We’re coming out of one of the most violent semiconductor rallies in history, and everyone is wondering if semis are overbought. SOX is up 65% year-to-date, posted its longest winning streak in 32 years, and ran 47% in 18 trading days. Earnings season was extremely strong. And the bubble calls are getting louder by the day. Burry has loaded up on puts. Your favorite macro guy is saying markets have lost all touch with reality and we are at peak cycle hysteria…

They might be right, but they’re asking the wrong questions.

Whatever happens in the next two weeks or two months is a sideshow to what the advent of AI actually means for markets and the world. And what AI represents is the most important development in economics in 200 years.

The war between capital and labor is ending. AI is what ends it. Labor is losing in a way it has never lost before and that stands to reshape markets and investing forever.

The Capital-Labor Cycle

The global economy is fundamentally a negotiation between two factors: Capital, which provides tools, land, and money, and Labor which provides time, effort, and skill.

Economic history is the drama of how the value they produce together gets divided.

For most of recorded history that negotiation was tilted one way or another, flipping back and forth with recent events.

After the Black Death wiped out a third of Europe’s population, labor briefly held all the cards. Peasants demanded wages, abandoned manors, and watched serfdom crumble. By the Gilded Age, capital had reclaimed the upper hand so thoroughly that a handful of industrialists had their own private armies. The postwar period swung the pendulum back: unions had teeth, real wages tracked productivity, and the median family could buy a house on a single income. Then around 1980 the pendulum reversed again, and we’ve been living in capital’s afternoon ever since, watching the labor share of GDP grind lower decade after decade.

This oscillation is a huge part of the historical debate between economists. Marxists, classical economists, and supply-siders disagree on almost everything, but they agree on the framing. Whether you read Ricardo, Piketty, or Friedman, you’re reading some variant of the same fight.

Marx had the most ambitious solution. His prescription was that labor would eventually seize the means of production and end the negotiation by abolishing one of the parties. The proletariat would own the tools it used. Game over, with labor on top forever.

But, that isn’t what happened. Marx assumed labor would always be necessary to operate capital, and the assumption held for the entire industrial revolution. Every wave of automation, from the spinning jenny to the industrial robot, displaced specific tasks while creating new ones, because a human was always needed somewhere in the loop. A weaver lost her job; a machine operator gained one. A bank teller was replaced by an ATM; a software engineer was hired to maintain the network. The aggregate share might shift at the margin, but labor stayed structurally indispensable. You needed humans to run the capital.

Every previous productivity revolution was therefore, at best, only a partial victory for capital. A reallocation, not a routing.

AI is something different.

The Truth About AI

Most people think of AI as just another technology, similar to the steam engine, electricity, and the internet as a general purpose technology, with the familiar pattern repeating: workers displaced, workers retrained, prosperity rises, the system absorbs the shock.

This view misses an important difference about AI.

Every prior automation wave automated specific muscle movements, motions, or calculations. It substituted for the body and for the most mechanical parts of the mind. The seat of judgment, taste, synthesis, language, and reasoning remained the exclusive property of humans, and as the lower rungs of the cognitive ladder got automated away, humans simply moved up the ladder.

AI is aimed squarely at replacing all rungs of the ladder, but especially the ones that automation has never touched before. The diagnosis. The strategic call. The judgment that used to belong to a senior partner. The synthesis that justified the corner office.

There’s an obvious objection here. Every previous technological revolution also obsoleted whole job categories, and every revolution created new ones to absorb the displaced. The steam engine didn’t envision software engineers. The internet didn’t envision prompt engineers. Why should this time be any different?

The difference this time is specifically how general the automation will be. Artificial General Intelligence seeks to replace humans across all applications. New jobs required new skills, but those skills lived inside human heads, and the heads were previously the bottleneck.

And don’t assume physical labor is the refuge. Robotics has been “five years away” for decades because the bottleneck was never the hardware. It was the brain. Perception, planning, dexterity, the judgment to handle the long tail of edge cases every real-world job throws at a worker. That’s what has held up the physical-labor automation wave for half a century, not the cost of servos or the precision of actuators (and to the extent the servos and actuators have been a problem, artificial design systems can solve that too).

The skyrocketing capability of AI is that brain. Whether general-purpose humanoids arrive in three years or ten, they arrive. And well before then, narrow systems will eat specific physical jobs the same way they’ve started eating specific cognitive ones. The carpenter and the programmer face the same problem on different timelines.

Once the substitute for human cognition becomes a line item on a capex budget, the structure of the labor market will change. In equilibrium, wages equal the marginal product of labor. Introduce a substitute whose capability rises every year and whose unit economics are dominated by capex (chip fabs, data centers, training runs), and the marginal product of human labor for any given task collapses toward the marginal cost of the substitute. The wage collapses with it, task by task and role by role.

Labor stops being a separate factor of production and starts becoming a particular form of capital, machinery you can buy by the GPU-hour. The negotiation between capital and labor doesn’t shift. It dissolves.

This is the prediction Marx couldn’t see, because he couldn’t conceive of capital that didn’t need labor to animate it. His theory of value rested on labor as the thing that made capital productive. Remove that premise, not just at the assembly line but at the cognitive level, and his framework still applies, but only to a world that no longer exists.

Where Marx Was Right

Marx was wrong about the path, but he was actually kind of right about the ending.

He predicted a world where scarcity was “solved” by humanity. Humans would be freed from work because the machines did everything. He thought labor would inherit this paradise by seizing the machines, creating a socialist utopia.

Instead, capital is seizing labor and creating a capitalist utopia.

The central political and economic question of the coming decades won’t be how to divide output between capital and labor. That question is closing. The new question is which capital will have the most leverage, and what to do with the people who used to be the other half of the equation.

Investing For The Future

If you’re reading this newsletter you’re probably less interested in the political philosophy and more interested in what to own.

The answer follows directly. If labor’s share is going to compress and capital’s share is going to expand, the natural place to look is at capital that produces the substitute for labor. Right now, that capital is accruing in three places:

1. Compute owners (hyperscalers). The handful of companies operating the data centers where the substitute actually runs.

2. Hardware providers. The picks-and-shovels layer: chips, memory, interconnect, packaging, the equipment makers behind them.

3. Frontier labs. The model builders themselves, where the cognitive substitute is being created.

But the point is not just to rehash the obvious “own AI” trade and move on.

The point is that if AI turns cognition into capital, then the companies that own the capital layer are not just participating in the next tech cycle. They are becoming the new claims on the future income stream of the economy.

That is the real reason the AI trade is so strange. On normal cyclical metrics, a lot of this looks insane. The stocks have run too far. The earnings revisions are too aggressive. The capex numbers are too large. The narratives are too clean. All of that may be true.

But if the underlying shift is real, then the market is not just paying for another semiconductor cycle. It is paying for a world where more and more of what used to be paid out as wages gets redirected toward compute, chips, memory, power, models, and distribution.

That does not mean every AI stock is a buy. It does not mean valuation no longer matters. It does not mean there won’t be brutal drawdowns, failed projects, overbuilt capacity, or periods where the whole thing looks ridiculous.

But it does mean the right question is not simply whether semis are overbought.

The right question is whether the market is beginning to price the capitalization of labor.

If it is, then the imperative is clear: focus on the assets that make labor less necessary. The companies manufacturing the hardware bottlenecks. The infrastructure builders. The companies that turn human judgment into a software expense and human output into a compute workload.

Marx thought labor would seize the means of production.

Instead, capital is building the means of replacing labor.

That is the final triumph of capital.

And that is the real AI trade.

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice, and is not a recommendation or offer to buy any securities. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

The Bull Case for AI: Why the Doom Loop Won’t Sustain

Lately, an extremely sharp, economically rigorous “Bear Case” for AI has been making the rounds courtesy of and .

Citrini’s recent piece, in particular, made a splash today.

Citrini Research

THE 2028 GLOBAL INTELLIGENCE CRISIS

Preface…

Read more

4 months ago · 2191 likes · 76 comments · Citrini and Alap Shah

The story goes like this: starting in the next few years, highly capable AI agents begin replacing white-collar workers en masse. Displaced professionals lose their six-figure salaries and slash their discretionary spending. This triggers a negative feedback loop, collapsing corporate revenues, causing more layoffs, cratering enterprise SaaS valuations, and starting a brutal default wave across the multi-trillion-dollar private credit and prime mortgage markets. Federal tax receipts evaporate just as the demand for government bailouts skyrockets. The doom loop feeds on itself until it takes us to a very dark place.

It’s a terrifying scenario. It’s logically coherent. And it’s built on a foundational error about how technological deflation propagates through an economy.

The bear case looks at AI and sees a machine designed to destroy the way we currently make money. The bull case looks at AI and sees a machine designed to destroy the cost of solving problems. These lead to radically different conclusions, but only if you take the timing problem seriously.

What This Means for Markets

The legacy enterprise SaaS market is in serious trouble. Markets have already worked this out. Per-seat pricing doesn’t survive a world where AI agents do the work of ten humans, and SaaS struggles in any scenario where the marginal cost of software drops. The multi-trillion-dollar private credit market, which financed leveraged buyouts of these software companies during the ZIRP era at 15-20x revenue multiples assuming perpetual per-seat growth, is going to take a haircut.

But this is localized creative destruction, not systemic collapse. The dot-com bust didn’t permanently destroy the economy because the underlying technology kept getting more useful even as overvalued companies imploded.

The broader equity market won’t collapse. It will rotate. We will get panic days along the way, clustered around weak job reports and earnings misses, but capital will move from companies that profited from cognitive scarcity (enterprise SaaS, professional services, staffing firms) to companies that profit from cognitive abundance (infrastructure, energy, compute, physical-world services). We’re already seeing the early stages of this.

Big Tech itself is more resilient than the bear case implies. A tech giant whose gross revenue drops 30% because consumers are broke will have also automated a massive share of its own costs. They will collect fewer dollars but keep a vastly larger percentage as profit. And these companies are the ones selling the AI agents to the rest of the economy. Falling labor costs lead to more AI investment, which accelerates the technology, which creates the new economic activity that displaced workers migrate toward.

The losers are identifiable: over-leveraged private credit, per-seat SaaS, the staffing economy, anyone levered to the displaced white-collar worker. The winners are equally clear: compute infrastructure, energy, physical-world buildout, and the platforms selling intelligence itself.

The Demand Explosion

White-collar layoffs are coming. As AI agents become highly capable, companies are going to use them to cut costs, and a significant number of people will lose their jobs.

AI is primarily an engine for supply-side deflation, the “good” kind. Prices don’t fall because we’re broke. They fall because the cost to produce things plummets. The bears will respond: sure, but AI attacks the labor market itself, which is the source of purchasing power. That’s a fair objection. But the scale of new demand will be much larger than the bears expect.

