for elite, complex institutions, but not the default architecture for enterprise AI. Growth decelerates, margins stay strong but not extraordinary, and the multiple compresses. At $144, you lose modestly.

The base case is the old bull case rebased to demonstrated reality. Palantir compounds through U.S. commercial and government, NRR stays elevated, and the company crosses $15 billion by 2028 with elite profitability. The stock works, but not spectacularly, because the quality is largely recognized.

The bull case requires the line shaft to actually come out across the enterprise. Applications become interfaces, agents multiply governed actions, and Palantir becomes the operating layer for institutional AI. That is the case where Palantir enters the club of fourteen and the current price looks cheap in retrospect.

What Would Prove Us Wrong

The framework requires specific things to remain true. U.S. commercial growth below 80% for two consecutive quarters without a program-transition explanation would mean the Jevons mechanism is weaker than Q1 suggested. Net dollar retention below 130% would undermine the agent TAM thesis, if existing customers are not expanding at that rate, the non-human population is not growing fast enough. Adjusted operating margins below 50% for two quarters would reopen the consulting critique. And if management language shifts from “we cannot meet demand” to “competitive intensity is increasing,” reassess everything. Tone is the leading indicator.

International remains the open condition. It grew 37% blended in Q1, up from near-zero, but it is not yet compounding at a rate that changes the three-year math. The NVIDIA Sovereign AI OS partnership is aimed at this gap. If international commercial reaches 40%+ growth by late 2027, the runway extends materially. If it does not, the U.S. has to carry the entire story.

Economic historians estimate it took roughly thirty years from the availability of electric power to the full redesign of American manufacturing. Factory owners kept bolting motors onto steam-era shafts because they could not see that the power source had changed the optimal architecture, not just the energy bill. The enterprises ripping out their shafts today, redesigning around the Ontology rather than adapting it to legacy systems, are the Highland Parks of the AI era. They will look obvious in retrospect. The rest will spend the next decade wondering why the new motor did not change very much.

Palantir is betting the world needs far more rewiring than it currently imagines. Q1 is the strongest evidence yet that the bet is right. Whether that justifies today’s price depends less on the quality of what is being built and more on how quickly the rest of the world figures out that the shaft is already obsolete.

• The business got better: KLA’s moat widened through higher process-control intensity, advanced packaging momentum, services growth, and sustained share leadership.

• The selloff was rational: KLA is priced like the best AI equipment story, but for now it is growing slower than etch/deposition peers because the AI fab cycle is still in the capacity-building phase.

• The key question is conversion: Gross margin recovery, memory mix rebound, free-cash-flow conversion, and FY2027 EPS revisions will determine whether the complexity tax becomes near-term shareholder returns.

KLA beat on revenue. It beat on earnings. It beat on gross margins. Management gave the most bullish forward commentary in the company’s history, explicitly breaking their own precedent to discuss 2027 growth expectations in April of 2026. Advanced packaging revenue estimates were revised upward by nearly double. The stock fell 5%+.

This is not irrational. It is informative. And it is where I want to start this update, because the gap between a nearly flawless quarter and a meaningful selloff tells you something important about where KLA sits in the market’s mental model of the AI infrastructure buildout.

The Complexity Tax

Two weeks ago, in A Perfect Business at an Imperfect Price, I argued that KLA is best understood not as a semiconductor equipment company but as a tax on semiconductor complexity. The company does not sell the tools that build chips. ASML, Applied Materials, and Lam Research do that. KLA sells the tools that determine whether the chips you built actually work. As chips become larger, more customized, more vertically stacked, and more packaging-intensive, the cost of not knowing what went wrong in the manufacturing process rises exponentially. KLA monetizes that rising cost.

That framing still holds. In fact, this quarter strengthened it. Management tied demand not to a generic “AI is good for semiconductors” narrative but to a specific, multi-dimensional complexity story: leading-edge foundry and logic, HBM stacking, custom silicon for hyperscalers, advanced packaging, larger die sizes, faster product cycles, rising design variability, and higher-value wafers. Each of these independently increases the amount of inspection and metrology a fab requires. Together, they represent a structural expansion of KLA’s tax base.

The compounding loop I described in the prior article is visibly at work. KLA reported 58% process-control market share, up 360 basis points since 2021 and roughly seven times larger than the nearest competitor. It achieved the number-one position in advanced wafer-level packaging. Services revenue grew 16% year-over-year. The installed base is expanding, the switching costs are deepening, and the proprietary defect data that trains KLA’s detection algorithms is accumulating across every advanced fab on earth.

But the question this quarter needed to answer was not whether the loop exists. We established that in the prior article. The question was whether it is accelerating, stable, or at risk. Three things could slow it. First, computational lithography and AI-driven process simulation could begin substituting software for physical inspection, reducing the number of tools customers need per wafer. Second, a competitor could assemble enough defect data through synthetic generation or customer collaboration to close KLA’s three-decade data advantage. Third, chip economics could shift such that customers preferred accepting lower yields to paying the rising complexity tax, which would only happen if the devices being manufactured became less valuable, not more.

This quarter provided evidence against all three. Process control intensity rose, with the semiconductor process control systems business growing over 20% against a wafer equipment market growing roughly 17%. No competitor gained meaningful share; KLA widened the gap. And the economics moved further in KLA’s favor. Advanced packaging, where a single sub-micron particle trapped during hybrid bonding can destroy an entire HBM stack worth tens of thousands of dollars, is the most extreme expression yet of the asymmetry between what inspection costs and what its absence costs. The loop is not just intact. It is compounding into new domains faster than alternatives are emerging.

That is the structural answer. The stock, however, fell because the market already knew the business was exceptional. What it wanted to know was whether that exceptionalism would show up in near-term earnings and free cash flow fast enough to justify a price that had run to nearly $1,940 in the days before the print. On that question, the quarter was good but not sufficient.

That is the tension this update tries to resolve: the business improved more than the stock reflected. Understanding why requires looking at what we got right, what we missed, and what the quarter revealed about KLA’s position in the AI cycle that our prior framework did not fully capture.

Scoring Our Own Homework

The prior article made several specific claims. Most held up. Three did not, and the three that did not are more interesting than the ones that did.

The central thesis was confirmed cleanly. The valuation warning played out precisely, the stock ran past $1,900 and sold off after a clean beat, exactly the “execution gets punished, only acceleration gets rewarded” dynamic we flagged. Services compounded within the target range. Advanced packaging exceeded even our estimates.

Where we were wrong matters more. First, the memory mix is far lumpier than our framework implied. We modeled HBM as a smooth, compounding tailwind. The yield math on 16-high stacks is correct structurally, but in the June quarter guidance, memory dropped from 38% to 18% of semiconductor process control systems revenue, a halving in one quarter. The explanation is that foundry and logic pulled forward while memory capex shifted to 2027 greenfield builds. Our steady-state model did not capture this volatility.

Second, we failed to name a structural vulnerability precisely enough. KLA’s inspection tools contain image-processing computers built with DRAM chips. When DRAM prices rise, which is broadly positive for the semiconductor industry, KLA’s own input costs rise with them. Management confirmed the headwind worsened from 75-100 basis points to a firm 100 basis points. No peer faces this circular cost exposure. Lam and Applied Materials do not have DRAM chips inside their etch and deposition tools.

Third, and most importantly, we analyzed KLA in isolation when the market prices it in comparison. In a quarter where Lam Research grew revenue 24% on the same $140 billion wafer equipment outlook, KLA growing 11% makes it the slowest grower at the highest multiple in the semiconductor equipment group. That relative gap is the single biggest reason the stock sold off, and our prior article did not frame it.

The Inspector Arrives After the Builder

This is the insight the prior article lacked, and I think it explains most of the stock’s behavior.

When a semiconductor fab gets built, there is a sequence. You install deposition chambers and etch tools first, because without them there is nothing to inspect. The tools that create capacity ship before the tools that optimize it. KLA is present throughout the manufacturing lifecycle, R&D, ramp, and high-volume production, so it is too simple to say it only matters later. But its relative growth advantage becomes most visible when the industry’s bottleneck shifts from building fabs to making those fabs yield.

The AI infrastructure buildout is in the building phase. Hyperscalers are committing historic capital, Meta raised its 2026 capex outlook to $125-145 billion. Dozens of fabs are under construction. Every incremental dollar flows first to the tools that create capacity. That is Lam territory. That is why Lam grows 24% while KLA grows 11%. It is not a share story. It is a sequencing story.

The analogy I keep returning to is cloud computing. The hardware vendors, Dell, Arista, benefited first. Then the platforms, AWS, Azure. Then the application layer, SaaS. Each wave created enormous value. The timing was staggered. AI semiconductors are following the same pattern: etch and deposition first, then inspection and metrology, then chip output and foundry revenue. KLA is in the second wave, not the first.

Management sees the second wave approaching. Rick Wallace described customer urgency around securing capacity as unlike anything in his career. Bren Higgins broke company protocol to say 2027 wafer equipment growth would exceed 2026. The backlog is building for fabs that do not yet exist. When those fabs come online and yield becomes the constraint, every deposition tool and etch chamber already installed will need KLA systems to become economically productive.

The tax collector does not arrive first. The tax collector always arrives.

The Margin That Won’t Move

If the sequencing thesis is the reason for patience, gross margin is the reason for caution.

KLA delivered 62.2% non-GAAP gross margin in the March quarter, 45 basis points above the guided midpoint. But the June quarter guides to 61.75%, a sequential decline despite higher revenue. Three headwinds are stacking: DRAM component costs at 100 basis points, tariffs at 50-100 basis points, and product mix. Management was asked directly whether KLA could pass costs through given the strong demand environment and said no, the company prices on value and cost-of-ownership improvement, not on supply scarcity. That protects the franchise. It also caps near-term margin expansion.

The 2030 target model requires gross margins expanding from roughly 62% to 63.5%. Every year of delay costs approximately a dollar of earnings per share relative to the path. This is the swing variable between the base case and the bull case. Right now, it is not swinging.

What the Market Sees vs. What’s Actually Happening

Consensus sees KLA as a best-in-class semiconductor equipment compounder levered to AI. Correct, but incomplete.

What consensus misses is that KLA is a lagged AI beneficiary whose growth inflection arrives after etch and deposition peers, not alongside them. The market is pricing KLA at 37 times forward earnings, the highest multiple in the group, while it delivers the lowest growth rate. That combination only holds if the growth gap narrows, which requires the second phase of the buildout cycle to arrive in reported numbers. That is a late-2027 to 2028 event.

What consensus is right to question is whether duration alone justifies the premium. KLA can be strategically essential and still not be the best stock to own in the earliest phase of a capacity buildout.

Our view is that the structural thesis, complexity compounds KLA’s value, is stronger after this quarter than before it. The compounding loop is widening into advanced packaging, deepening through services, and facing no credible competitive challenge. But the stock needs four to six quarters for the growth inflection to show up in the income statement. At current prices, you are paying full freight for that future.