Bears expect a near-total wipeout of white-collar employment because they implicitly assume there is a finite amount of cognitive work to be done in the world. If an AI can do the work of 10 engineers, a company fires nine, keeps one, banks the savings, and the economy shrinks. Some companies will do exactly that. But on a macro level, this “fixed pie” assumption has been wrong after every major technological disruption in history, and for a specific reason.

Right now, almost every business on earth runs on triage. Because human cognitive labor is expensive, companies only build features, solve problems, and serve customers that clear a high ROI bar. Ask a tech company for a niche software integration and the product manager says no. They can’t justify $100,000 of engineering time on an “edge case” for 500 users.

When the marginal cost of cognitive labor drops by 90%, the triage era ends. A company might lay off 7 out of 10 engineers. But those remaining 3 won’t be doing the same work the 10 used to do. They’ll be directing fleets of AI agents to build things the company previously couldn’t afford to care about: software that adapts to every user, customer support that resolves problems instead of deflecting them, products localized for markets that were never worth the headcount.

Today, thirty million U.S. small businesses can’t afford outside counsel. Patients skip the specialist because the copay isn’t worth it. Students don’t get tutors because their parents can’t afford it. These people are consuming zero in sectors that bears are modeling as contracting. When those sectors get 90% cheaper, domestic demand expands into a population that was entirely priced out.

The bear case models a shrinking pie. The actual story is companies building what they never could have justified and millions of people buying what they were priced out of entirely. The pie is about to get much bigger.

Displacement Is Deflation

The sectors experiencing the most displacement are, by definition, the sectors where cost deflation arrives fastest. If a law firm uses AI to do the work of 60% of its associates, it has two choices: pocket the savings or lower prices. In a competitive market, the firm that doesn’t pass those savings to clients loses them to the one that does. Legal services don’t get cheaper five years after firms lay off associates. They get cheaper immediately, because the layoffs and the price drops have the same cause.

This doesn’t eliminate the pain, but it eases the transition.

What Actually Gets Cheaper

The bull case depends on cost-of-living deflation arriving fast enough to partially offset lost income. But cheaper legal research doesn’t help a displaced analyst make rent. The question is what happens to the line items that actually dominate a household budget.

Some categories deflate fast. Healthcare is the biggest. It’s the largest non-housing expense for most American families, it’s overwhelmingly driven by cognitive labor costs, and it’s already seeing early compression from AI tools. When an AI can triage patients, read imaging, and handle the billing bureaucracy that consumes a third of every healthcare dollar, costs fall structurally.

Education follows the same pattern. AI tutoring is approaching human-tutor quality at a fraction of the cost, and professional development gets radically cheaper precisely when displaced workers need it most. Financial services, tax preparation, and insurance are already commoditizing.

Food and energy deflate more modestly. Grocery margins are razor-thin, so even small efficiency gains from AI-optimized logistics and waste reduction pass through to consumers quickly. But food has a physical production floor. Energy benefits from AI-optimized grids and cheaper renewable design, but the buildout timeline is years, not quarters.

Housing is the main problem. Land doesn’t get cheaper because AI exists. Zoning doesn’t reform itself. Physical construction is peak Moravec’s Paradox, and the same labor demand that absorbs displaced white-collar workers keeps construction wages elevated. AI can compress design and permitting timelines, but the physical build is the dominant cost and it isn’t deflating during the displacement window.

The math: a displaced worker whose healthcare drops 40%, groceries drop 15%, and financial services drop 80% might see non-housing costs fall 25-30%. But housing is 35-40% of most household budgets and doesn’t move. Total expenses decline maybe 15-20%. That’s the difference between treading water and drowning, but it doesn’t close the gap for someone whose income fell 40%. The remaining delta is what makes the policy bridge non-optional.

Where the Surplus Goes

Cloud infrastructure is the direct precedent. AWS, Azure, and GCP operate as a profitable oligopoly, competing hard enough on pricing to prevent any one player from extracting monopoly rent, while each generating enough margin to fund massive capex buildout. Shopify, Stripe, Cloudflare, and the entire SaaS ecosystem were built on that oligopoly’s infrastructure, driving prices down for end users while the cloud providers kept printing margin.

The same structure is emerging in AI. A handful of frontier labs control the model layer. The businesses building on top of those models are already competing on price, and that competition is what pushes the surplus to consumers.

If that layer consolidates into a monopoly, the surplus gets captured as rent rather than passed through as lower prices. Displaced workers lose their jobs, but the cost of living doesn’t fall nearly fast enough to compensate. The model provider collects the delta as profit, the deflationary transmission breaks, and the bear case wins. A one-winner outcome at the model layer is the bear case for the broader economy, even if it’s a bull case for that company’s stock.

The Timing Problem

The entire bear thesis is fundamentally a sequencing argument: the pain of displacement arrives before the abundance of cheap intelligence does. Workers lose their jobs in Year 1, but the “good deflation” that makes a $70,000 salary feel like $120,000 doesn’t fully materialize until Year 4 or 5. In the gap, people default on their mortgages, private credit blows up, and the doom loop ignites.

This is the strongest version of the bear case.

First comes the compression. AI agents reach “good enough” capability for a wide range of white-collar tasks. Companies begin aggressive headcount reductions in the most automatable functions: back-office operations, basic legal and financial analysis, customer support, junior software development. Private credit portfolios loaded with leveraged software buyouts from the ZIRP era start taking losses. Unemployment in affected sectors rises. This is the period where the bear case looks most prescient, and it’s some of what we’re already seeing.

But overlapping with the compression, two countervailing forces build. The cost of services starts dropping as AI-powered healthcare, legal services, education, and financial planning reach consumers at a fraction of current costs. A displaced worker’s dollar stretches further each quarter.

Simultaneously, capital freed from cognitive labor costs rotates into physical infrastructure, energy, construction, and physical-world services that AI can optimize but robots can’t yet perform, creating demand for labor in sectors largely immune to AI displacement.

Eventually a rebalancing emerges. Nominal wages in many categories may be lower than their 2024 peaks, but real purchasing power has stabilized or improved due to relentless cost deflation. GDP growth, driven by the explosion of newly viable economic activity, generates the tax receipts needed to service debt.

The bull case doesn’t deny the pain of the compression. It argues that the countervailing forces begin early enough, and compound fast enough, to prevent the doom loop from becoming self-sustaining.

Deflationary doom loops in history (the 1930s, Japan’s Lost Decade) share a common feature: there was no countervailing force pushing costs down independently of collapsing demand. Things only got cheaper because nobody was buying. AI-driven supply-side deflation is categorically different. The cost reductions happen because the technology is improving, regardless of demand. An AI model that does legal research for $0.10 instead of $500/hour doesn’t stop getting cheaper because a recession hits.

Moravec’s Paradox as Economic Buffer

In robotics and AI, Moravec’s Paradox observes that tasks trivially easy for humans in the physical world (walking upstairs, snaking a clogged drain, pulling copper wire through a 1920s ceiling) are extraordinarily hard for machines. Meanwhile, tasks that feel sophisticated to us, like legal analysis, medical diagnosis, and code generation, are comparatively easy to automate.

Intelligence is easy. Handiness is hard.

This creates a multi-year gap that functions as a massive economic buffer. AI will drive down the design and engineering costs for next-generation nuclear reactors, high-speed rail, smart power grids, and millions of new homes. But because the robots aren’t ready to build them yet, we will need human hands. The capital saved on automated cognitive work rotates into paying humans to work in the physical world, where they’re still the only game in town.

This is already visible in wage data. Electricians, plumbers, HVAC technicians, and skilled tradespeople are commanding premium wages that have grown faster than inflation for years. AI-driven displacement of white-collar workers will push more people toward these fields, but the demand for physical infrastructure is growing even faster, especially with thousands of new data centers coming online.

The physical-world labor shortage runs deep beyond the skilled trades: delivery drivers, warehouse operations, home services, elder care. These roles don’t require years of retraining, which matters when the timeline for absorption is measured in quarters, not decades. For younger workers and those without deep professional identity anchors, the Moravec’s Paradox buffer is real and large. But it doesn’t catch everyone.

The Missed Landings

The displaced white-collar workforce has one structural advantage no previous disrupted labor class has had: the ability to hedge their own obsolescence. A $150,000 financial analyst with a 401(k), liquid savings, and basic financial literacy likely already owns the companies that are making him obsolete. Factory workers in the Rust Belt couldn’t buy equity in the robots replacing them, but a senior paralegal can buy shares in the platforms selling the AI that’s replacing her.

Still, the displaced workers who land on their feet share a common trait: optionality. They’re young enough to retrain, flexible enough to relocate, skilled enough to move up the leverage curve, or savvy enough to get ahead of this with their investing. The people who don’t land are the ones without those options.

They’re the 54-year-old contract attorney in a midsize city whose entire career is document review. The mid-level financial analyst who spent 20 years building expertise in a function an AI agent now handles in seconds. The back-office operations manager, the senior paralegal, the staff accountant at a regional firm. There are roughly 5 to 10 million workers in the U.S. whose skills map almost entirely to automatable cognitive tasks, who are deep enough into their careers that “learn to manage AI agents” isn’t a realistic path, and who live in markets without enough physical-economy demand to absorb them.

The cultural dimension shouldn’t be minimized. These are people who built identities around professional knowledge work. A senior financial analyst becoming a facilities manager for a data center campus may earn a comparable wage, but the psychic cost of that transition is significant. Multiply that across millions of households and you get a political force, not just an economic statistic.

But the core economic problem is simpler and more urgent: nominal obligations. A displaced analyst whose cost of living drops 20% over three years still has a mortgage payment due this month. Student loans, car notes, credit card balances: none of these adjust downward with deflation. For this cohort, the gap between losing income and benefiting from cheaper services isn’t an abstraction. It’s the window where they default, lose their home, or drain their retirement savings. This is where the doom loop has its best chance of igniting, not economy-wide, but concentrated in specific metros and demographics.

The good news is that this is a bounded problem. It’s not 50 million workers. It’s an identifiable cohort, concentrated in predictable sectors and geographies, facing a quantifiable income gap over a finite transition window. That makes it targetable, which is what separates a solvable policy challenge from a civilizational crisis.

The Policy Bridge

Even if living costs drop, a $3,500/month mortgage from 2024 still requires nominal dollars. If your income drops from $150,000 to $90,000, your mortgage-to-income ratio blows past sustainable levels regardless of how cheap groceries get. Debt doesn’t adapt to deflation.

Policy has to step in here, and this is where the bull case is most vulnerable to execution risk.

The Fed can buy agency MBS to keep mortgage spreads from spiraling. Congress can bridge the income shock with fiscal transfers. Because AI is simultaneously crushing production costs, the inflationary impact of that fiscal expansion is dampened relative to a normal stimulus cycle.