Three Years From Here

The probability-weighted expected value is approximately $1,780. That is roughly where the stock trades. The market is pricing KLA correctly for the base case. You make money only if the bull case is more likely than the 25% the market implies.

What changed from the prior article: the probability-weighted value moved from roughly $1,600 to $1,780, reflecting the improvement in forward visibility and the advanced packaging acceleration. But the stock moved too. The conclusion has not changed. KLA is fairly priced for the base case and not cheap enough to compensate for the timing lag between the structural thesis and its financial expression.

What to Watch

Seven numbers will tell us which scenario is unfolding.

Gross margin is the most important. Above 62.5% by the September quarter means the DRAM headwind is fading and the 2030 path is intact. Below 61.5% means the margin ceiling is real. KLA versus Lam relative revenue growth is the new addition, if the gap narrows below five percentage points, Phase 2 is arriving and KLA re-rates; if it widens above fifteen, the multiple compresses regardless of absolute results. Memory mix recovery above 25% by mid-2027 would confirm the greenfield builds are converting into KLA shipments. Advanced packaging must sustain above $275 million per quarter to validate the second-engine thesis. Free cash flow conversion needs to exceed 80% for the full fiscal year; Q3’s 50% conversion was the single biggest quality blemish, and a repeat would be a genuine concern. Insider purchases remain at zero from a management team guiding to $84 in earnings per share by 2030. And consensus FY2027 EPS revisions moving up more than 5% within ninety days of the print would confirm that the selloff was a buying opportunity.

The complexity tax is permanent. The compounding loop behind it is widening. The conversion question is the only one still open, and the numbers, one quarter at a time, will close it.

Subscribe now

• The AI storage bottleneck is now real, not theoretical. Datacenter is already a quarter of revenue, enterprise SSD demand is being pulled directly by inference, KV cache, RAG, longer context windows, and agentic workloads.

• Q3 proved SanDisk has pricing power — maybe too much pricing power. Revenue, EPS, gross margin, and free cash flow all exploded, but the quarter was heavily ASP-led, which keeps cyclicality risk front and center.

• The investment case has shifted from demand to durability. The key variable is no longer whether SanDisk can earn extraordinary EPS; it is whether NBM coverage, contract guarantees, and margin floors can hold through the next downcycle.

The Thesis Has Been Walked

When I initiated SanDisk at $244, the argument was simple: NVIDIA had built the twenty-lane GPU highway, but the on-ramps were still gravel roads. AI infrastructure had a missing middle, High Bandwidth Memory was too expensive for petabyte-scale data, hard drives were too slow for active workloads, and the only technology that filled the gap was high-capacity enterprise NAND. I called it warm storage and argued it was becoming mandatory infrastructure, not a commodity input. My bull case was $450.

After Q2, at $539, I wrote that Jensen Huang had effectively confessed the problem on stage at CES. He described context memory that no longer fit inside HBM, KV cache traffic choking inference clusters, and 150 terabytes of SSD-backed context memory designed directly into Vera Rubin. Three weeks later SanDisk’s earnings confirmed the demand. ASPs were up 35-57% sequentially, gross margins were guided to 65-67%, and the company was operating in allocation, choosing which customers to serve rather than competing for orders. I set a new bull case at $1,200 and wrote that “the next 2x may be harder, but the path is visible.”

The stock is now $1,127. The path has been substantially walked. My original bull case has been exceeded by 2.5 times while we are still in fiscal 2026.

The awkward part about being right on direction is that it does not mean being right on magnitude. My framework for the architectural bottleneck was correct: datacenter is now a quarter of revenue, enterprise SSDs are being co-designed into custom silicon, and pricing power has transferred from buyer to seller. But my imagination for how violently SanDisk would capture the economics of that bottleneck was far too small. That makes this update harder than the previous two. This is not a victory lap. It is a framework reset forced by a stock that has outrun the thesis.

Q3: Numbers That Force a Reset

The numbers almost defy annotation.

Source: SanDisk press releases and earnings presentations.

Revenue beat consensus by 26%. EPS beat by 61%. Gross margin beat the company’s own guide by eleven hundred basis points. Datacenter revenue grew 233% sequentially and 645% year-over-year. Q4 was guided roughly 30% above what the sell side had penciled in.

The old question, does AI make enterprise SSDs strategic?, has been answered. Datacenter is a quarter of the business, not a future slide. Management explicitly tied NAND demand to inference, reasoning, agentic AI, KV cache, RAG, longer context windows, and token generation. The warm-storage thesis is no longer theoretical.

But this was not a clean bit-growth quarter. Revenue nearly doubled while BiCS8 shipments were down high-teens sequentially, implying ASPs exploded roughly 130-140% quarter-over-quarter. That is both the bull case and the warning sign. Pricing-driven earnings are the ones that disappear fastest when the cycle turns, which is exactly why memory investors flinch even when the numbers are extraordinary.

The Toll Booth Emerges

The more important disclosure than the headline beat is what SanDisk is doing with its leverage.

Source: SanDisk Q3 FY26 earnings call and press release.

In the initiation, I argued SanDisk could be valuable as the “intentional fifth wheel”, a strategically necessary supplier that hyperscalers would preserve to avoid Samsung/SK Hynix concentration. That framing was right but incomplete. It described why SanDisk had value. It did not describe how SanDisk would capture it.

Q3 supplied the mechanism: contracts. Five multi-year agreements with prepaid collateral and auto-execute breach clauses, backed by third-party financial instruments. SanDisk is no longer merely paving the on-ramps to the GPU highway. It is charging tolls before customers are allowed onto them.

There is an underappreciated structural reason why customers signed with SanDisk specifically. SanDisk’s joint venture with Kioxia, historically the biggest analytical knock on the company, prevents SanDisk from unilaterally redirecting capacity in a downturn. Flash Ventures requires Kioxia’s joint approval for material capacity changes. Customers will pre-fund commitments with a supplier that cannot defect from contracts under cyclical pressure. They will not pre-fund the same commitments with Micron or SK Hynix, who own their fabs outright and retain the operational flexibility to redirect bits whenever the math changes. The constraint became the credibility device. The credibility device enabled the toll booth.

But toll booths are not magic. The notes following the print captured a nuance the bulls were underweighting: NBMs improve multiple durability but may cap peak-cycle upside. If customers lock supply at fixed or semi-fixed terms near current pricing, SanDisk may be trading some peak-cycle optionality for a more durable base. That is good for the multiple. It is potentially bad for the most aggressive EPS scenarios. Durability has a price, and the price is participation if NAND keeps ripping above contracted levels.

The Problem With Being Right

My Q2 bull case was $1,200 with 25% probability. The stock is at $1,127, within 6% of the scenario I assigned one-in-four odds. Either my Q2 probabilities were wrong and the bull case deserved higher weight, or the stock has moved faster than the evidence supports. Both readings lead to the same conclusion: the analytical framework needs to be rebuilt around different variables.

The stock reaction confirms this. SanDisk fell roughly 4% on a quarter that beat consensus by 26% on revenue and 61% on EPS, with a Q4 guide 30% above the street. That is not irrational. It is revealing. Consensus was not the bar; buy-side bogeys were. The most aggressive investors were already underwriting CY26 EPS around $140 and CY27 EPS above $210. At $1,127, the stock trades at roughly 8x CY26 and 5.4x CY27 on those bogeys.

The market is not exuberant. It is skeptical. A 5x multiple on CY27 EPS is the market saying those earnings may not be durable. The market has the thesis. It is waiting for proof that the floor holds.

The New Bear Case Is Margin Durability

The demand picture is as strong as it has ever been for any semiconductor subsector. Big 5 hyperscaler capex went from $620 billion to $720 billion in three months. All five guided higher for CY27. Token consumption is growing exponentially. Agentic AI, multi-agent systems that consume 100x or more tokens per task compared to single-shot inference, is in its first year. KV cache, RAG, longer context windows, and reasoning workloads are all NAND-hungry, not HBM-hungry. The demand side of the warm-storage thesis is not just intact; it is accelerating faster than I modeled at initiation.

The bear case is no longer about demand. The bear case is about margin durability on the uncontracted portion of the book. NBMs currently cover just over one-third of FY27 bits. That means roughly two-thirds of SanDisk’s output remains exposed to spot pricing. If ASPs roll over, because Samsung eventually responds, because Micron pivots back from HBM, because hyperscalers hit a digestion cycle in late 2027, the uncontracted book gets hit. NBMs cushion the trough. They do not repeal cyclicality.

The question is no longer whether SanDisk can earn extraordinary EPS. It obviously can. The question is whether the gross-margin floor is 55%, 45%, or 30%. The answer to that question determines whether the stock is cheap, fair, or expensive at $1,127, and the answer depends almost entirely on whether NBM coverage scales past 50% and whether the variable-pricing components in years three through five of those contracts hold through a downturn.

Variant Perception: The Debate Is The EPS Floor

The consensus view is that SanDisk is in the greatest NAND upcycle in history, with AI demand providing structural support, but earnings are probably near peak and the cycle will resolve as cycles always have.

The bull view is that NBMs and AI architecture integration have changed the earnings profile permanently, that contracts convert commodity pricing spikes into infrastructure-like cash flows, and the stock deserves a multiple closer to differentiated semis than to cyclical memory.

My view is that both sides are arguing about the wrong variable. The multiple is roughly right, 5-8x near-peak earnings is what a skeptical market pays for memory at the top of the cycle, and that skepticism is historically well-founded. The debate is not about the multiple. It is about whether normalized EPS is $210 (in which case 5x is absurdly cheap) or $60 (in which case 5x is generous). The opportunity is not multiple expansion. It is proving that the EPS floor is structurally higher than old-cycle investors assume. That proof requires NBM coverage scaling, and we are at one-third, not one-half.

Three-Year Framework

The bull case requires NBM coverage above 60%, datacenter mix at 45-50%, HBF contributing revenue by FY29, and Samsung maintaining HBM-first capital allocation through 2028. The demand backdrop, $720B of hyperscaler capex rising into CY27, agentic workloads in year one, makes this more plausible than it was three months ago.

The base case requires NBM coverage stabilizing at 45-50%, margins normalizing well above historic NAND levels but well below 78%, and a mid-cycle correction in 2027-2028 that tests the contracted floor. This is the scenario where SanDisk proves it is a better business than old NAND without proving it has fully escaped the cycle.

The bear case requires a hyperscaler digestion cycle, competitor capacity response, and NBM variable-pricing components resetting lower in years three through five. Even here, the floor is higher than the $140 bear I wrote at initiation, the contracted base provides genuine protection that did not exist in prior cycles.

What We Track

Eight items, with falsifiable thresholds:

1. NBM coverage, above 50% of forward-year bits strengthens bull case; stuck at 33-40% confirms base

2. Gross-margin floor, sustained above 50% = bull; normalizing to 45% = base; below 35% = bear

3. Datacenter mix, above 35% and rising = infrastructure supplier; stalling below 28% = mix-shift fading

4. ASP vs bit decomposition, pricing-led is the current story; bit-led growth needed for durability

5. QLC Stargate ramp, must become a real second datacenter growth vector, not just a press release

6. Competitor supply, Samsung greenfield NAND announcement = old-cycle risk returns

7. Hyperscaler language, “scaling” and “deploying” remain bullish; “digesting” and “right-sizing” are warnings

8. Contract liabilities on balance sheet, should rise if NBM prepayments are real; flattening = stalling

From Thesis to Test

The first article asked whether AI needed a warm-storage layer between HBM and hard drives. The second article showed that NVIDIA’s architecture confirmed the problem. This article has to ask a harder question: can SanDisk turn that problem into durable economics?

SanDisk has not become software. It has not escaped memory cyclicality. But it has found a mechanism, contracts, guarantees, committed bits, a JV structure that makes those commitments credible, that could make part of the cycle contract-backed. Whether that mechanism scales from one-third to one-half of bits, whether the margin floor holds through the next downturn, and whether the demand from agentic workloads and inference scaling sustains the capex cycle through 2028, those are the questions the next 18 months will answer.

At initiation, SanDisk was a thesis. After Q2, it was a confirmation. After Q3, it is a test. The easy intellectual work, identifying the bottleneck, explaining why warm storage matters, understanding why the fifth wheel has strategic value, is done. What remains is watching whether the toll booth collects, and at what price.

The GPU highway is twenty lanes wide and full to capacity. SanDisk paved the on-ramps faster than I expected and built toll booths I did not anticipate. Whether the tolls hold through the next cycle is the only question that matters now.

Subscribe now

• Microsoft’s Q3 FY26 was strong, but the real debate is not capex or Azure growth; it is whether AI is a CIO-led cycle or a user-led cycle.

• If Agent 365, Work IQ, Entra, and Purview become the default governance layer for enterprise agents, Microsoft can capture durable AI value even without owning every frontier model.

• The risk is that Claude, ChatGPT, Cursor, or another AI-native agent owns the user intent layer, leaving Microsoft as essential middleware rather than the operating system of AI-era work.

When the Perimeter Moves

In 1998, corporate security meant a locked door. Data lived on servers inside buildings, accessed through desktops connected by Ethernet cables. The IT department controlled security by controlling physical access to the local area network. Then laptops arrived with WiFi cards, and data started leaving the building every night in thousands of laptop bags. The physical perimeter dissolved.

Microsoft’s answer was not to fight the laptop. It was to move the perimeter. Active Directory shifted the control point from the building to the user’s identity. It no longer mattered where you were; what mattered was who you were. If you wanted to access the file server, you authenticated through Microsoft. The principle was elegant: when technology dissolves an old perimeter, don’t rebuild it. Build a new one at a different layer.

Cloud computing dissolved the next perimeter, the server room. Microsoft moved the control point again, to Entra and the enterprise cloud identity plane. And now agents are dissolving the application perimeter itself. When an autonomous agent reads your Salesforce database, updates your CRM, and commits code to GitHub without a human ever opening a UI, the application interface is no longer the control point. Work happens without anyone touching the software. The perimeter must move again.

The question, and it is the most important strategic question facing Microsoft, is whether the perimeter moves to Microsoft’s governance layer, or to the agent itself.

What We Got Right, and What We Missed

I have spent the past three quarters building a thesis about Microsoft and AI control. In October, I argued that AI’s “jagged intelligence” would force enterprises back toward integrated platforms. I introduced a three-layer framework, the Grid (infrastructure), the Engine (models), and the Cockpit (identity, security, compliance), and concluded that Microsoft would win by dominating the Cockpit. I titled the section “Control is Everything.”

In January, I extended the thesis. Agent 365 was not just integrated governance; it was portable governance. Microsoft’s bet was not that AI always runs on Azure, but that every enterprise agent must authenticate through Microsoft’s control plane. I compared it to Active Directory and concluded that the perimeter always moves but never disappears.

Those frameworks were not wrong. But they were incomplete. My prior thesis was that AI would move control from applications to governance. That still looks right. What I underweighted is that AI may also move user intent away from applications entirely, and if the agent owns intent, governance may be valuable without being sovereign.

What changed my mind was not an earnings print. It was watching Microsoft bring a competitor’s agentic technology into its own flagship product, and call it progress.

The Quarter: Strong Metrics, Wrong Debate

Microsoft’s Q3 FY26 results were excellent by any conventional measure. Revenue was $82.9 billion, up 18%, beating consensus by $1.5 billion. Azure grew 39% in constant currency, above the 37-38% guide, with Q4 guided to 39-40%, the first signal of re-acceleration. M365 Copilot paid seats crossed 20 million, up from 15 million the prior quarter. For the first time, Microsoft disclosed an AI revenue run rate: $37 billion, growing 123% year over year. Satya Nadella said Copilot’s weekly engagement has reached the same level as Outlook.

The stock faded roughly 2% because Microsoft guided calendar-year 2026 capex to approximately $190 billion, $35 billion above consensus. Wall Street’s debate became: great quarter versus scary spending. That is the stock analyst’s debate. The strategic debate is about something the earnings call surfaced but the market has not fully processed: whether the AI cycle rewards enterprise distribution or product velocity, and which one Microsoft actually has.

The Copilot Cowork Moment

In January 2026, Anthropic launched Claude Cowork, an agentic tool for non-developers that handles spreadsheets, file management, reports, and automated workflows. The launch coincided with a sharp selloff in enterprise software stocks as investors recognized that Anthropic’s products looked a lot like traditional software and could replace much of the sector.

Seven weeks later, on March 9, Microsoft launched Copilot Cowork. Microsoft’s own blog described it as integrating “the technology platform that powers Claude Cowork” into Microsoft 365 Copilot. The enterprise version runs inside a customer’s Microsoft 365 tenant, grounded in Work IQ, secured by Entra, governed by Purview. According to Fortune’s reporting, Microsoft’s CMO for AI at Work, Jared Spataro, called Anthropic’s offering “a fantastic tool” and said: “What Anthropic has done is demonstrate the value of these agentic capabilities and show us practically what it could look like. Microsoft is all about commercialization.”

This is not a disproof of Microsoft’s platform thesis. It is the cleanest possible stress test of it.

If Microsoft can turn every external agent innovation into a Microsoft-governed enterprise product, then importing Anthropic is not weakness, it is the platform working as designed. Microsoft adds identity, compliance, Work IQ context, and enterprise data protection. The agentic capability becomes safe for the Fortune 500. That is a powerful value proposition.

But the implication of Spataro’s framing, Anthropic demonstrates, Microsoft commercializes, captures a division of labor that should make long-term investors uncomfortable. Microsoft, with exclusive OpenAI access since 2019, $190 billion in annual AI capex, and 450 million M365 users, did not build the agentic work product first. It integrated someone else’s. The question is whether that pattern reflects platform strength or product-layer dependence.

Is AI a CIO Cycle or a User Cycle?

Every major computing cycle is ultimately a fight over the default interface, the surface where users express intent.

CIO cycles reward integration, distribution, and enterprise trust. User cycles reward product velocity and new interaction models. Microsoft wins the first kind. It historically loses the second.

The evidence from the past eight months suggests AI is behaving like a user cycle at the product layer. Anthropic reported that its run-rate revenue has surpassed $30 billion, up from approximately $9 billion at the end of 2025, the fastest growth trajectory in enterprise software history, with roughly 80% coming from enterprise and developer API usage. Claude Code, which Anthropic said reached $1 billion in run-rate revenue just six months after public launch, had more than doubled to over $2.5 billion by February 2026. Developer surveys suggest Claude Code has overtaken GitHub Copilot in professional usage, with substantially higher satisfaction scores. One analysis estimated that 4% of all public GitHub commits are now authored by Claude Code. Cursor, another AI-native coding tool, reportedly reached $2 billion in annualized revenue.

Developers did not wait for CIO approval. They adopted the tool that solved the problem. Enterprise followed. This is what a user cycle looks like: bottom-up adoption, product quality as the selection criterion, enterprise rationalization after the fact.

If Context Is the OS, Microsoft Wins

Microsoft’s strongest argument is not about Copilot-the-chatbot or Azure-the-infrastructure. It is about enterprise context as a structural moat.

AI agents need organizational context to be useful in enterprise settings. They need permissions, identity, organizational memory, audit trails, and policy enforcement. Microsoft owns that context layer. Work IQ spans more than 17 exabytes of organizational data, growing 35% year over year. Agent 365 governs tens of millions of agents across tens of thousands of companies. Ninety percent of the Fortune 500 have active agents built with Microsoft’s tools. The $627 billion in commercial remaining performance obligations represents enterprises pre-paying for governed AI environments.

If Agent 365 achieves Active Directory-level ubiquity, then it does not matter whether developers prefer Claude Code or whether ChatGPT has a better consumer experience. Every agent that operates inside an enterprise must still authenticate through Microsoft, query Microsoft’s context layer, and comply with Microsoft’s governance. The Copilot Cowork launch demonstrates exactly this: Anthropic provides the agentic capability, but the enterprise version runs inside Microsoft 365’s tenant, wrapped in Work IQ and enterprise data protection.

The strongest bull reading is that Microsoft can absorb the best agents. The model doesn’t matter. The agent doesn’t matter. The context and governance layer, the new Cockpit, is what matters.

But there is a historical echo worth noting. Active Directory had Windows beneath it. Microsoft controlled the desktop operating system, which meant it controlled the surface that applications ran on. Agent 365 may not have the equivalent advantage. Agents can originate in Claude, ChatGPT, Gemini, Cursor, local terminals, browser layers, or future environments Microsoft does not control. Can governance achieve ubiquity without owning the surface beneath it? That is the open question.

The Product Layer Problem, A Thesis Revision

This is where I need to revise my own framework. In January, I wrote that “the Cockpit can win even if other layers face pressure.” I still think that is directionally right. But the evidence since January has accumulated in a pattern I can no longer dismiss as noise.

In every product category where Microsoft had an AI first-mover advantage, search, coding, enterprise productivity, consumer AI assistant, it has either lost the product lead or failed to convert the head start into dominance. Google Search revenue has accelerated from 12% to 19% year-over-year growth since Bing got GPT-4, based on Fiscal.ai data from SEC filings, Google Cloud is growing 63% versus Azure’s 39%, with Google Cloud margins expanding while Microsoft’s compress. GitHub Copilot, with a four-year head start, now trails Claude Code in developer satisfaction surveys. And when the defining agentic work product appeared, Microsoft’s response was to integrate it rather than build it.