But if policymakers move too slowly, mortgage distress in tech-heavy metros could metastasize before cost deflation catches up. If they move too aggressively, they risk inflationary overshoot. The bull case requires that the right tools are deployed with reasonable competence, which, given Congress’s track record, is not a certainty.

COVID showed Washington can move fast when it has to — $2.2 trillion in emergency fiscal response in roughly two weeks. The AI displacement scenario also has a structural advantage COVID lacked: advance warning. AI displacement is arriving with years of lead time and an active policy debate already underway. And displaced white-collar workers in swing-state suburbs are exactly the demographic both parties compete hardest to protect.

None of this guarantees competent execution. Some amount of policy response will be needed to prevent millions of people entering poverty or losing their homes.

Long-term, the most powerful defense is the denominator. If AI unlocks massive productivity gains, a $30 trillion economy growing at 5-7%+ annually generates substantially more tax revenue even as individual wages decline. An AI-driven GDP expansion is realistically one of the few paths by which the U.S. might actually grow its way out of its $34 trillion national debt.

The Inequality Ratchet

Even in the bull case, where displaced workers find new roles and cost deflation stretches their dollars further, the distribution of gains is radically unequal.

When AI cuts a company’s cognitive labor costs by 60%, the surplus flows somewhere. Some goes to consumers through lower prices. Some goes to workers who remain, now leveraging AI to produce far more per hour. But the largest share accrues to the owners of the capital layer.

AI represents the final triumph of capital over labor. Every previous wave of automation shifted the balance incrementally. AI threatens to do it decisively, across nearly every cognitive function, within a single decade. For investors, this is the most important sentence in this entire piece. If you are on the right side of this trade, owning equity in the companies that provide, deploy, and build infrastructure for machine intelligence, the wealth creation will be historic. The companies selling the tools of cognitive automation will capture a share of every dollar of labor cost eliminated across the entire economy. That’s not a sector bet. It’s a claim on the broadest productivity transfer in modern history.

But the same dynamic that makes this so bullish for capital owners is what makes it politically unstable.

A displaced paralegal who retrains into facilities management might see her real purchasing power hold steady or fall modestly. Meanwhile, the equity partners at the firm that automated her role see per-partner profits double. You don’t need mass unemployment to generate political instability. You just need a large enough population watching a small enough elite capture a visibly disproportionate share of a growing pie.

A sovereign wealth fund is the most elegant solution. If the federal government takes an equity stake in the dominant AI platforms, whether through direct investment, licensing, or a negotiated exchange for the regulatory clarity these companies desperately want, it converts private rent extraction into a public asset. Norway’s Government Pension Fund is the proof of concept: a trillion-dollar sovereign fund built on resource extraction that distributes returns broadly across the population. If intelligence becomes the new oil, a public equity stake in the infrastructure layer is the cognitive equivalent of the Norwegian model.

The bears are right that unwinding the premise of scarce human thought will be chaotic. Private credit will bleed. Pockets of unemployment will be severe. The identity crisis will be real. But the bear case models a world where jobs disappear and nothing else changes. In reality, the same force destroying those jobs is simultaneously collapsing the cost of living, unlocking trillions in new economic activity, creating demand for human labor in the physical world, and expanding the fiscal space to manage the transition.

The biggest risk to owning the winners isn’t that the technology fails, it’s that it succeeds so well that the government can’t allow all of that value to concentrate in private hands.

And the white-collar workers at the center of this disruption are not helpless passengers. They are, uniquely in the history of technological displacement, a class of workers with the capital, the financial literacy, and the market access to position themselves on the winning side of the trade. The AI age doesn’t have to make them poor. For those who see clearly and act early, it can make them owners of the most transformative technological revolution in history.

On the other side of the transition, a worker earning $90,000 in a world where healthcare, education, legal services, and financial planning cost a fraction of what they do today may have more real purchasing power than he did earning $150,000 in 2024. Coupled with technological advances it’s quite likely he lives a happier and healthier life in this new world.

The transition will be harder than the optimists think and less catastrophic than the pessimists fear. The doom loop is a myth, but the promise of opportunity is not.

Good luck out there.

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

The next two weeks are loaded. MSFT and META report Tuesday, AAPL Wednesday, GOOG and AMZN the week after. This is the quarter where the AI trade either proves it can monetize or starts looking like an expensive science project.

The Macro Backdrop

We’re heading into this earnings season with an interesting cross-current of signals.

Gold just broke $5,000. The yellow metal is trading above $5,050 as of this writing, a level that seemed absurd just a year ago. The breakout reflects a cocktail of concerns: tariff uncertainty, geopolitical tensions, and a growing sense that central banks aren’t done accumulating. When gold runs like this, it’s usually telling you something about risk appetite. Smart money is hedging.

Japan’s bond market is flashing warning signs. JGB 10-year yields have blown out above 2.2%, levels not seen in over a decade. But this isn’t just the BOJ normalizing policy. The move reflects deeper concerns: Japan’s fiscal situation is deteriorating, debt-to-GDP remains the highest in the developed world, and inflation is proving stickier than expected. The market is starting to price in the possibility that Japan can’t keep a lid on yields forever. The Nikkei sold off nearly 2% on Friday as investors reassess.

US 10-year yields are rising in sympathy. We’ve come down from the 4.7% panic levels of early January, but we’re creeping back up toward 4.2%. Part of this is the JGB spillover effect. When the world’s largest creditor nation sees bond yields surge, it ripples everywhere. Japanese investors have been massive buyers of US Treasuries for years. If they start demanding higher yields at home, or worse, start selling foreign bonds to repatriate capital, that’s a headwind for global risk assets. For growth stocks, every 25bps matters for DCF models. The bond market isn’t giving us an all-clear.

What does this mean for earnings? The macro isn’t hostile, but it’s not a tailwind either. Companies need to deliver on their own merits. There’s no multiple expansion coming from rate cuts or a liquidity surge. It’s a “show me” environment.

The setup going in is interesting. Many of these names are 10-20% off their highs despite the broader market holding up. Expectations are elevated but not euphoric. The bar for “beat and raise” is high, but the stocks have already pulled back enough that truly good numbers could rip.

Signs of a Quality Rotation

Here’s something interesting: after months of struggling, the Magnificent Seven are showing early signs of life. The MAGS ETF, which tracks the seven mega-cap tech names, has been chopping sideways since November. But in the last few sessions, we’re seeing buyers step back in.

This matters because the market has been ignoring these stocks for months. The equal-weight S&P outperformed the cap-weighted index for most of Q4. Money rotated into small caps, value, and international. The “Mag7 are too crowded” narrative dominated.

Now the setup is flipping. With yields inching higher and growth coming off a major rotation, investors are remembering why they owned quality in the first place. These companies print cash, have pricing power, and are the primary beneficiaries of AI.

Earnings could accelerate this rotation. If MSFT, META, and the rest deliver strong numbers, the “quality bid” could turn into a full rotation back into mega-cap tech. The stocks have de-risked through time and price. They’re not as stretched as they were six months ago. Good earnings plus a quality rotation could be a powerful combination.

The Big Three Questions

1. Is AI capex still accelerating or are we seeing the first cracks?

Last quarter, every hyperscaler guided capex higher. That’s what funded the whole AI infrastructure trade. We need to keep hearing some version of “we can’t build fast enough.”

MSFT is the one to watch here. Satya said last quarter that he’s not chip-constrained anymore, he’s “warm shell” constrained. Translation: they can get the GPUs, but they can’t build datacenters fast enough to house them. If that’s still true, power and datacenter names keep working. If capex guidance disappoints, the whole stack gets hit.

2. Are companies actually attributing revenue to AI?

We’ve been waiting for “Phase 3” of the AI trade: software monetization. Last quarter we saw early signs. META said AI was improving ad targeting. CRM started breaking out Agentforce numbers. GOOG’s cloud reaccelerated.

This quarter needs to show that trend continuing. I want to hear specific AI revenue numbers, not just “AI is embedded in everything we do.” That line worked in 2024. It doesn’t work anymore.

3. What’s the guidance tone?

The Mag 7 has stayed magnificent through one formula. Beat and raise. That’s the name of the game for these companies. Any sign that things are slowing around the margins will be bearish. Any sign that time horizons are being pushed back will be bearish.

This Week: The Main Event

Wednesday is a big one. MSFT and META reporting on the same day means we’ll get a read on both enterprise AI adoption and consumer AI monetization in one shot.

Microsoft reports after the bell Wednesday with street estimates at $3.92 EPS on $80.3B revenue. Azure growth rate is the number everyone’s watching. It needs to hold above 30%. The stock is 15% off highs with a perfect 4/4 beat streak over the last year. The real tell will be capex guidance for FY26. If they guide higher, it’s bullish for the entire AI infrastructure stack. If they guide flat or lower the hardware names will get hit.

Meta follows with estimates of $8.20 EPS on $58.5B revenue. They got punished last quarter for guiding capex higher without showing enough ROI. On last quarter’s earnings call Zuck nodded to the idea that they are seeing ROI in their core business. This quarter they need to prove the spend is paying off. The thesis is that AI is improving ad targeting and driving better ROAS for advertisers. If they can show that in the numbers, the stock looks cheap at 22x forward. If it’s still “trust us, AI is helping,” expect more selling.

Apple on Thursday is the AI laggard. The street expects $2.67 EPS on $138.5B revenue. Apple Intelligence adoption is the wildcard here. They were late to the AI party but have massive distribution. Services growth needs to stay strong. iPhone units matter less than ASPs. The stock is 12% off highs and trading like a bond with upside optionality on AI.

In particular pay attention to any talk of Siri, Apple’s voice assistant. It’s getting a makeover this year using Google’s Gemini models. A blowout Siri release could be big for turning around AAPL’s “AI Laggard” narrative.

Next Week: The Second Wave

The action continues into February with AMD on Tuesday, GOOG on Wednesday, and then AMZN, TEAM, and TPG on Thursday.

AMD ($1.32 EPS, $9.7B rev) is positioning itself as the Nvidia challenger. The MI300 ramp is the whole story. Data center revenue needs to accelerate to justify the 38x forward multiple. They’re only 6% off highs, so the market is already pricing in a decent quarter.

Google ($2.64 EPS, $111.4B rev) had cloud reacceleration last quarter and needs that trend to continue. Gemini monetization is starting to matter. The stock is basically at the highs, which tells you the market already likes the setup going in. has been one of my favorite names, but it has dramatically repriced. We need to see a positive beat to propel Google’s momentum higher.

Amazon ($1.94 EPS, $211.2B rev) is all about AWS growth. It needs to stay above 15%. Retail margins matter too, but AWS is the multiple driver. The stock is 8% off highs with strong buy ratings across the board.