Fifteen months ago, I titled an article “Infrastructure as Destiny.” I am no longer sure that is right. Infrastructure is necessary. Whether it is destiny depends on whether the user surface forms inside Microsoft’s products or outside them. Microsoft is not at risk of missing AI revenue. It is at risk of missing AI sovereignty, the ability to set the terms of how AI-era work gets done rather than merely facilitating it.

We were right that AI dissolves the application perimeter and makes governance more valuable. What we may have missed is that AI also moves intent away from applications, and the owner of intent may sit above the owner of governance.

Three Scenarios for Microsoft’s AI Future

The strategic question maps to three distinct outcomes, each with different implications for what Microsoft is worth.

The observation that matters: even the bear case has meaningful upside from $430. At 24x forward earnings, the stock is priced closer to the bear scenario than the base. The market has already discounted the capex risk and competitive pressure. What it has not priced is the possibility, still the most likely outcome, that Microsoft’s governance layer captures durable value even if it does not define the era. The asymmetry favors the long side. But the type of return differs across scenarios: the bull is a platform compounder that re-rates to 30x; the base is a steady mid-teens grower that normalizes to 26x; the bear is a utility generating excellent cash flow at a compressed multiple.

The Perimeter Always Moves

Six things will determine which scenario unfolds, not Azure growth in isolation, not quarterly EPS, but the questions that separate platform ownership from profitable participation.

The perimeter always moves. It never disappears. For the past year, I have been confident that the next perimeter is Microsoft’s governance layer, that control would migrate from applications to identity, compliance, and context, and that Microsoft would own it the way it owned Active Directory.

I still believe governance is a massive and durable business. Microsoft’s enterprise distribution, identity infrastructure, and organizational context layer are assets no competitor can replicate quickly. The most likely outcome is that Microsoft captures enormous value from the AI cycle through this control layer.

But I now take seriously the possibility that governance is not the operating system of the AI era, it is the middleware. The operating system may be the agent itself: the AI that users form a relationship with, that remembers their work, that executes on their behalf, that improves over time. If that agent is Claude, or ChatGPT, or something that has not been built yet, then Microsoft’s governance layer becomes essential infrastructure wrapped around someone else’s platform, the way IBM wrapped enterprise relationships around an operating system it did not own.

The distinction between governance as the operating system and governance as middleware is the difference between Microsoft setting the terms of the AI era and Microsoft merely participating in it. Copilot Cowork, integrating Anthropic’s agentic technology inside Microsoft 365’s governance boundary, is what that distinction looks like in practice. It may be exactly how Microsoft wins. It may also be evidence that the agent layer is being invented somewhere else.

The next four quarters will tell us which one it is.

$MSFT

General Disclaimer: The information presented in this communication reflects the views of the author and does not necessarily represent the views of any other individual or organization. It is provided for informational purposes only and should not be construed as investment advice, a recommendation, an offer to sell, or a solicitation to buy any securities or financial products.

While the information is believed to be obtained from reliable sources, its accuracy, completeness, or timeliness cannot be guaranteed. No representation or warranty, express or implied, is made regarding the fairness or reliability of the information presented. Any opinions or estimates are subject to change without notice.

Past performance is not a reliable indicator of future performance. All investments carry risk, including the potential loss of principal. This communication does not consider the specific investment objectives, financial situation, or particular needs of any individual.

The author and any associated parties disclaim any liability for any direct or consequential loss arising from the use of this material and undertake no obligation to update or revise it.

Subscribe now

• Our prior view was that Amazon had become an Engine Room funding a Foundry. Retail, Prime, sellers, and advertising generated the cash; AWS, AI infrastructure, and custom silicon consumed it to build long-duration advantage. Q4 sharpened the point: the Foundry was the organizing principle of Amazon’s capital allocation.

• Q1 was the first real answer to the $200B capex question. AWS grew 28%, its fastest in 15 quarters, on a $150B annualized base. Amazon’s chips business crossed a $20B revenue run-rate. Trainium commitments reached $225B. AWS backlog surged to $364B, 2.7 times trailing revenue, up from 1.7 times last quarter, and that figure excludes the recently announced Anthropic commitment exceeding $100B.

• The thesis is stronger on demand, more conditional on returns. Operating cash flow rose to $148.5B on a trailing basis. Free cash flow fell to $1.2B as capex consumed nearly all of it, and Amazon raised $53B in new long-term debt. The Foundry is not a fantasy. It still has to become a cash-flow machine.

The Framework Survived. The Funding Model Changed.

Our framework across Q3 and Q4 was that Amazon is an Engine Room (commerce, Prime, sellers, advertising, generating traffic and cash) and a Foundry (AWS, AI capacity, Trainium, Bedrock, consuming cash to build infrastructure advantage). Q3 showed the Engine Room had the power to fund the Foundry. Q4 raised the stakes: $200B in capex, free cash flow likely turning negative, the Foundry becoming the company’s capital allocation priority.

Q1 updates, but does not replace, that framework. The Engine Room and Foundry are still distinct. But the relationship is no longer one-way. The Foundry is beginning to improve the Engine Room, Trainium powers Bedrock inference, Bedrock powers Rufus, Rufus drives engagement and conversion. That is not yet the same as saying Amazon has one fully fused AI grid. It is too early for that. But it means the Q4 question, can Amazon afford the Foundry?, is now incomplete.

The better question is: can the Foundry generate enough demand, cost advantage, and product leverage to justify the cash it is consuming?

Q1 starts to answer yes.

The Foundry Is No Longer a Promise

AWS grew 28% year-over-year to $37.6B, the fastest growth rate in 15 quarters. AWS operating margin was 37.7%, expanding while growth accelerated, a combination that is exceptionally rare at this scale. This matters because the bearish version of the $200B capex story was that Amazon might be building capacity before demand existed. Q1 argues strongly against that.

The backlog is the strongest single data point. AWS committed backlog surged to $364B, up 49% sequentially from $244B. At the current $137B trailing twelve-month run rate, that is 2.7 years of revenue visibility. Nine months ago, it was 1.7 years. The demand pipeline is not merely holding; it is widening faster than revenue can recognize it.

The more important disclosure was custom silicon. Amazon’s chips business, Graviton, Trainium, Nitro, crossed a $20B revenue run-rate, growing triple digits. Jassy went further: measured like an external chip vendor, the run-rate would be closer to $50B. Trainium commitments exceed $225B. Trainium2 is largely sold out. Trainium3 is nearly fully subscribed. Much of Trainium4, still 18 months from broad availability, has already been reserved.

That is the real narrative shift. Amazon is not renting GPUs and calling it an AI strategy. It is trying to own the inference cost curve. Jassy quantified this for the first time: Trainium saves Amazon “tens of billions of dollars of CapEx each year” and provides “several hundred basis points of operating margin advantage” versus third-party chips. If that advantage is real and durable, AWS becomes more than a cloud platform participating in the AI boom. It becomes one of the few companies that can shape the economics of the boom.

The asterisk is that Trainium still has to broaden. OpenAI, Anthropic, Meta, and Uber are powerful names. But the platform question is not whether enormous customers will reserve capacity. It is whether the next layer of enterprises chooses Trainium because it is simply the best economic answer, not because Amazon invested in them.

The Engine Room Is Still Throwing Off Heat

It would be a mistake to turn Q1 into only an AWS note.

Unit growth reached 15%, the highest since COVID lockdowns. North America operating income rose 42% to $8.3B at a 7.9% margin. The cost spread confirmed structural operating leverage: units grew 15%, fulfillment costs grew 9%, shipping costs grew 12%. The logistics network is getting denser and more efficient at the same time volume is accelerating.

Advertising deserves its own recognition. At $17.2B in Q1 and over $70B on a trailing basis, advertising is the single highest-return-on-capital business inside Amazon. It requires no fulfillment centers, no delivery vans, no inventory. It operates at estimated 70%+ margins, layered on traffic the retail business already generates. Every dollar of advertising revenue drops almost entirely to operating income. This is the business that, more than AWS, more than Prime, actually funds the operating cash flow surplus the Foundry consumes.

The tariff bear case, which dominated pre-earnings sentiment, was a non-event in Q1. Average selling prices declined year-over-year. Amazon absorbed costs through efficiency rather than raising prices and still expanded margins. Jassy did not mention tariffs in prepared remarks. The silence is itself a data point.

Together, the Engine Room produced $148.5B in trailing twelve-month operating cash flow, up 30%. The funding mechanism is not a theory. It is accelerating at the exact moment the Foundry needs it most.

The Balance Sheet Enters the Story

The income statement was excellent. Revenue grew 17% to $181.5B. Operating income rose to $23.9B, above the guide high-end by $2.4B. Operating margin hit a record 13.1%. Reported EPS of $2.78 was inflated by a $16.8B pre-tax gain from Anthropic investments; the operating performance, not the bottom line, is the cleaner signal.

The cash-flow statement was more uncomfortable. Operating cash flow rose 30% to $148.5B on a trailing basis. Free cash flow fell to $1.2B. Capex consumed 99.2% of all cash the business generated.

Then the balance sheet. Long-term debt rose from $65.6B at year-end to $119.1B, with $53.4B of new long-term debt proceeds in Q1. This received zero discussion on the earnings call.

That does not break the thesis. Interest coverage is 29.8 times. The cash balance grew to $101.8B. But it refines the thesis.

Our earlier version was: the Engine Room funds the Foundry. The more accurate version after Q1: the Engine Room funds most of the Foundry; the balance sheet is funding the rest. At $200B of annual capex on ~$150B of operating cash flow, Amazon needs approximately $50B per year in external financing. This is no longer a self-funded reinvestment cycle. It is a reinvestment cycle with leverage.

The Market Is Still Using the Wrong Capex Frame

Here is what the market sees: a company spending $200B per year with near-zero free cash flow and $119B in debt. A capex risk. A “show me” stock.

Here is what the market is underweighting: the $200B in capex is not serving one business. It is serving four, AWS (28% growth), Stores (15% unit growth), Advertising ($70B+, 24% growth), and Prime (15% subscription growth), all of which use the same Trainium-Bedrock-AI infrastructure layer. Each business’s growth reduces the others’ unit costs: more retail units create more ad impressions, more ad revenue funds more capex, more capex builds more Trainium capacity, more Trainium capacity lowers inference cost, lower inference cost makes Rufus and Alexa+ and Health AI cheaper to run, and cheaper AI applications drive more retail engagement.

The market applies a single-business capex framework to a four-business grid. That is the analytical error. The correct framework amortizes the infrastructure across all four loops, which changes the return-on-invested-capital math. It does not make the capex cheap. It makes it less reckless than it appears on a single-segment basis.

What would make this wrong: a third-party shopping agent captures the consumer entry point, severing the Stores-to-Advertising loop. AWS growth decelerates below 22%, suggesting Q1 was a cyclical peak. Tariffs compress retail margins enough to weaken the Engine Room’s cash generation. Debt exceeds $150B without a clear FCF inflection timeline. Any of those would impair the grid’s economics. All of them simultaneously would break the thesis.

The Price of the Grid

The range is wide because the outcome genuinely depends on execution over the next two to three years, not on information available today.

The bear case requires AWS to decelerate to low-20s, tariffs to compress margins, advertising to slow to mid-teens, and the FCF trough to extend into 2028. The base case requires AWS at 25%+ through 2027, advertising above 20%, margins expanding as the depreciation window closes, and FCF inflecting to $60B+ by 2028. The bull case requires Trainium to become a platform, AWS to sustain high-20s growth longer than expected, and the market to re-rate as four compounding loops become visible in the financial statements.

The skew matters: the bear case requires multiple loops to break simultaneously; the bull case requires the existing trajectory to continue.

The New Proof Points

The Q4 signposts have largely been met. The next phase requires stricter markers.

1. AWS growth ≥27% through Q3 2026. Below 24%, the overbuild question returns.

2. Advertising ≥22% ex-FX. This is the Engine Room’s highest-margin turbine. Below 18% signals the agentic entry point is shifting.

3. Unit growth ≥12% with the cost spread remaining positive, units must grow faster than fulfillment and shipping costs.

4. Three named Trainium enterprise deployments outside Amazon’s investment portfolio by H2 2026.

5. TTM FCF inflects to $10B+ by Q4 2026. Near zero into 2027 means the depreciation window becomes a funding problem.

6. Long-term debt does not exceed $150B by year-end 2026. Above that, the conversation shifts from investment cycle to structural leverage.

Evidence Is Not Yet Cash

Q4 was when Amazon asked investors for patience. Q1 was when it gave them evidence.

The Foundry has numbers behind it: AWS reacceleration, Trainium commitments, Bedrock momentum, custom silicon at scale, and demand arriving before capacity. The Engine Room still has power: units, ads, fulfillment leverage, Prime, and segment profits all improved.

But the proof that matters most is still ahead. Revenue is not enough. Operating income is not enough. Even $364B in backlog is not enough.

The proof is cash.

Amazon is no longer asking investors to believe in the Foundry on faith. Q1 gave them evidence. Now the company has to turn that evidence into free cash flow.

Subscribe now

• Alphabet is no longer just a Search company defending itself from AI; it is four AI-powered businesses sharing one infrastructure layer. Search compounds intent, Cloud compounds compute demand, YouTube compounds attention, and Waymo compounds autonomous miles.

• Q1 2026 showed that the grid is starting to pay off. Search grew 19%, Cloud grew 63% with 32.9% margins and $462B of backlog, YouTube is layering subscriptions on top of living-room scale, and Waymo crossed 500K autonomous rides per week.

• The debate has shifted from disruption risk to capital allocation. The question is no longer whether AI kills Google; it is whether Alphabet’s $180–190B+ capex cycle can generate adequate returns across four separate compounding loops.

Sundar Pichai summarized the quarter neatly: Alphabet’s AI investments and full-stack approach are now “lighting up every part of the business.” The numbers back him up. Revenue grew 22% to $109.9B, Search & Other grew 19%, Google Cloud grew 63%, operating income grew 30%, and operating margin expanded to 36.1%.

Those are excellent numbers. They are not the most interesting thing about the quarter.

For the past year, the right way to understand Alphabet was that it was not trying to win AI by building the best chatbot. It was trying to build the grid: the infrastructure, silicon, models, products, and distribution layer that could make intelligence cheap, ubiquitous, and profitable.

Q1 2026 sharpens that thesis. The grid is not the end state. The grid is the shared production function for four businesses: Search, Google Cloud, YouTube, and Waymo. Each uses the same AI infrastructure. Each compounds differently. Search compounds intent. Cloud compounds compute demand. YouTube compounds attention. Waymo compounds autonomous miles.

That is why Alphabet is harder to analyze than a normal mega-cap. It is not one AI story. It is four AI-powered business-model upgrades funded by one balance sheet.

The question is no longer “Will AI kill Google?” That question looks increasingly stale. The question is:

Can Alphabet turn one massive AI infrastructure investment cycle into four separate compounding loops?

Q1 suggests yes. It also makes the cost impossible to ignore.

The Four Loops

The mistake is to think of Alphabet’s businesses as reporting segments. They are really four markets in which Alphabet is trying to change the unit of competition.

The same stack, TPUs, GPUs, data centers, Gemini models, agents, and developer tools, increasingly powers all four. That is the point.

The Agentic Pivot

For 25 years, Google’s Search business was simple. A user had intent. Google matched that intent to information and advertising. The user clicked. The transaction happened somewhere else. Google took a toll for the referral.

That was an extraordinary business. It was also why generative AI looked so dangerous. If AI answered the question directly, what happened to the click? If the click disappeared, what happened to the ad? If the ad weakened, what funded the rest of Alphabet?

Q1 narrows that bear case. Search & Other grew 19% to $60.4B, and management said AI Mode and AI Overviews are bringing users back to Search more, with queries at all-time highs. Pichai also said Google reduced the cost of core AI responses by more than 30% after upgrading AI Overviews and AI Mode to Gemini 3.

That matters. AI is not merely adding cost to Search; it is expanding usage while Google pushes down unit cost.

But the more important change is the mechanism of monetization. Philipp Schindler said Gemini is being deployed across the ads infrastructure, improving intent understanding, ad relevance, and bid strategies. More than 30% of customer Search spend now uses AI-enabled campaigns such as AI Max or Performance Max. Google is also testing direct offers in AI Mode and new ad formats around organic product recommendations.

Then came the bigger reveal: agentic commerce.

The Universal Commerce Protocol added Amazon, Meta, Microsoft, Salesforce, and Stripe to its technology council, joining Shopify, Etsy, Target, Wayfair, and Google. Partners such as Sephora, Macy’s, and Ulta Beauty are rolling out UCP, and Ulta launched agentic commerce inside AI Mode, Search, and Gemini. Users can review recommendations, compare products, and complete checkout inside the experience.

That is a different business.

To be clear, this is not yet proven. Alphabet has not disclosed GMV, take rate, conversion rate, or transaction revenue from agentic commerce. Google has tried commerce before and failed more than once. The difference this time is that the agent changes the user behavior: the user does not need to leave, the merchant does not need to rebuild the storefront, and the payment rails can sit underneath the assistant.

The old Google monetized intent by sending users elsewhere. The new Google wants to monetize intent by completing the task.

The bear case is no longer simply “AI reduces queries.” Q1 argues against that. The new bear case is that someone else becomes the starting point for agentic commerce.

If the user begins with Google, Search expands from referral layer to transaction interface. If the user begins with ChatGPT, Perplexity, Apple, Amazon, or a vertical agent, Google’s historic monopoly over commercial intent becomes less relevant.

That is the real fight.

The Grid Gets Paid

Cloud is where the grid becomes an external revenue stream.

Google Cloud revenue grew 63% to $20.0B. Operating income tripled to $6.6B. Operating margin expanded from 17.8% to 32.9%. Backlog nearly doubled sequentially to $462B.

This is no longer optionality.

Management’s commentary was even stronger than the numbers. Pichai said enterprise AI solutions became Cloud’s primary growth driver for the first time, revenue from products built on Google’s GenAI models grew nearly 800% year-over-year, new customer acquisition doubled, and multiple $1B+ deals were signed.

This is the grid thesis made tangible. Alphabet owns models, silicon, data centers, cloud infrastructure, enterprise applications, agent platforms, and now a security layer through Wiz. That vertical integration is no longer just a technical claim. It is appearing in revenue, margins, and backlog.

The new wrinkle is TPU hardware.

Pichai said Google will begin delivering TPUs to a select group of customers in their own data centers. CFO Anat Ashkenazi later clarified that TPU hardware agreements are included in Cloud backlog, that the majority of backlog is still typical GCP contracts, that just over 50% of backlog should convert into revenue over the next 24 months, and that TPU hardware revenue will be lumpy, with a small percentage recognized later in 2026 and the majority in 2027.

This is not Google “becoming NVIDIA.” It is subtler. Google is selectively opening the TPU stack to prevent the highest-value AI workloads from defaulting to NVIDIA-only infrastructure.

That may be the right trade. It also creates a new disclosure problem.

A dollar of recurring GCP consumption is not the same as a dollar of lumpy TPU hardware revenue. If the backlog is mostly durable GCP and AI services, it deserves a high-quality multiple. If a material portion becomes lower-margin hardware, investors will discount it. Management said the majority is still typical GCP contracts, which is reassuring; the fact that they needed to say it tells you what to watch.

YouTube Is Not an Ad Line Item

YouTube is the business most often underweighted in Alphabet analysis.

YouTube ads grew 11% to $9.9B. That is solid, but it looks underwhelming next to Search growing 19% and Cloud growing 63%.

The problem is that YouTube ads are no longer the whole story.

Pichai said U.S. viewers watch more than 200M hours of YouTube in the living room daily. More than 10M channels now publish Shorts each day. YouTube Music and Premium had their largest quarterly increase in non-trial subscribers since YouTube Premium launched. Schindler added that YouTube subscriptions revenue continues to grow faster than ads, especially YouTube Music and Premium.

This is not just a video ad platform. It is a media operating system.

Traditional media companies decide what to produce, spend the money upfront, and hope the audience shows up. YouTube inverts the model. Creators produce the content. Users reveal demand. YouTube shares revenue when the content works.

AI makes that more valuable. Gemini can improve creator matching, ad matching, content discovery, translation, summarization, clipping, and eventually video search. YouTube may become the world’s largest AI-queryable video knowledge base.

That is why the disclosure gap matters. If YouTube ads are growing 11%, but YouTube subscriptions are growing faster, then the ad line understates the business. Alphabet should disclose total YouTube revenue, including subscriptions. Until it does, the market will likely continue to undervalue YouTube as a standalone platform.

Search is the intent graph. YouTube is the attention graph. Both become more valuable when AI can understand and act on them.

Waymo Becomes Measurable

Waymo used to live in the “someday” bucket.

Someday autonomy works. Someday rides scale. Someday the unit economics matter. Someday the valuation shows up.

Q1 moved Waymo closer to today.

Waymo surpassed 500,000 fully autonomous rides per week, doubling in less than a year. It launched in Nashville, added six new cities so far in 2026, and now operates in 11 major U.S. cities.

This is still not a financial story. Other Bets revenue was only $411M, and Other Bets lost $2.1B.

But Waymo is no longer merely a science project. It is an operating network.

The other important disclosure is what Alphabet is doing around Waymo. Verily completed an external capital raise and was deconsolidated. GFiber announced a combination with Astound Broadband and is expected to be deconsolidated when the deal closes. Management said Alphabet will continue allocating significant resources to businesses where it sees meaningful opportunities, such as Waymo.