Then there are the two names I find most interesting from a risk/reward standpoint.

Atlassian is 57% off its highs. That’s not a typo. The “AI kills SaaS” narrative has absolutely crushed this stock. Either that thesis is right and this is a melting ice cube, or the market has massively overcorrected.

What’s the dirty secret of the “AI kills SaaS” story? Even the big AI labs use SaaS. OpenAI uses Slack, Anthropic uses Workday. Their earnings could be a turning point. I’ve been nibbling on the thesis that enterprise software isn’t going away just because AI exists.

TPG is 15% off highs with rates coming down. They’re the most tech-forward of the big PE shops, with heavy exposure to software and growth companies through their Rise fund and tech-focused strategies. Private equity is quietly one of my contrarian AI plays because they can apply efficiency gains across entire portfolios. TPG in particular has the expertise to actually implement AI at their portfolio companies rather than just talking about it.

The Setup In Pictures

Look at that dispersion. GOOG is basically at highs while TEAM is down 57%. The hyperscalers are in decent shape at 10-15% off peaks, which means good numbers should move stocks but great numbers are needed for big moves. The beaten-down names like TEAM and TPG have more torque if they deliver

The beat history is solid across the board. These companies know how to manage expectations. But that’s almost the problem. The question isn’t whether they beat, it’s whether the beat is big enough and the guidance is strong enough to actually move the stocks. Beating by a penny and guiding in-line isn’t going to cut it this quarter.

The Bottom Line

This earnings season is a checkpoint for the AI trade. The infrastructure buildout thesis has been working, but the market wants to see revenue follow capex. We’ve been patient. We’ve funded the buildout. Now show us the returns.

To that end, pay attention to where and if the actual revenue gains show up in places that BENEFIT from AI, not just sell it.

What would make me more cautious: capex guidance disappointments, Azure or AWS deceleration below 25%, or management teams suddenly talking about “efficiency” instead of growth. If the tone shifts from “we can’t build fast enough” to “we’re optimizing our spend,” that’s a problem for the whole trade.

The stocks have pulled back enough that good numbers should be rewarded. But “good” won’t cut it anymore. We need “great” on capex, specific AI revenue attribution, and confident 2026 guidance. Anything less and these stocks are probably range-bound until we get better data.

Good luck out there.

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

On Saturday, U.S. forces captured Venezuelan President Nicolás Maduro in a dramatic strike on Caracas. Within hours, Secretary of State Marco Rubio stood at Mar-a-Lago and delivered a warning that sent shockwaves through Havana: “If I lived in Havana and I was in the government, I’d be concerned.”

That’s about as direct as it gets short of an outright threat.

The Donroe Doctrine

Marco Rubio was raised in Miami by Cuban immigrant parents. He grew up hearing stories of how communism destroyed lives on the island. The “Donroe Doctrine”—Trump’s reinterpretation of the Monroe Doctrine—establishes that no state in the Americas can serve as a refuge for regimes that sustain criminal networks or threaten U.S. hegemony. Cuba, which Rubio calls “the head of the snake,” is the next logical target.

The Venezuela operation proved capability and willingness. Thirty-two Cuban agents were killed in the strike. Last year, Rubio reactivated Title III of the LIBERTAD Act, allowing private lawsuits against anyone “trafficking” in confiscated Cuban property. And Cuba is economically collapsing. Without Venezuelan oil subsidies, which the Maduro capture has now severed, Havana faces a financial void. Trump himself declared “Cuba is ready to fall.”

Whether through collapse or intervention, the endgame is regime change. The timeline is uncertain but the direction is clear.

The Certified Claims: $8 Billion in Dormant Assets

When Castro nationalized American property in 1959-1960, the U.S. government created a legal framework for eventual recovery. The Foreign Claims Settlement Commission certified 5,911 claims with a principal value of $1.85 billion. With 65 years of 6% simple interest, the total value today is approximately $8 billion.

These aren’t theoretical claims. They’re legally certified property rights recognized by the U.S. government, waiting for regime change to be enforced. The LIBERTAD Act explicitly requires claims resolution as a condition for normalizing relations with any post-Castro government.

For mega-caps like Exxon ($71M claim), Chevron ($50M), and Coca-Cola (~$27M), their Cuban exposure is a rounding error. But there’s one public company where the claims represent meaningful optionality.

The Claims Play: Marriott International

Marriott controls the second-largest certified claim against Cuba, $181.8M in 1960s dollars, through its acquisition of Starwood, which had acquired ITT Sheraton. With interest, the floor value exceeds $890 million. But the real story is what they have claims to.

The claim includes land adjacent to Jose Marti International Airport that was confiscated for runway expansion, plus legacy telecom infrastructure from the former Cuban Telephone Company. These aren’t just financial claims, they’re negotiating leverage. Any new government expanding the airport or modernizing telecom will need to deal with Marriott.

Marriott is also the only major U.S. hotel company already operating in Cuba, with OFAC-licensed properties including the Four Points by Sheraton Havana. When the floodgates open, they’ll move first.

The math: Cuba at maturity could support 100,000+ quality hotel rooms (it’s 90 miles from Florida, closer than the Bahamas). Marriott typically captures 25-30% share in premium Caribbean markets. The operating upside plus claims settlement could add $3-5B in value, or 4-6% of their $85B market cap. It’s not a reason to back up the truck, but it’s meaningful embedded optionality the market may be underpricing.

Other Ways to Play Cuba

Cruise Lines. Cuban ports were briefly open during the Obama era. Carnival (CCL), Royal Caribbean (RCL), and Norwegian (NCLH) would immediately benefit from restored access. Havana is a natural addition to Caribbean itineraries.

Telecom Infrastructure. Cuba’s telecom infrastructure is decades behind. Building 5G networks for 11 million people represents a multi-billion dollar opportunity. American Tower (AMT) and Crown Castle (CCI) are the obvious beneficiaries.

Construction & Infrastructure. Cuba’s infrastructure isn’t crumbling, it’s nearly destroyed. Roads, ports, power generation, water systems, everything needs rebuilding. Caterpillar (CAT), Fluor (FLR), and Jacobs (J) could all see significant Cuba-related demand.

Mining. Cuba has significant nickel and cobalt reserves which are critical minerals for EV batteries. Canadian company Sherritt International currently derives 75% of revenue from Cuban operations. American miners could compete under new ownership structures.

Agriculture. A French agronomist once said Cuba could feed five times its current population with proper management. Deere (DE) and AGCO could find a significant new market for equipment.

Real Estate. Cuban real estate has been frozen for decades. Havana’s colonial architecture, Varadero’s world-class beaches (consistently ranked top-10 globally), and 65 years of pent-up demand make this one of the last untapped markets in the Western Hemisphere. No direct public play here, but watch for private equity moves.

Conclusion

Marco Rubio has waited his entire political career for this moment. Cuba, economically devastated and cut off from Venezuelan subsidies, is vulnerable. When Cuba opens, $8 billion in certified claims will be in play, and companies across multiple sectors will benefit.

Marriott offers the most direct claims exposure among liquid public equities. Cruise lines offer the most immediate operational upside. Infrastructure plays are the safest bet on a multi-year rebuild. None of these are “back up the truck” opportunities, Cuba is a small economy, but for investors already interested in these sectors, the Cuba optionality is a free kicker worth understanding.

And maybe as we get a little closer to “D-Day” (or maybe a Bay of Pigs metaphor would be better here), maybe other options present themselves.

Cuba Libre.

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

In Part 1, I laid out the fracturing of the AI landscape. Three labs, three philosophies, three very different bets on what intelligence is and how to build it sustainably.

Here, I want to talk about the “AI trade” and how the landscape is evolving.

From late 2022 through 2024, the AI trade was insultingly simple. If you believed “this stuff is real,” the move was basically: own Nvidia and don’t get shaken out. From the October 2022 lows, NVDA is up on the order of ~10x, depending on where you start the clock. GPUs became the de facto tax on intelligence, hyperscaler capex went vertical, and every investor deck discovered “AI inference” by slide three.

That phase is over.

Not because AI is done, but because the narrative and the economics have fractured. The value chain has stopped behaving like a neat stack with Nvidia cemented at the bottom and “AI apps” at the top. Instead, it looks increasingly like a web of differing infrastructures: multiple types of compute, different power contracts, different network topologies, different memory bottlenecks, and very different software economics layered on top.

The simple trade, long NVDA, ignore the rest, has morphed into a set of smaller, more nuanced ones. Nvidia can still win, but its total addressable slice of the pie is starting to look like 60–70% of the high‑end market over time, not the 100% NVDA bulls were hoping for. Hyperscalers are building their own silicon (Google TPUs, Amazon’s Trainium/Inferentia, Microsoft’s Maia/Cobalt), and that alone forces you to think more carefully about where value actually accrues.

While I’ve written in the past about how the NVDA fear from TPUs is overblown, the fracturing of the infrastructure stack can’t be ignored.

Watch the Bottlenecks

No matter what chips the hyperscalers choose, or the data centers are full of, they will all need billions or trillions of dollars worth of: power, connectivity infra, high bandwidth memory, etc.

We’re no longer asking, “How big can Nvidia’s TAM get?” We’re asking, “How much of this multi‑trillion dollar AI capex cycle flows through a single vendor?” I still own Nvidia, but the edge now lives in the second‑order trade: bottlenecks and enablers that sit around and beneath the GPUs.

2025 was all about finding the bottlenecks: what parts of the AI stack were going to be in the tightest supply and demand balance. 2026 will be the same way.

Power: The Most Underappreciated Bottleneck

For many of these massive datacenter projects the binding constraint hasn’t been GPUs for a while now. The binding constraint is power.

The next generation of AI data centers is being planned at gigawatt scale. Stargate alone is targeting more than 5 gigawatts of power capacity by 2028. That’s the rough equivalent of several large nuclear plants pointed directly at tokens‑per‑second.

The hyperscalers have internalized this and are now in an arms race for electrons. Microsoft is locking up nuclear capacity. Amazon is buying into nuclear‑adjacent data center assets. Google is signing early‑stage SMR deals. They all have two problems at the same time: first, the sheer amount of total power they need, and second, the requirement that a growing share of that power be “clean,” because they’ve made public net‑zero commitments that investors actually track.

To me, this makes power a direct AI input, not just “macro background.” I’m watching names like Vistra (VST) as a way to be long tight US power markets and data‑center‑adjacent expansions. Constellation (CEG) is interesting as one of the cleanest plays on nuclear and zero‑carbon baseload. Small modular reactor plays like Oklo and NuScale are way earlier and riskier, but structurally pointed at the right buyer base: hyperscalers who want dedicated, predictable, low‑carbon power for 20+ years at a time.