This matters because Other Bets has historically obscured as much as it revealed. If Verily and GFiber move out of the consolidated segment, Waymo becomes more visible. That does not mean an IPO or tracking stock is imminent. It does mean the market will increasingly have to ask what Waymo is worth.

The answer is not zero anymore.

The Cost of the Grid

Alphabet’s income statement looked excellent. Revenue grew 22%, operating income grew 30%, and operating margin expanded to 36.1%. But GAAP EPS of $5.11 was flattered by $37.7B of other income, primarily unrealized gains on non-marketable equity securities. The better operating read is revenue and operating income, not EPS.

The cash flow statement is where the cost shows up. Operating cash flow was $45.8B. Capex was $35.7B. Free cash flow was only $10.1B. TTM free cash flow was $64.4B. Alphabet ended the quarter with $126.8B of cash and marketable securities, but long-term debt rose to $77.5B.

Then management raised the stakes. FY26 capex guidance increased to $180–190B, up from $175–185B, and management said 2027 capex will “significantly increase” compared to 2026.

This is not the old Alphabet.

The old Alphabet was a capital-light advertising monopoly. The new Alphabet is still immensely profitable, but it is becoming a capital-intensive AI infrastructure platform.

The bear case is straightforward: AI is turning Alphabet’s high-margin, high-FCF advertising business into a lower-FCF infrastructure utility.

The bull case is more subtle: Alphabet’s capex should not be judged against a single business. The same infrastructure improves Search, monetizes Cloud, enhances YouTube, and trains Waymo.

That is the capital allocation claim embedded in “full stack.” Alphabet is saying it can spend more because it has more surfaces over which to amortize the spend.

The Conglomerate View

Alphabet trades and reports like one company. Strategically, it increasingly looks like four.

A blended multiple hides businesses at different stages, with different economics and different valuation frameworks:

This is intentionally rough, but useful. A crude SOTP points to $3.8–4.0T of equity value before giving credit for agentic commerce optionality or faster Waymo scaling. Add net cash, and the argument is not that Alphabet is obviously cheap on consolidated earnings; it is that the consolidated multiple obscures the range of assets inside the company.

Search should be valued like a mature but still-growing intent monopoly with agentic commerce upside. Cloud deserves a growth infrastructure framework if it can sustain high growth and high margins. YouTube deserves more than an ad-line multiple if subscriptions and living-room viewing keep scaling. Waymo deserves option value that is no longer theoretical.

The old description, “advertising company with Cloud optionality”, is obsolete.

The Investor Test

The old investor test was simple: does AI kill Search?

That is no longer sufficient. Search is growing faster with AI. The harder question is whether Alphabet can generate adequate returns across all four loops while capex rises.

The disclosure gaps are equally clear: AI Search monetization, Cloud backlog quality, total YouTube revenue including subscriptions, and Waymo operating metrics.

The more Alphabet becomes four businesses on one grid, the more investors need business-specific metrics.

The New Question

A year ago, the question was whether AI would kill Google.

By late 2025, the question became whether Google had built the intelligence grid. The answer was increasingly yes: AI did not kill Search, Cloud was breaking out, custom silicon mattered, capex was offensive, and Alphabet’s five-layer moat was beginning to compound.

Search compounds intent. Cloud compounds compute demand. YouTube compounds attention. Waymo compounds autonomous miles. The same infrastructure increasingly powers all four.

That is the opportunity. It is also the risk.

Alphabet’s Q1 2026 was not just a strong quarter. It was the quarter that clarified the next version of the company.

The question is no longer whether Google built the grid. The question is how many markets that grid can power.

Subscribe now

• The AI dividend is real and accelerating. Revenue grew 33%, driven by 19% more ad impressions at 12% higher prices, the rare combination that only happens when AI is making the product genuinely better. The Family of Apps generated $26.9B in operating income at a 48% segment margin.

• The AI tax is rising and locked in. Capex guidance rose to $125–145B, but the spending isn’t just for better ads, it’s building a consumer agent platform that doesn’t have revenue yet. Revenue grew 33%; capex grew 45%. Those two slopes are the entire debate.

• The stock fell because the market can’t value what it can’t see. Meta has a history of making massive bets on attention shifts before the business model is clear, mobile, Instagram, Stories, Reels. The AI agent bet follows the same pattern at 100x the cost. The market is not rejecting the strategy. It is struggling to price a future it cannot yet measure.

What Is Meta?

Before the quarter: Meta becomes more valuable as its market grows because every unit of human attention flowing through its apps generates behavioral data that improves AI models, which makes the next unit of attention more valuable, to the user, to the advertiser, and to Meta. More engagement creates more data; better data creates better models; better models create more engagement and higher-performing ads; higher ad performance generates the cash to fund more compute; more compute improves the models.

The output of the system is also the input. That is what makes it compound.

This mechanism strengthens at scale because Meta has 3.56 billion daily active people, massive advertiser liquidity, real-time conversion feedback, and enough cash flow to fund the infrastructure required to keep the loop spinning. It breaks if the cost of compute rises faster than the monetization it produces, or if something intercepts user intent before it reaches Meta’s apps.

That is the fundamental question this quarter needed to answer: is the loop still compounding, or is the cost of running it starting to outpace the returns?

AI and the Attention Business

The answer, on the operating side, was unambiguous.

Revenue grew 33% year-over-year to $56.3 billion, the fastest growth since Q4 2021, on a revenue base nearly twice as large. More importantly, ad impressions grew 19% and price per ad grew 12%. That combination is rare. Usually more ad volume dilutes pricing. Meta delivered both because the underlying product is getting better.

Susan Li explained the mechanism:

On Instagram, the ranking improvements that we made in Q1 drove a 10% lift in Reels time spent. On Facebook, total video time increased more than 8% globally in Q1, the largest quarter-over-quarter gain in four years... We doubled the length of user interaction sequences we use for training on Instagram and increased the richness of how each user interaction is described, enabling our systems to develop a deeper understanding of user interests.

These are not plans. They are engineering changes running in production, generating measurable engagement gains right now. The ad improvements are not separate from the AI investment. They are the first return on it.

The monetization evidence goes further. The value optimization suite, which helps advertisers find the highest-value conversions, not just the most conversions, has surpassed $20 billion in annual revenue run rate, more than doubling year-over-year. The adaptive ranking model, a trillion-parameter LLM-scale inference system, drove measurable conversion rate gains across major ad surfaces. Eight million advertisers are using generative AI creative tools. Partnerships ads surpassed $10 billion in run rate.

The Family of Apps generated $26.9 billion in operating income at a 48% segment margin. That is the profit pool funding everything else. No other consumer internet company generates cash at this rate.

The AI dividend is real, measurable, and accelerating. Price per ad re-accelerated from just 6% in Q4 to 12% in Q1. The flywheel is working.

The Capex, and What It’s Actually Building

That brings us to the number the market cared about. Meta raised its 2026 capex guidance from $115–135 billion to $125–145 billion, and disclosed a $107 billion step-up in contractual commitments in Q1 alone, multi-year cloud deals and infrastructure agreements that lock in the spending trajectory for years. Li was candid:

Our experience so far has been that we have continued to underestimate our Compute needs, even as we have been ramping capacity significantly.

If this spending were only about making ads better, the market would have a point. The ad stack improvements are valuable but incremental, 6% conversion lifts, 1.6% ranking gains, better creative tools. Important, but probably not $135-billion-a-year important.

The spending makes sense only if you understand what Meta is actually building: not just a better ad machine, but a consumer agent platform.

Zuckerberg was explicit about this on the call:

My view of AI is very different from many others in the industry. I hear a lot of people out there talk about how AI is going to replace people. Instead, I think that AI is going to amplify people’s ability to do what you want... We are building a personal agent focused on helping people achieve the diverse goals in their lives. We’re also building a business agent focused on helping entrepreneurs and businesses across the world.

The distinction between an assistant and an agent is the distinction between a feature and a platform. An assistant answers questions inside someone else’s product. An agent understands your goals and works proactively to achieve them, it becomes the interface. If successful, it opens monetization beyond advertising: commerce commissions, premium subscriptions, business services.

The earliest evidence is already visible. Business AI conversations on WhatsApp and Messenger grew from 1 million weekly at the start of the year to 10 million, a 10x increase in one quarter. Li noted these are “currently free for most businesses,” with plans to establish a longer-term monetization model.

But Zuckerberg was also honest about readiness. Asked about timing, he said the quality bar is building something he would “want to give to my mother.” That tells you the consumer agent is not ready for mass-market deployment. 2026 is a development year. Monetization is 2027+.

So the capex funds two things: the ad stack improvements generating returns today, and the agent platform that is supposed to generate returns tomorrow. The first justifies some of the spending. The second is supposed to justify the rest. And the second does not have revenue yet.

The Risk Under the Risk

Most of the market’s debate is about cost. I think the deeper risk is disintermediation.

Meta’s flywheel depends on attention and intent flowing through its apps. If an on-device AI agent, Apple Intelligence, most obviously, handles shopping, recommendations, and messaging at the operating system layer before the user opens Instagram or WhatsApp, Meta never sees the interaction. The data loop breaks at its source. Apple’s ATT changes already cost Meta an estimated $10 billion in annual revenue by restricting data. On-device agents could do worse, not by restricting Meta’s data, but by making Meta’s apps unnecessary.

This is why Reality Labs matters despite $4 billion in quarterly losses on $400 million in revenue. Meta does not control a device platform. Glasses are not strategically interesting because of hardware revenue. They are interesting because they could become a computing surface Meta owns, one where the agent sees what you see and the data loop is secured at the hardware layer rather than rented from Apple at the app layer.

Capex, agents, and glasses are not three separate bets. They are one bet.

Why the Market Is Confused

The stock fell because the market cannot value what it cannot see.

Meta told investors two things simultaneously.

1. First: AI is already making the core business better, here are the metrics.

2. Second: we are spending $135 billion a year building something new that does not have revenue yet, and we cannot tell you when the spending peaks.

The first story had numbers. The second had vision. The market priced the second.

This is a familiar pattern for Meta, though not always a comfortable one. The company has a history of making large, disorienting bets on where attention is moving, often before the business model is clear, and often at a cost that alarms investors.

The mobile transition in 2012–2013 is the closest parallel. Meta saw usage migrating to phones, invested heavily in mobile ad formats while desktop revenue slowed, and watched the stock trade below its IPO price for over a year. The market could see the cost of the transition but not the revenue. Then mobile ads worked, and the stock went from $18 to $300 over the next eight years.

Instagram looked reckless, $1 billion for thirteen employees and no revenue. Stories looked self-destructive, cannibalizing News Feed. Reels looked derivative, copying TikTok. In each case, Zuckerberg bet on where attention was heading and absorbed near-term pain.

But not every bet has worked. The metaverse pivot was the most expensive strategic miscalculation in Meta’s history, costing tens of billions before the company course-corrected toward AI. The Libra/Diem cryptocurrency initiative collapsed under regulatory pressure. Meta’s commerce push on Instagram and Facebook in 2021–2022 was quietly wound down.