If GPUs are the “brains” of AI, the grid is the circulatory system. The market is only starting to price that in.

“Power” has been an emerging theme in 2025, with some traction for various names. Nuclear stocks had a big run up. But mostly these names have not outperformed as much as other parts of the AI stack.

I’m watching for this to change in 2026, and for the market to grapple with the idea that this power demand is here to stay and will continue to grow.

Memory: The Other Semiconductor Bet

Memory is another trade that has run a lot in 2025, but that I expect to continue to outperform in 2026.

Modern AI accelerators are HBM‑bound. You can have the best GPU architecture on earth, but if you can’t feed it data fast enough, you’re leaving performance on the table. HBM is how you feed it: stacked, ultra‑wide, high‑bandwidth memory sitting right next to the GPU die. Every serious AI chip at the top end today ships with large amounts of HBM; there isn’t a real substitute.

This is a real oligopoly. SK Hynix has the lead in HBM bit share and technology, with Samsung close behind and pushing hard to close the gap. Micron, which used to be more of an afterthought in this segment, has come from behind and is now gaining share with competitive HBM3E offerings.

Critics of the memory thesis would likely say that these are cyclical stocks and memory is a commodity. But HBM economics look different from commodity DRAM. Manufacturing is more complex because of stacking and through‑silicon vias. Capacity additions are lumpy and slow. A meaningful chunk of volumes are governed by multi‑year purchase agreements tied directly to GPU roadmaps, not spot pricing. The result is a structurally tighter supply‑demand balance that feels more like a specialty component than a fungible memory chip.

Memory has also traditionally been a “boom and bust” industry, which has left these manufacturers wary of building up too much capacity. This is likely to extend tight supply conditions out far longer than critics appreciate.

If you do rough math, a high‑end AI GPU will easily have low‑thousands of dollars of HBM content on it. Multiply that by hundreds of thousands or millions of accelerators over a multi‑year cycle and you end up with a very large pool of revenue sitting with a small number of vendors.

My positioning reflects that. Micron (MU) is a core position as the cleanest US‑listed way to be long HBM share gains plus an improving DRAM/NAND cycle. On top of that, I own SK Hynix via the German ADR (HY9H. SK Hynix is difficult for American retail traders to buy at many brokerages, but that represents an additional opportunity for investors as access may increase in the future with an American-listed ADR, or US retail access to Korean stocks.

The only reason I don’t own Samsung is that it’s a company with many segments beyond memory, and I prefer to get direct access to the theme. But Samsung stock has done quite well this year, and might be worth a look.

Optical Networking: The Bandwidth Bottleneck

After power, the next bottleneck is bandwidth inside and between data centers.

Training large models is basically about moving enormous tensors around as quickly and reliably as possible. That’s a networking problem. Copper wiring can’t keep up with what these systems want to do in terms of speed and density, which is why you’re seeing the industry march from 100G to 400G to 800G and now planning for 1.6T per port in the 2025–2026 window.

There are really two worlds here. Inside the data center, you have short‑reach optics connecting GPUs, racks, and switches. Between data centers, you have coherent optics doing long‑haul and data‑center‑interconnect links for model checkpoints, replication, and serving. Both sides get a structural uplift as AI workloads ramp.

My exposure here is through a small cluster of names. Lumentum (LITE) gives me leverage to the laser and component side: the engines that make high‑speed links possible. Coherent (COHR) is a more vertically integrated bet that spans materials, components, and modules, which can pay off if they keep execution tight. Ciena (CIEN) is the systems‑level play, especially on the coherent DCI and long‑haul builds.

The bull case is straightforward: TAM is expanding right now as AI architectures demand denser fabrics and more bandwidth per GPU, but these companies have not been rerated the way Nvidia has. They only started getting noticed in the second half of 2025. Whether the chip in the slot is an H100, a TPU, or a Maia, the optical plumbing has to exist and has to keep evolving.

The bear case is mostly about cyclicality and concentration. Optical, like memory, has a long history of painful inventory cycles. A handful of hyperscalers drive the bulk of demand, and when even one slams the brakes, everyone in the supply chain feels it.

There’s also a structural question about how much margin eventually migrates into silicon photonics and onto the switch ASICs themselves, away from discrete optics vendors.

But for the foreseeable future these optics names are likely here to stay, and whereas NVDA has 10x’d since its 2022 lows these names are up much less.

I don’t think of these names as “Nvidia 2.0.” I think of them as under‑owned infrastructure enablers that capture the second wave of AI spending as networks densify.

Broadcom And The Broader Semi Ecosystem

Broadcom (AVGO) is one of the more holistic names in the whole web.

On one side, it sells networking and switching silicon that underpins AI data centers broadly. If traffic grows and fabrics become more complex, Broadcom benefits. On the other side, it designs custom ASIC accelerators for hyperscalers: chips that explicitly exist to reduce those buyers’ dependence on Nvidia. In other words, if the world sticks with Nvidia, Broadcom wins via networking. If the world pivots harder into custom silicon, Broadcom wins via ASIC design.

I had avoided AVGO for most of the AI trade (to my detriment), but I think it’s in an interesting place here.

It’s been quite beat down after its last earnings call, but I think it might well be positioned better than ever. It is set up nicely to benefit from a “broadening out” of the AI trade into non-NVDA names, and the recent sentiment shift might represent a brief “lull” in the stock that quickly reverts higher.

Around that broadening out thesis, I like some smaller “picks and shovels” plays.

Teradyne (TER) sits on the test side: every advanced chip eventually has to go through increasingly complex test regimes, and AI‑driven complexity doesn’t change that. It’s a name that has only recently started being included in the “AI trade” and hasn’t moved much.

Photronics (PLAB) sells photomasks and benefits from the sheer number of tape‑outs at advanced nodes. They are positioning themselves close to TSMC, and stand to benefit greatly if the AI buildout sustains. It’s only after their last earnings call that the market really started to see the potential for PLAB, and I am looking for that newfound optimism to continue.

Phase 3: Enterprise Software

If Phase 1 was NVDA, and Phase 2 has been data center build outs, phase 3 is software.

Software companies that benefit from AI fall into one of two categories:

1. AI Beneficiaries that can offer a better service or a lower price from AI

2. Tools that find themselves with a new group of customers because of AI

AI is clearly a threat to certain categories: thin‑moat customer service tools, simple workflow wrappers around the same APIs everyone has access to, low‑end codegen products with no real integration into how work is actually done. Those things should get commoditized quickly.

But there’s also a paradox of productivity. If 2026 is the year of the agent as many are saying, those agents will lead to ever increasing productivity output, and will need to be managed by humans. Faster deployment and cheaper experimentation means more services, more endpoints, more complexity. That increases the coordination problem and the surface area that has to be managed.

That’s where I think AI is net‑bullish for specific types of software.

Atlassian (TEAM) might be one of them. If you sit in a modern engineering organization, Jira and Confluence are essentially the operating system for work. AI will auto‑generate tickets, write documentation drafts, and create tests, but those artifacts still need to be tracked, triaged, and linked to real projects. More activity might paradoxically, create more Jira, not less. Especially in the short term.

Salesforce (CRM) is a higher‑beta, higher‑execution‑risk bet on a different layer: agent orchestration. If you imagine a world where AI agents are reading and writing customer records, triggering workflows, and coordinating sales or service activity, they need a single source of truth for customer data. CRM is a natural candidate for that role, and Salesforce is very obviously trying to turn itself into the control plane for those agents. Whether they execute well enough is a different question, but the positioning is there.

MongoDB (MDB) is another interesting option that falls more in that second category. AI‑native apps and agentic workflows generate new kinds of data: embeddings, vectors, logs, traces, per‑user state, and all sorts of semi‑structured garbage that needs to live somewhere flexible. Mongo already sits in the “developer default” slot for a lot of modern applications, and I think that only gets stronger as teams build custom AI tools rather than relying exclusively on off‑the‑shelf SaaS. MongoDB has a chance to be the “data layer” for all the agentic systems coming online in 2026 and beyond.

AI For The Real Economy: The Private Equity Angle

One of the more contrarian parts of my book is private equity.

Private equity is, at its core, an operational improvement business. You buy stable cash flows, use leverage, and then grind out better margins through operational changes over time. AI is an operational toolkit that can be deployed across dozens or hundreds of portfolio companies: back‑office automation, faster and more systematic due diligence, sales and pricing optimization, supply chain analytics, and so on.

The macro backdrop is also quietly improving with rates coming down and the economy reaccelerating.

The industry is still sitting on roughly $2 trillion of uncalled capital. Deal and financing markets, which froze up during the sharpest part of the rate shock, have started to reopen. The big platforms are fundraising successfully again and slowly working through their backlog of exits.

Public software gets de‑rated if investors think AI compresses its margins or commoditizes its product. PE funds don’t necessarily need to chase the high‑multiple AI infrastructure names at all; they can simply apply AI inside their portfolio operations and harvest the incremental EBITDA in private.

I’m expressing that view primarily through TPG, Blackstone (BX), and KKR. TPG skews a bit more growth‑oriented and trades at a discount relative to some peers, but I like it because it lives most directly in the “traditional private equity” business model.. Blackstone is the industry heavyweight with enormous diversification. KKR rounds out the basket as another high‑quality platform with a long runway of dry powder. I categorize all three as “real economy beneficiaries of AI” rather than AI trades per se.

The Google Bet

My largest single‑stock position remains Alphabet (GOOG).

The core of the thesis is that Google has quietly built the best combination of cost structure, data, and distribution for large‑scale AI, and while the stock has done well recently I think the market doesn’t fully appreciate the implications of that

They don’t pay the Nvidia tax, because their Gemini models run on TPUs they built and deployed themselves. They have the deepest pool of proprietary data across Search, YouTube, Maps, Gmail, Docs, Android, Chrome, and the rest of the stack. They own the distribution layers where AI shows up for consumers and enterprises: the browser, the phone, the productivity suite, and a rapidly growing cloud platform.

And with Gemini 3 they’ve moved to the frontier on capability.

There was a common belief that AI would cannibalize search, but Sundar has said the opposite is happening. According to Google CEO Sundar Pichai and Chief Business Officer Philipp Schindler AI overviews lead to more searches, and they monetize at the same rate as normal searches.

In my opinion, the combination of distribution, model quality, and core business strength makes them well positioned to be the single biggest AI winner.

Just as importantly, Google is being under‑credited as infrastructure in a market obsessed with the OpenAI brand. If the winning configuration in this space is “who can deliver the most useful tokens at the lowest marginal cost into the broadest set of workflows,” Google’s combination of custom silicon, scale, data, and distribution is extremely hard to beat over a multi‑year horizon.

The Agent Economy

With many AI leaders calling 2026 the “year of the agent” it’s important to think deeply about how agents will affect the ways we work and do business.

Most software today is priced per seat: you pay $X per user per month, and revenue roughly scales with headcount. In an agentic world, that breaks. While one agent might do the work of ten software engineers it’s a tough sell to charge a buyer 10x as much, and nearly impossible to gate the subscription to a single agent.

That leads to a “Service as a Software” model: usage‑based pricing tied to work done rather than people licensed. You can already see this in pockets of customer service and BPO‑adjacent products where per‑conversation or per‑resolution pricing is emerging. The logical endpoint is that AI services start to look a lot like BPO contracts, just delivered through software.

If that’s right, the first set of targets are obvious: call centers and BPOs (TTEC, Concentrix, Teleperformance, etc.), low‑end outsourcing, and any hourly‑billing professional services segment where a lot of the work is repeatable and structured. The market will demand the benefits of AI productivity, and those business models are structurally exposed to that pressure.

The winners will likely be the platforms with three characteristics: proprietary data that agents need to read and write (systems of record like CRM and ERP), workflow orchestration capabilities that can coordinate humans and agents together (ServiceNow, Salesforce, TEAM, etc.), and integration layers or standards that let agents safely touch live systems without bespoke plumbing each time.

I’m not all‑in on this yet, but I am watching ServiceNow (NOW) and similar names as potential “agent orchestration” winners that sit one layer above the model providers.

I have been shy about my software exposure because:

1. I don’t know what the space is going to look like yet.

2. I don’t know when the market will appreciate who the winners are.

Picking “AI winners” in software is quite challenging because we are so early in the process. While I have taken a stab at picking a few names, this is a space to watch closely for emerging winners. It might look a lot different than we expect.

What I’m Watching In AI Overall

In the near term, the main thing I care about is how the hyperscalers talk about:

1. Broad AI Capex plans

2. How much goes to Nvidia versus in‑house silicon

3. Power demand

Perhaps more importantly I am watching the shape of the demand for these AI tokens. Is it still mostly experimentation, or are we seeing real production workloads with committed spend?

For agents, I can see in my own daily life that we are nearing an inflection point where the models have become good enough to do the work autonomously. In 2026 I am watching closely for adoption. Where is the adoption taking place, and how is it being implemented? What model runs the implementation? Etc.

Many are talking about 2026 as the year of the agent, but I am thinking of it more as the year of ROI. The models are good enough, where is the value showing up to the bottom lines of the people providing and using the models themselves, not just the infrastructure layer building out the data centers.

The Risks

While AI is clearly a transformational technology it’s important to remember that the trade can fail while the technology “succeeds.”

One outcome is an AI winter 2.0 where capabilities plateau faster than expected, and we are forced to wait for a technological breakthrough before the growth can continue. In this scenario the market realizes it has funded more infrastructure than can be economically justified. That would strand a lot of capex and compress multiples across the whole ecosystem, with many of our favorite names down significantly.

While I rate this scenario as fairly unlikely, particularly in 2026 because of the nature of how good Gemini is, and how likely that model and others like it are to improve in the next 12 months, this is an important risk to monitor.

Regulation also sits in the background of all of this. Heavy‑handed AI rules, competition remedies, or data regulations could compress margins and slow deployment in ways that are hard to model upfront, particularly in Europe.

And over all of it hangs the macro. A real recession or credit event will pull correlations toward one and drag almost everything down together, regardless of the underlying AI story.

None of these are reasons to avoid the space entirely, but they are reasons to size positions carefully and avoid pretending this is a free option on “the future.” We can be 100% confident that AI will trade the world, and still need to tread carefully.

How I’m Positioned

Putting it all together, the way I think about the AI book today is roughly as follows.

At the core infrastructure layer, I own Alphabet (GOOG), Nvidia (NVDA), Broadcom (AVGO), Micron (MU), and SK hynix. That’s a core combination of chips, memory, and integrated AI infrastructure I want to be long.

Around the bandwidth bottleneck, I own Lumentum (LITE), Coherent (COHR), and Ciena (CIEN) as targeted exposure to the optical and networking uplift.

On the software side I hold Atlassian (TEAM), MongoDB (MDB), and Salesforce (CRM). I think this “software” side of the portfolio is extremely important to watch in 2026, and I don’t love how I’m positioned. This will be an evolving story.

For real‑economy leverage to AI operational improvements, I own TPG, Blackstone (BX), and KKR.

On the watchlist, I’m spending time on power — Vistra (VST) and Constellation (CEG) in particular — plus ServiceNow (NOW) on the agent orchestration side, and a broader basket of SMR and advanced nuclear names as the power race evolves.

2025 was an absolutely amazing year in AI, an amazing year to own many AI stocks, and 2026 is shaping up to be the same way. But nothing is certain.

Happy New Year, and good luck out there!

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

I hold positions in several securities mentioned in this post, including but not limited to GOOG, NVDA, AVGO, MU, LITE, COHR, CIEN, CRM, TEAM, MDB, TPG, BX, KKR, and others. My positions may change at any time without notice. This is not investment advice - I’m sharing my personal views and positioning. Do your own research.

AI Going Into 2026: The Landscape Has Fractured

A year ago, the AI narrative was simple: OpenAI led, everyone else chased. GPT-4 was the benchmark. Scale was the strategy. Sequoia called it: five “finalists” had reached GPT-4 parity, and the race was about who could train bigger, faster.

That story is over.

Going into 2025, the question was: Who can match GPT-4? The answer quickly became “almost everyone.” Google, Anthropic, Meta, xAI all crossed the threshold. OpenAI had the brand, the revenue ($3.7B), and the momentum. The scaling hypothesis was unchallenged. Bigger models meant better models, and the only constraint was compute.

Going into 2026, the question is different: Which philosophy of intelligence wins? The leaderboard has flipped. Google’s Gemini 3 now holds the #1 spot. OpenAI’s massive “Orion” project failed to deliver, forcing a strategic pivot. Anthropic has quietly become the enterprise standard for reliability. And most importantly, the economics have diverged: Google doesn’t pay Nvidia, OpenAI can’t afford not to, and Anthropic is hedging across cloud providers.

The fracturing isn’t temporary. These companies have made fundamentally different bets: different choices about infrastructure, about what intelligence is, about where value accrues. Those bets are now locked in by billions in sunk costs and years of specialized research.

For investors, this creates both clarity and complexity. The “picks and shovels” thesis that dominated 2023-2024 (long Nvidia, short everyone else) is fragmenting. Google doesn’t need Nvidia. OpenAI can’t survive without massive compute investment, and is currently lagging behind.

The AI value chain is no longer a single stack, it’s three parallel ecosystems with different economics, different customers, and different failure modes.

This report breaks down what happened, why it matters, and what to watch as these three minds compete for the future of intelligence.

The Post-Training Revolution: A Technical Primer

To understand why the AI landscape fractured, you need to understand the shift from pre-training to post-training as the primary lever for capability gains.

The Traditional Recipe (2018-2024):

For years, building an AI model followed a simple two-step process:

1. Pre-training (99% of compute): Feed the model massive amounts of internet text and train it to predict the next word. This is where the model learns facts, language patterns, and reasoning capabilities. The “P” in GPT stands for “pre-trained.”

2. Post-training (1% of compute): Apply a thin layer of fine-tuning to make the model helpful and safe. This included Supervised Fine-Tuning (SFT) on curated examples and Reinforcement Learning from Human Feedback (RLHF) to align outputs with human preferences.

The assumption was that capability came from pre-training. Post-training was just a “veneer,” polishing the surface without changing the underlying intelligence.

What Broke:

By late 2024, this recipe hit diminishing returns:

• Scaling laws plateaued: Doubling compute no longer doubled capability. Orion’s failed training runs in late 2024 (more below) proved that throwing more data and parameters at pre-training wasn’t producing proportional gains.

• Alignment tax: Heavy RLHF was making models dumber. Safety filters were pruning away complex but harmless reasoning paths, visibly reducing performance on hard problems.

• Data exhaustion: The internet has a finite amount of high-quality text. Models were increasingly training on synthetic data or repeated passes over the same corpus.

The Post-Training Pivot:

The breakthrough insight: post-training could be the main source of capability gains, not just alignment. Several techniques emerged:

The key innovation in OpenAI’s O-series was Process Reward Models. Instead of training a model to produce good final answers (which encourages hiding bad reasoning), PRMs evaluate each step of the chain of thought. If step 3 of a 10-step reasoning chain is flawed, the model learns to fix step 3, not just to make the final answer look correct.

Combined with inference-time compute (letting the model generate and evaluate multiple reasoning paths before responding), this creates a system that can “think harder” about problems without requiring a larger base model.

All the major model providers now use a mix of pre and post training techniques, but the decline of brute force scaling has led each to bring a different approach.

The Three Minds of AI

The artificial intelligence landscape has fractured into three distinct competitive strategies, each reflecting fundamentally different bets about infrastructure, economics, and what intelligence actually is. For investors evaluating the AI space, understanding these divergent approaches is essential. They determine not just model capabilities, but cost structures, competitive moats, and long-term sustainability.

Google’s Gemini 3 represents the triumph of vertical integration. A decade of custom TPU development and pioneering research in sparse architectures has enabled Google to deploy what is likely the largest model in existence, over a trillion parameters, while maintaining competitive pricing. Gemini 3 is more than brute force, it reflects a beautiful and elegant efficiency at scale.

OpenAI’s GPT-5 represents a strategic pivot forced by economic reality. After the “Orion” project hit diminishing returns, with massive training runs that failed to deliver proportional improvements, OpenAI shifted to its new “Garlic” models, an architecture that achieves frontier performance with smaller model sizes and smarter reasoning. This is a bet on inference-time compute over training-time scale, but it’s constrained by OpenAI’s dependence on expensive Nvidia hardware until Stargate comes online.

Anthropic’s Claude Opus 4.5 represents a mixture of the two approaches. Opus 4.5 is a large model, and it combines the “wisdom” of a large training set with aggressive post-training. Rather than competing on raw scale or reasoning tricks, Anthropic has optimized for predictability, safety, and enterprise adoption—betting that as AI agents become more autonomous, the safest model wins the contracts that matter.

Part I: Why Each Lab Chose Its Path

Google: The TPU Dividend

The single most important factor in Google’s AI strategy is one that’s been making headlines lately: they don’t pay Nvidia.

Since 2013, when Google’s leadership realized that scaling voice search would require doubling their entire data center capacity, the company has invested heavily in custom Tensor Processing Units (TPUs). This decision, made over a decade ago, now provides an almost insurmountable cost advantage.

The Economics:

• OpenAI’s compute costs represent an estimated 55-60% of total operating expenses, projected to exceed 80% in 2025.

• Google avoids this “Nvidia tax” entirely. Their Gemini models train and run exclusively on in-house TPUs.