The difference between the successes and failures is instructive. The successes, mobile, Instagram, Stories, Reels, were bets on observable shifts in user attention. The failures, metaverse, crypto, commerce 1.0, were bets on behavior changes that had not yet occurred.

Which category does the AI agent bet fall into? The engagement data says users are already spending more time on AI-improved content. The business AI data says businesses are already adopting AI-mediated conversations. The agent quality, by Zuckerberg’s own admission, is not yet at the “give to my mother” standard. The spending is locked in. The product is still catching up.

The Whole Quarter

Revenue grew 33%. Capex grew 45%. The ad flywheel is compounding. The cost of building the agent platform is rising. The stock reflects uncertainty about which curve is steeper.

The near-term test is price per ad. Q1’s 12% growth re-accelerated from 6% in Q4, the strongest evidence that AI compute is improving the value of Meta’s inventory. If that holds above 10% through the second half while capex stays within $145 billion, the dividend is outgrowing the tax. If price per ad falls back toward mid-single digits despite the spending, the marginal return on compute is compressing.

Meta is not becoming a worse business. It is becoming a more expensive business to run, in service of a platform that does not yet exist in finished form. The market has seen Meta make this kind of bet before. Sometimes it worked spectacularly. Sometimes it didn’t.

That uncertainty is reasonable. I think it is too pessimistic. But I understand why no one wants to say that out loud with $107 billion in contractual commitments on the table.

Subscribe now

• The quarter reframed Intel’s problem from demand to supply. Server CPU demand is strong enough that management suggested unfilled demand “starts with a B,” while DCAI margins expanded sharply to 30.5%.

• The CPU thesis is finally visible, but the foundry thesis is still mostly prospective. Google LTAs, Nvidia DGX Rubin NVL8, and custom ASIC growth validate Intel’s AI infrastructure role, while external foundry revenue remains small and losses remain large.

• Intel is now a race between two clocks. The product flywheel is accelerating, but the foundry still consumes cash; the bull case depends on CPU-led cash generation arriving before AI demand slows.

From Bloomberg’s Ian King:

Intel Corp. delivered a blockbuster sales forecast that shattered Wall Street expectations, signaling that the long-struggling chipmaker is benefiting from the giant build-out of artificial intelligence computing. Revenue will be $13.8 billion to $14.8 billion in the June quarter. Analysts estimated $13 billion on average.

I’ll admit it: I did not expect Intel to report a quarter where the most important number was the one management didn’t report, the revenue they couldn’t ship. When CFO David Zinsner was asked on the call to quantify how much demand Intel is leaving on the table, his answer was five words: “Let’s just say it starts with a B.”

That single sentence reframes the entire Intel debate. For the past three years, the question has been whether Intel’s products are competitive enough to sustain the company while it tries to become a foundry. After this quarter, the question is whether Intel can manufacture enough chips to capture the demand that already exists. That’s a fundamentally different problem, and a much better one to have.

Here are the numbers:

Sixth consecutive beat. Every line above the high end of guidance. Stock up 15–20% after hours toward $76–82.

But a single quarter, even a blowout, only matters if it reveals something about the company’s trajectory. Nine quarters of context tell the story:

Revenue has been range-bound between $12.7B and $14.3B for two years. This is not yet a revenue inflection. It is a margin and narrative inflection, and the distinction matters for everything that follows.

Why the CPU-to-GPU Ratio Is the Quarter’s Most Important Number

There is a narrative that has dominated the semiconductor conversation since late 2022: the GPU is eating the data center. Nvidia’s ascent from a $300 billion company to a $3+ trillion company was built on a simple and largely correct observation, AI training requires massive parallel compute, and GPUs deliver that better than anything else.

What this narrative missed, and what Intel’s Q1 results illuminate, is that training was only the first act. The industry is now moving into inference, agentic AI, and production deployment, and each of these stages reinserts the CPU as a critical component of the stack. Here’s how Lip-Bu Tan described it:

“The CPU now serves as the orchestration layer and critical control plane for the entire AI stack. This is not just our wishful thinking. It is what we hear from our customers, and it is evident in the demand profile for our products.”

The numbers back him up. DCAI revenue hit $5.05 billion, up 22% year-over-year, with operating margins expanding from 13.9% a year ago to 30.5%, a trajectory that looks less like cyclical recovery and more like structural re-rating:

That margin expansion, from 14% to 31% in four quarters, is remarkable. It’s driven by three forces operating simultaneously: volume growth (more units), pricing power (supply constraints letting Intel raise ASPs), and cost improvement (Intel 3 yields maturing). When all three move in the same direction at once, the operating leverage is enormous.

But the number that really matters is the CPU-to-GPU ratio. Zinsner laid out the math: training runs at 1 CPU for every 7–8 GPUs. Inference drops to 1:4. Agentic workloads approach 1:1 or potentially flip in the CPU’s favor. If you believe, as I do, that inference and agentic are going to be substantially larger markets than training, then the total addressable market for server CPUs is expanding, not shrinking.

Two data points from the quarter make the CPU-demand thesis concrete rather than theoretical. First, Xeon 6 was selected as the host CPU for Nvidia’s DGX Rubin NVL8, the most important AI server platform on earth. Nvidia could design its own CPU; it has Grace. It could use AMD. It chose Intel. That is not a legacy relationship being maintained out of inertia; it is an active architectural decision by the company with the deepest understanding of AI data center requirements. When Nvidia puts Intel inside its flagship rack, it validates the x86 orchestration thesis in a way no amount of management commentary can.

Second, Intel’s custom ASIC business, purpose-built silicon for specific customer workloads, nearly doubled year-over-year and is now running above a $1 billion annual rate. Zinsner said on the call that “people have been surprised about how big the business is already,” and Tan described it as a fast-growing opportunity over the next five years. This matters because it shows Intel can participate in the accelerator-adjacent TAM without competing head-on with Nvidia’s GPU franchise. Intel brings x86 CPU IP, advanced packaging, and captive manufacturing to the table; the customer brings the workload-specific design. It’s a business model that didn’t exist two years ago.

These customer commitments sit underneath the structural argument that HSBC’s Frank Lee is building his $95 price target around: 20% server CPU shipment growth in both 2026 and 2027, with 20% ASP increases in 2026 and 10% in 2027. His DCAI revenue estimates of $22.8B (2026) and $29.1B (2027) sit 16% and 33% above Street consensus. Those are aggressive numbers. After this print, they look more plausible than they did a week ago.

What Intel Wants You to See, and What It Buried

I find it instructive to examine what Intel’s earnings presentation highlighted and what got buried. The slide deck’s executive summary had four bullet points:

1. Tangible Progress Building the New Intel

2. 6th Consecutive Quarter Exceeding Expectations

3. Intel CPUs Foundational to Inference and Agentic

4. AI Driving Demand for Silicon and Advanced Packaging

Notice what’s missing: the foundry. External foundry revenue. 14A customer names. Foundry breakeven timeline. These were the centerpieces of Intel’s narrative under Pat Gelsinger. Under Lip-Bu Tan, the foundry is repositioned as a supporting character, important, progressing, but not the headline.

This is smart messaging management. Tan understands that the market will pay for what’s measurable today (server CPU demand, margin expansion, supply constraints) and discount what’s prospective (external foundry revenue, 14A customer conversion). By leading with the product franchise and making foundry progress a secondary talking point, he lets the numbers speak first and the narrative follow.

That said, when you dig into the foundry details, the news is more encouraging than the understated presentation suggests:

• 18A yields: Zinsner revealed that Tan’s end-of-2026 yield target will likely be hit by mid-year. That’s a six-month pull-in on the company’s own internal plan. He also noted that Panther Lake volume is ramping 6–7x from Q1 to Q2, a genuine production ramp, not a token volume.

• 14A progress: Tan said 14A “maturity, yield, and performance are outpacing Intel 18A at a similar point in time.” If true, this is significant, it suggests Intel’s learning curve is compressing, which is exactly what you’d expect from an organization that has gotten its manufacturing discipline back.

• Advanced packaging: This is where Zinsner’s tone shifted from measured to almost excited. His prior guidance was “hundreds of millions” of external packaging revenue. On this call, he upgraded that to demand “in the billions of dollars per year kind of level,” with margins at or above foundry average. Packaging may end up being Intel Foundry’s first genuinely profitable external business, arriving before the wafer revenue that gets all the attention.

But here’s what Intel didn’t mention, and what bulls need to sit with: Intel’s own product designers are routing Nova Lake desktop compute tiles and Diamond Rapids server compute tiles to TSMC’s N2 process, not Intel 18A. The GPU and I/O tiles for Panther Lake also go to TSMC. In other words, Intel’s highest-performance 2026–2027 silicon, the chips that will determine whether Intel stabilizes server CPU share, will be manufactured by Intel’s foundry competitor.

Management can explain this as rational chiplet-era engineering: use TSMC N2 where density matters most, use 18A where PowerVia delivers a power-delivery advantage. That’s technically defensible. But it creates an uncomfortable credibility problem for the external foundry pitch. If Intel’s own chip architects, who have the most intimate knowledge of 18A’s capabilities and limitations, choose to send their flagship compute tiles elsewhere, why should an external fabless customer trust 18A with their flagship silicon? This is the single hardest question in the Intel bull case, and nobody on the earnings call asked it.

The GAAP Problem Nobody Wants to Talk About

Here’s where I need to push back on the euphoria. Intel reported non-GAAP EPS of $0.29. It reported GAAP EPS of $(0.73). The gap, over a dollar per share, is explained by $4.1 billion of restructuring and other charges (primarily a Mobileye goodwill impairment), $621 million of stock-based compensation, and $1.1 billion of mark-to-market losses on the government’s escrowed shares.

You can argue that any one of these is non-recurring or non-operational. But when the adjustments consistently dwarf the “real” earnings, when non-GAAP operating income is $1.7 billion but GAAP operating loss is $3.1 billion, it’s worth asking what the actual economic earnings of this business are.

Here’s the five-quarter trend:

The Q1 2026 gap is the widest in a year, driven almost entirely by the Mobileye impairment. Fair enough, goodwill impairments are non-cash and backward-looking. But the pattern of chronic GAAP losses alongside improving non-GAAP earnings is a feature of Intel’s financials that bulls need to at least acknowledge.

The other number that deserves scrutiny: adjusted free cash flow of negative $2.0 billion. Intel is still spending more on capital investment than it generates from operations, even in a blowout revenue quarter. Operating cash flow was $1.1 billion; gross capex was $5.0 billion. Government incentives ($107M) and partner contributions ($2.0B) close part of the gap, but the business is not self-funding its investment program. Zinsner reiterated positive adjusted FCF for the full year excluding the Fab 34 buyback, but that “excluding” is doing a lot of work given that the buyback was $14.2 billion.