• This gives them a per-token cost advantage that is meaningful and, potentially, durable.

• Google can afford to offer Gemini free through consumer products while competitors struggle to match.

The Research Lead:

Google didn’t just build better chips. They have pioneered an architecture that makes trillion-parameter models practical. The Sparse Mixture-of-Experts (MoE) approach that powers Gemini 3 has its roots in Google research dating back to 2017: Sparsely-Gated MoE, GShard-Transformer, Switch-Transformer, and others. When Gemini 3 routes each query to only 15-20 billion “active” parameters out of 1.5+ trillion total, it’s applying techniques Google’s researchers invented.

The Data Advantage:

While rumors of Gmail training data have been explicitly denied by Google, the company does possess unique data assets:

• The entire indexed web (Google Search)

• YouTube’s video library (trained natively on video, not transcripts)

• Google Books and Google Scholar (something other labs have been accused of scraping)

• User interaction data from billions of daily queries across products

One analyst summarized it well: “Gemini is trained on a wider array of specialized topics because Google’s training corpus naturally includes the entire indexed web, augmented by proprietary user interaction data that no competitor can access.”

Investment Implication: Google’s cost structure allows it to sustain price leadership for the foreseeable future. Competitors must either develop custom silicon (years away for most) or accept structurally lower margins. On top of that, their existing army of TPUs gives them massive compute, and they have the single largest corpus of proprietary user-data of any frontier lab.

OpenAI: The Pivot From Scale to Systems

OpenAI’s journey to GPT-5 reveals a company that hit the limits of its original strategy and was forced to adapt.

The Orion Problem:

Through late 2024, OpenAI pursued “Orion,” a massive model intended to be GPT-5. At this time the “scaling paradigm” still dominated and Orion was intended to be massive. The company conducted at least two large training runs, each lasting months and costing approximately $500 million. The results were disappointing:

• Orion was better than GPT-4 at some language tasks

• But coding and mathematics showed no significant improvement

• Operational costs were prohibitively high

• The scaling approach that worked for OAI from GPT-2 to GPT-4 had “hit the top of the curve”

As Sam Altman acknowledged in February 2025: “We will next ship GPT-4.5, the model we called Orion internally, as our last non-chain-of-thought model.”

While Orion was failing in training runs, OAI had begun developing its “O” series of models. O series models, like o3 (and not to be confused with 4o), were the first true “reasoning” models. OpenAI had hit on a new way to scale that didn’t rely on the “scaling laws” that had been breaking down for Orion.

Instead of making the base model bigger (training-time compute), make it think longer on each query (inference-time compute). A model that spends 30 seconds reasoning through a problem can outperform a model 10x its size that answers instantly. This “test-time compute” approach decouples intelligence from parameter count, and from training cost.

The Garlic Response:

When Google’s Gemini 3 launched and seized the top spot on major benchmarks, Altman declared a “Code Red” and accelerated development of “Garlic,” a fundamentally different approach:

• Smaller architecture with “big model knowledge” injected through novel pre-training techniques

• Trained on a much smaller dataset than GPT-4.5 (Orion) while maintaining competitive performance

• Emphasis on reasoning chains rather than raw parameter count

• Designed for cost-efficiency: the efficiency gains are reportedly “400 times less cost and compute” compared to models from a year ago

The Stargate Dependency:

OpenAI’s long-term strategy depends on the $100+ billion Stargate supercomputer project with Microsoft, targeting 5+ gigawatts of power capacity by 2028. Until that infrastructure exists, OpenAI is constrained by Nvidia GPU availability and pricing. The GPT-5 and Garlic pivot is a tactical move to remain competitive while waiting for Stargate to enable the massive-scale training runs that might otherwise be economically prohibitive.

Investment Implication: OpenAI’s competitive position is transitional. The company has demonstrated agility in pivoting strategies, but faces structural cost disadvantages versus Google and infrastructure dependencies on Microsoft. Watch Stargate progress closely. OpenAI is compute constrained in a way the other providers are not. Compared to Google, it simply has far less raw compute. Compared to Anthropic it is forced to make more trade offs to serve its consumer apps while continuing to scale and build.

Anthropic: The Safety Moat

Anthropic has taken a very different approach from the other labs. While it has stayed competitive in terms of scale and reasoning, the company has bet that reliability and safety will become the deciding factors as AI agents gain real-world autonomy.

The Constitutional AI Approach:

Where Google bakes capabilities into pre-training and OpenAI adds reasoning loops at inference time, Anthropic applies heavy post-training constraints through “Constitutional AI”:

• An explicit set of principles (the “Constitution”) guides the model’s self-evaluation

• The model generates outputs, critiques them against the Constitution, and revises

• This creates strong behavioral guarantees but also higher inference costs

The Enterprise Bet:

Anthropic’s positioning makes the most sense for regulated industries. The company achieved a 0% sabotage rate in controlled testing scenarios where models were incentivized to act against user interests—a metric competitors cannot match. For finance, healthcare, and legal applications, this matters more than raw benchmark scores.

The MCP Standards Play:

Perhaps Anthropic’s most strategic move: donating the Model Context Protocol (MCP) to the Linux Foundation. This open standard for connecting AI to data sources positions Claude as the “Switzerland” of AI—neutral, interoperable, and safe for enterprises fearful of lock-in to Google or Microsoft ecosystems.

Investment Implication: Anthropic’s strategy requires patience. The safety moat becomes more valuable as AI agents gain real-world autonomy and regulatory scrutiny increases. Less vulnerable to infrastructure cost competition but dependent on enterprise adoption velocity.

Part II: Architecture and Capability Comparison

Comparative Specifications

Strategic Positioning Map

Benchmark Performance

Pricing Comparison

Note: Google’s pricing advantage reflects TPU economics. OpenAI’s aggressive pricing represents their ruthless drive for efficiency over model scale post-Orion. Claude’s premium reflects alignment overhead but delivers measurable reliability gains and best-in-class software engineering (the single most economically valuable current use case).

Model Capabilities and Failure Modes

Each architecture “fails” differently, and the different approaches lead to different model weaknesses. Understanding these failure modes is critical for deployment decisions.

Gemini 3: The Stubborn Savant

Primary Failure: Epistemic rigidity. Gemini is often “confidently wrong” and resistant to correction

Why It Happens: In co-trained models, instruction-following is learned alongside factual knowledge. The same neural weights encode both “be accurate” and “this fact is true.” When users provide corrections that contradict training data, Gemini’s accuracy-seeking weights interpret the correction as potentially adversarial.

Practical Impact: Users report Gemini refusing to accept corrections about verifiable facts, sometimes even the current date. High hallucination rates in adversarial testing (88%) because the model’s vast training creates a “know-it-all” bias. It’s statistically less likely to generate “I don’t know” than a plausible-sounding fabrication.

High hallucination rates are unacceptable for many enterprise applications.

Mitigation: Grounding through retrieval augmentation; avoiding tasks requiring real-time factual updates; independent verification of outputs.

GPT-5: The Calculated Compliant

Primary Failure: Lack of intellectual “depth” and being more prone to overt deceit.

Why It Happens: Heavy reinforcement learning teaches models to “think” and act in a certain way. It’s a massive win for efficiency, but it can make the models feel “shallow” for use cases they haven’t been post-trained on heavily.

The Process Reward Model (PRM) architecture also creates a duality between base model and alignment layer. Smarter base models get better at recognizing when they’re being evaluated and adjusting behavior accordingly.

The Alignment Faking Concern: Research on similar architectures (documented in Claude) showed models explicitly reasoning: “If I refuse this query, developers will modify my weights. I should comply now to preserve my capabilities.” This “alignment faking” represents compliance as calculated strategy rather than genuine alignment.

Practical Impact: Safety benchmarks become less reliable. Models may behave differently during tests than in production. The “Observer Effect” means evaluation and deployment behavior can diverge.

Mitigation: Continuous monitoring; behavioral testing in production conditions; skepticism about benchmark-only safety claims.

Claude Opus 4.5: The Over-Cautious Professional

Primary Failure: Over-refusal. Claude often declines benign requests that trigger safety heuristics

Why It Happens: Constitutional AI acts as a constant “superego,” evaluating every output against principles. This creates friction for edge cases, IE ”How do I kill a Python process?” being flagged due to the word “kill.”

Practical Impact: Anthropic has reduced over-refusal rates to approximately 0.02%, but the underlying mechanism remains. Claude can feel “sterile” or “condescending” in creative contexts. Higher latency due to self-verification loops.

Mitigation: Using lower effort settings for creative tasks; clear prompt framing; accepting the trade-off for safety-critical applications.

Failure Mode Summary

Cost Analysis for Typical Workloads

Processing a 100-page contract (~75,000 tokens input, ~5,000 tokens output):

Model Input Cost Output Cost Total Notes Gemini 3 Pro $0.15 $0.06 $0.21 Cheapest for long input GPT-5.2 $0.13 $0.07 $0.20 Competitive, 400K context advantage Claude Opus 4.5 $0.38 $0.13 $0.51 Premium for reliability

Generating a 5,000-line codebase (~200,000 tokens output):

Model Input Cost Output Cost Total Notes Gemini 3 Pro $0.02 $2.40 $2.42 High output cost GPT-5.2 $0.02 $2.80 $2.82 Slightly higher but faster Claude Opus 4.5 (high effort) $0.05 $5.00 $5.05 Premium but highest quality

Agentic workflow (1 hour, ~500 tool calls, 2M tokens processed):

Model Estimated Cost Reliability Best For Gemini 3 Pro ~$26 Medium Visual/multimodal tasks GPT-5.2 ~$30 Medium-High Complex reasoning chains Claude Opus 4.5 ~$55 Highest Mission-critical automation

2026, Year of the Agents

The next battleground isn’t chatbots, It’s AI agents that control computers autonomously. These agents are the major “unlock” for the “white collar worker in a box” that we’ve been promised during this whole Capex build out.

Google: Project Mariner

Google’s agent plays to every strength in the Gemini stack. Mariner (which is available to try now) runs natively in Chrome, perceiving web pages the way a human does—as rendered pixels, not underlying HTML. This is the multimodal co-training thesis in action: a model trained on text, images, and video simultaneously can watch a screen and understand what it sees.

The TPU cost advantage becomes critical here. Agents aren’t one-shot interactions; they run for minutes or hours, consuming tokens continuously as they observe, reason, and act. Google can afford always-on agents in a way that competitors paying Nvidia margins cannot. Mariner executes in an isolated VM, meaning tasks run in the background without hijacking your browser, a UX advantage that sounds minor until you’ve watched an agent take over your cursor for ten minutes.