This matters because the entire turnaround thesis assumes Intel’s product franchise can fund the foundry transition. Right now, the product franchise is generating $4B of segment operating income per quarter ($2.5B CCG + $1.5B DCAI), but after corporate overhead, foundry losses ($2.4B), and capex, the company is still cash-flow negative. The trajectory is improving, but it’s not there yet.

Using Scarcity to Lock In Customers

The most strategically important disclosure in the quarter wasn’t a financial metric, it was the Google long-term agreement. Google committed to deploying multiple generations of Intel Xeon processors across its data centers and co-developing custom IPUs with Intel.

Why does this matter? Because Google is one of three hyperscalers (along with AWS and Microsoft) that has invested heavily in custom Arm CPUs for its data centers. Google has Axion. If Google, a company that designs its own TPUs, its own CPUs, and its own networking chips, is signing multi-year Xeon commitments, it tells you something about the indispensability of x86 in production AI infrastructure.

Zinsner described the LTA structure: “Most of these agreements are structured with volume and pricing. They’re usually somewhere between three and five years.” He also noted that some customers prefer confidentiality, implying there are additional agreements beyond Google that Intel hasn’t announced.

The LTA strategy serves three purposes simultaneously. First, it gives Intel volume visibility, which lets the company plan capacity more rationally. Second, it locks in pricing in a supply-constrained market, protecting Intel from future price competition if supply loosens. Third, it creates switching costs, once a hyperscaler commits to three-to-five years of Xeon volume at negotiated prices, they’re not migrating that workload to Arm mid-contract.

This is Intel at its strategic best: using a temporary supply advantage to create durable demand commitments. The question is whether the commitments survive the end of the supply constraint.

The Terafab Question

I want to address the Terafab partnership separately because I think the market is conflating two different things.

The Tesla/SpaceX/xAI Terafab project is being treated as a 14A foundry customer win. The stock jumped 3% on Musk’s comments, and it was cited as a catalyst in virtually every post-earnings note. But listen carefully to what Tan actually said on the call:

“Elon and I, we believe that global supply chain is not keeping pace with the rapid acceleration in demand. We both share the vision that we’re going to learn a lot together, exploring the innovative way... in the process of the manufacturing.”

And when pressed by Tim Arcuri on whether this is a traditional foundry arrangement or something different, possibly a process licensing deal or even turning over an entire fab, Tan said: “Clearly it’s a very broad relationship and then, we would update you as we go.”

This is deliberately vague. “Very broad relationship” could mean many things. It could be a multi-billion-dollar 14A wafer commitment. It could be a co-investment in a new fab where Intel provides the process technology and Tesla provides the capital. It could be an advanced packaging deal for chips designed by Tesla and fabbed elsewhere.

What it almost certainly is not is a typical foundry engagement where Intel takes a GDS file and produces wafers. The language about “refactoring silicon process technology” and “unconventional ways to improve manufacturing efficiency” suggests Musk wants to rethink how fabs operate, which is interesting intellectually but commercially uncertain.

I’d treat Terafab as a long-dated option with high variance, not as a near-term revenue driver. The fact that Tan declined to provide further details, even after Musk publicly named Intel on Tesla’s earnings call, suggests the commercial terms are either not finalized or structured in a way that doesn’t map cleanly to traditional foundry revenue.

Intel’s Three Different Competitive Problems

It’s tempting to ask “who is Intel competing against?” as if the answer is one company. It’s not. Intel faces three structurally different competitive contests, and this quarter changed Intel’s position in each of them differently.

Against AMD, the fight is x86 versus x86: core count, performance-per-watt, roadmap cadence, pricing. AMD took 41.3% server revenue share in Q4 2025 and nothing in this quarter’s data shows that trend reversing. Tan acknowledged the gap, noting that Coral Rapids will restore multithreading to compete more effectively, which is an implicit admission that current products are disadvantaged. The supply constraint masks the share loss: Intel can sell everything it makes regardless of competitive position. When supply normalizes, the AMD question returns.

Against Arm and hyperscaler custom silicon, the fight is architectural control. Amazon, Google, and Microsoft are building their own CPUs not primarily for performance but to reduce dependency on merchant suppliers and capture margin internally. Intel’s response here isn’t just x86 compatibility, it’s the bundle: ASIC co-design, custom IPUs, advanced packaging, and captive manufacturing capacity. The Google LTA is evidence this response is working. Google signed a multi-year Xeon commitment despite having Axion. That means x86 offers something Arm doesn’t for those specific workloads, probably software ecosystem depth and orchestration maturity.

Against TSMC, Intel is simultaneously a customer and an aspiring competitor. Tan explicitly called TSMC “a very important partner” and described a multi-foundry approach. This is rational, Intel Products needs the best manufacturing, even if Intel Foundry wants external customers. But it creates the credibility tension described above: every tile Intel sends to TSMC is a tile that didn’t go to its own fabs.

The strategic implication is that Intel’s moat, if it rebuilds one, won’t be a single-product advantage. It will be a bundle: x86 installed base, server CPU orchestration, customer-specific IP, advanced packaging, domestic leading-edge manufacturing, and long-term supply commitments. No single element is defensible alone. Together, they might be.

What I’m Watching

The supply-demand balance through H2 2026. Intel’s pricing power exists because of supply constraints, not because of product superiority (though Granite Rapids early traction is positive). If supply catches up to demand, either through Intel’s own wafer start increases, TSMC expanding Arm CPU capacity, or a hyperscaler capex slowdown, the pricing premium evaporates. Zinsner said supply will increase every quarter going forward, but demand outlook has also improved. The sequencing matters enormously.

The 14A customer decision tree. Tan said he expects “earlier design commitments emerging beginning in the second half of 2026 and expanding into the first half of 2027.” This is the window. If H2 2026 passes without a named top-tier 14A customer (beyond the ambiguous Terafab arrangement), the foundry thesis weakens materially.

DCAI operating margin sustainability. The 30.5% margin in Q1 was flattered by pricing actions, improving Intel 3 yields, and volume leverage. Memory cost increases in H2 (”substrates are going up, T-glass, memory” per Zinsner) could compress margins even as revenue grows. If DCAI margins hold above 25% through year-end while revenue grows 20%+, the earnings power of the product franchise is genuinely transformed.

Gross margin trajectory. The 41% non-GAAP GM in Q1 included inventory reserve releases that won’t repeat. Q2 is guided at 39% due to heavier 18A mix (Panther Lake at 6–7x volume but below corporate-average margins). The path from 39% toward 45%+ requires 18A yields to keep improving and input cost inflation to moderate. Both are plausible but neither is guaranteed.

The Loop That Matters

Before addressing valuation, it’s worth making explicit the self-reinforcing mechanism that the bull case depends on, because if it’s real, it changes what Intel is worth, and if it’s not, nothing else in this quarter matters.

The loop works like this: AI-driven server CPU demand rises → Intel reallocates wafer starts from client to data center → DCAI revenue and margins expand (13.9% → 30.5% in four quarters) → higher internal wafer volume improves fab utilization → foundry losses narrow as fixed costs are spread across more units → improving 18A yields lower cost-per-good-die → better yields attract external foundry and packaging customers → packaging backlog converts to high-margin revenue → Intel generates more cash to invest in 14A and capacity → better process technology attracts more customers. Each turn of the loop makes the next turn easier.

This is the mechanism that distinguishes “Intel is having a good quarter” from “Intel is structurally more valuable.” The evidence from Q1 that the loop is engaging: DCAI margin expansion is real and accelerating; Zinsner confirmed 18A yields are ahead of plan; packaging demand upgraded from hundreds of millions to billions per year; Google, Nvidia, and undisclosed others signed multi-year LTAs. The evidence that the loop has not yet fully engaged: external foundry revenue was $174M; the foundry lost $2.4B; adjusted free cash flow was -$2.0B; and Intel still sends its own best silicon to TSMC.

The loop is spinning at the product level. It has not yet reached the foundry level. The question that determines whether Intel at $76–82 is cheap or expensive over the next 18 months is simple: Does the product-level flywheel generate enough cash and enough time for the foundry-level flywheel to engage before the AI infrastructure cycle peaks?

The Strategic Position

Intel is not a foundry company. It may never be a foundry company in the way TSMC is a foundry company, with dozens of fabless customers across the design ecosystem. What Intel is becoming, under Tan, is something different: a vertically integrated manufacturer of x86 server processors with a packaging services business and a selective external foundry operation funded by the U.S. government’s strategic interest in domestic leading-edge capacity.

That’s a less exciting pitch than “the Western TSMC.” But it may be a more honest and more achievable one. The product franchise is generating real cash, $4B of quarterly segment operating income, and the demand environment for server CPUs is the strongest it’s been in a decade. If Intel can convert that demand into sustainable earnings while gradually shrinking foundry losses through higher internal utilization and selective external wins (packaging first, wafers later), the financial model starts working without requiring Intel to become something it has never been.

The risk, as always with Intel, is that the current strength is cyclical, not structural. Server CPU demand is booming because AI infrastructure is being built out at unprecedented scale. That build-out will slow at some point. When it does, Intel’s supply constraints will vanish, its pricing power will erode, and the bears will re-emerge with the same arguments they’ve always had: AMD is eating x86 share, Arm is eating x86 TAM, and the foundry burns $10 billion a year.

The difference between this moment and 2024 is that Intel now has 18 months of demonstrated execution under a credible CEO, a balance sheet that can absorb another cycle, strategic investors who are economically aligned with the turnaround, and, for the first time, a demand environment where the product franchise and the manufacturing base reinforce each other rather than competing for capital. Those are real assets.

Whether they’re worth $400 billion depends on a race between two clocks: the product flywheel generating cash, and the foundry flywheel consuming it. This quarter showed the product clock accelerating. The foundry clock hasn’t slowed yet. The bull case is that the gap between them is closing. The bear case is that the AI cycle will peak before it does.

I don’t know which clock wins. But after Q1 2026, I know this is a race worth watching, and that’s a sentence nobody would have written about Intel eighteen months ago.

$INTC

General Disclaimer: The information presented in this communication reflects the views of the author and does not necessarily represent the views of any other individual or organization. It is provided for informational purposes only and should not be construed as investment advice, a recommendation, an offer to sell, or a solicitation to buy any securities or financial products.

While the information is believed to be obtained from reliable sources, its accuracy, completeness, or timeliness cannot be guaranteed. No representation or warranty, express or implied, is made regarding the fairness or reliability of the information presented. Any opinions or estimates are subject to change without notice.

Past performance is not a reliable indicator of future performance. All investments carry risk, including the potential loss of principal. This communication does not consider the specific investment objectives, financial situation, or particular needs of any individual.

The author and any associated parties disclaim any liability for any direct or consequential loss arising from the use of this material and undertake no obligation to update or revise it.

Subscribe now