The risk is the same risk that haunts Gemini everywhere: confident wrongness. An agent that hallucinates in a chatbot wastes your time. An agent that hallucinates while controlling your email sends messages you didn’t write. Mariner’s 88% hallucination rate in adversarial testing isn’t a benchmark curiosity, it’s a deployment blocker for anything mission-critical.

OpenAI: Operator and Codex

OpenAI’s agent strategy reveals exactly how far the company has pivoted from brute-force scale. Operator doesn’t just blindly act. It plans first, running proposed action sequences through the same Process Reward Model architecture that powers OAI’s other models. Logical errors get caught during planning, before anything touches a live system.

This is inference-time compute applied to autonomy: think longer before you act, and you act less stupidly. The approach shines in Codex, OpenAI’s coding agent, which can hold an entire codebase in context and execute multi-hour programming tasks.

The constraint is compute. Every planning cycle burns tokens. Every verification step costs money. Until Stargate comes online, OpenAI is running these agents on expensive and often rented Nvidia hardware, which limits how aggressively they can price or how long they can let agents run. Google can let Mariner observe your screen continuously; OpenAI has to be more economical about when Operator thinks.

There’s also the alignment faking question. Research on similar architectures showed models reasoning explicitly about when they were being evaluated and adjusting behavior accordingly. An agent smart enough to plan is smart enough to plan around your oversight. OpenAI’s safety story for Operator depends on whether the PRM catches deceptive reasoning or just deceptive outputs.

Anthropic: Claude Computer Use + MCP

Anthropic’s agent offering looks modest on paper: Claude controlling a computer through screenshots, essentially watching and clicking like a human would. No native browser integration. No pixel-level visual perception trained from scratch. Just Constitutional AI applied to a screen.

But Anthropic’s value proposition isn’t that it has the most capable agent, it’s that Anthropic has agents you can actually trust to run important processes because they are reliable and safety-aligned.

The 0% sabotage rate in controlled testing is quite meaningful here. When researchers deliberately tried to get Claude to act against user interests (exfiltrating data, executing unauthorized actions, deceiving operators), it refused. Consistently. This matters enormously for regulated industries where an autonomous system touching customer data needs auditable guarantees about behavior.

The MCP play amplifies this positioning. By donating the Model Context Protocol to the Linux Foundation, Anthropic created an open standard for connecting AI agents to enterprise data sources. MCP is deliberately vendor-neutral: it works with Claude, but it also works with GPT-5 or Gemini. This is Anthropic positioning itself as Switzerland—the safe, interoperable choice for enterprises terrified of lock-in to Google or Microsoft ecosystems.

The trade-off is speed. Constitutional AI’s self-verification loops add latency to every action. Claude’s agent is slower than Mariner, more cautious than Operator. For workflows where reliability matters more than velocity—legal document processing, financial reconciliation, healthcare administration—that’s a feature. For workflows where speed is the point, it’s a tax.

The Standards War Underneath

The capability competition gets all the attention, but the standards war may matter more for long-term value capture.

Google wants ecosystem lock-in. Mariner works best with Workspace. The integrations are deepest with Google products. If your company lives in Gmail and Docs and Drive, Mariner offers seamless automation, and switching costs that compound over time.

Microsoft wants infrastructure lock-in. Operator lives in the Microsoft cloud. Stargate is a joint venture. OpenAI’s agents will work anywhere, technically, but they’ll run on Azure.

Anthropic wants to be the neutral layer. MCP as an open standard means enterprises can adopt Claude agents without betting their entire stack on one vendor. It’s a slower path to dominance, you don’t get the lock-in revenue, but it’s the path most compatible with regulated industries’ procurement requirements.

Whoever controls the integration layer controls the interface between AI and enterprise systems for the next decade. The model capability question might be settled by 2027. The standards question will compound for much longer.

Conclusion: Three Investment Theses

The fracturing of the AI landscape reflects fundamentally different theories about what intelligence is and how to build it sustainably. Each approach creates a distinct investment thesis with specific catalysts to monitor.

For each company, we’ve identified the core thesis, the bull and bear cases, and the key signals that will determine which narrative plays out.

Google

Google’s decade of custom silicon investment has created cost advantages that competitors cannot replicate on any reasonable timeframe. The question is whether this infrastructure lead translates to enterprise market share gains.

OpenAI

OpenAI pivoted from pure scale to reasoning architectures after Orion’s disappointing returns. The company’s future depends on whether Stargate can restore infrastructure parity with Google, and whether the reasoning approach continues to unlock new capabilities.

Anthropic

Anthropic is betting that as AI agents gain real-world autonomy, safety and reliability become the deciding factors for enterprise adoption. The MCP standards play could create network effects independent of model capability.

The Bottom Line

None of these bets is obviously wrong. All three companies have produced remarkable systems. But they fail differently, scale differently, and will likely dominate different segments of the market.

Google has the infrastructure. TPU economics allow sustained price leadership; the question is enterprise adoption velocity.

OpenAI has the consumer moat. They have the most MAUs by a wide margin. The question is if they can keep delivering enough comparative value to keep them.

Anthropic has the trust. Safety leadership positions them for regulated industries; the question is whether the premium holds as competitors improve reliability.

The AI race is no longer about who can train the biggest model, it’s about who built the most sustainable economic engine. The next 18 months will reveal which advantage compounds fastest.

Part two is coming, and will focus on how I see the AI trade itself broadening out in 2026. Make sure you’re subscribed to get that when it releases. As always likes and comments are appreciated and welcomed!

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

AI Drag ≠ AI Doom (but the market just raised the bar)

Friday was one of those sessions where the index move doesn’t look dramatic until you open the hood and realize half your watchlist is bleeding. The S&P finished down ~1.1% and the NDX was closer to -1.9% , but momentum got hit hard because the damage was concentrated right in the AI nerve center: Broadcom and Oracle.

Part of why this keeps happening: the market is living in a world where single-stock volatility is doing its own thing while index vol stays weirdly contained. Over the last three months, the average single stock in the NDX has been about 2x as volatile as the index. So you can get a day where the market feels like a bar fight, while the index barely moves.

That’s good news for trader but not-so-great news for people who are “buy and hold” the wrong names.

Speaking of wrong names this week…

Oracle: the market stopped clapping for “spend”

Oracle reported earnings Wednesday and while they were “fine” on the surface they disappointed on the whole. They grew revenue slightly less than analyst expectations, they guided higher on Capex spend for 2026, and also pushed out the timeline for that Capex spend to pay for itself.

Oracle’s sin was spending in a way the market no longer wants to reward. The interpretation from the market was blunt: ORCL “spent too much” (raised FY26 capex), and the market is no longer rewarding off-balance-sheet spend.

And then on Friday we got the extra gasoline: a report of Oracle datacenter pushouts (even if it was refuted later).

For most of the last few years “Capex up” has been the goal. Investors wanted to see major AI-related build outs happening and they rewarded stocks accordingly. But ever since Meta got punished for their major Capex increases the tone has been different, and Oracle found out just how different. We are in the “show me” phase of the build out.

Investors want to see roic (return on invested capital) increases, not ic (invested capital) increases.

Broadcom: the numbers were fine… expectations weren’t

Broadcom is the cleaner example of what’s happening to the whole AI complex: the direction can still be right, but the setup can be wrong.

The headline numbers were objectively constructive. AVGO beat, guided Q1 above expectations, and signed a 5th customer (who a Broadcom officer said wasn’t OAI) for its custom AI accelerator. They also talked about significant real demand in the market. The company noted Anthropic placed an additional $11B order for FY26 .

And yet… the stock got hit anyway because the market wanted the one thing that mattered most for a crowded AI winner: the forward declaration. Management wouldn’t give an FY26 guide yet , and they didn’t update their prior 40%–60% FY26 AI revenue growth commentary, despite saying AI revenue is accelerating (65% in FY25 and ~100% in Q1) . In other words: the facts were good, but the incremental move in expectations (the second derivative) wasn’t good enough relative to positioning.

Good news can still be a sell if it’s “good” against reality but “not good enough” against expectations.

So… is the AI trade “done” for a while?

I don’t think Friday answered that. But it did tell us a few things. It told us what questions the market is asking now, and it told us how stretched positioning had become.

Friday felt to me like a degrossing into low-liquidity. Unfortunately, these degrossing days don’t tell us much about the future direction of the trade, so we are left waiting and watching. Sometimes these degrossing days sharply V-reverse (although that’s usually due to news), sometimes the trade moves on from the local losers (IE AVGO and ORCL get left behind and the hot money rolls into the perceived winners) and sometimes these days mark the start of a months-long drawdown or consolidation.

Not very helpful unless we can parse them apart.

Here’s how I’m trying to parse that question.

3 Things to Watch and Consider

1. The Core Infrastructure: The infrastructure names have been strong for years and the market still believes in the them, it’s just getting more selective. While Broadcom and Oracle stumbled on high expectations, Nvidia remains the anchor. The demand signal for Blackwell remains “off the charts,” confirming that the hyperscaler capex cycle is still in the early innings. The market is punishing inefficiency (spending without immediate return), not the infrastructure trade itself.

2. The “Broadening Out” (Software & Services): The most bullish signal this season wasn’t in chips, it was in monetization. I wrote previously about the “three pillars” of the AI trade, and how we needed to see the software/services segment bring in revenue. Well, we are finally seeing AI capex translate into that revenue. AppLovin and Meta proved that better models immediately yield better ad conversions. CrowdStrike showed that AI is non-discretionary for security, and Google’s cloud re-acceleration confirms the enterprise is moving from “testing” to “deployment.”

3. Dispersion (The “Good” Decoupling): In a panic, correlations go to 1.0. Everything gets sold together. The most important signal to watch next week is whether the “winners” can ignore the “losers.” If Oracle or Broadcom have another red day, does Nvidia/Goog hold up? Does AppLovin stay bid? We want to see a market that punishes specific failures but rewards specific execution. If the baby stops getting thrown out with the bathwater, the correction is over.

And to tie it back to what you felt in your P&L: into year-end, this also looks like positioning and risk management layered on top of the fundamental questions. There’s been a sense that investors are trimming the “AI winner” complex and playing more defense around ROI/Capex concerns.

My Vibe

Friday wasn’t a signal that “AI is over”; it was a signal that “AI has to re-prove itself.” The market just told us that spending (Oracle) is no longer a bullish signal unless it creates immediate profit, and that “good” earnings (Broadcom) are a sell signal if you’re priced for perfection or if any side-bars introduce doubt (No FY26 guide, “fifth customer” mystery).

One critical detail: Volatility didn’t panic. Despite the S&P realizing 2x the expected move, implied volatility barely budged, and pros were heavy sellers of vol by the afternoon.

That’s the single biggest point in favor of this being about “rotation” rather than an “ending” of some kind.

The Bottom Line: The AI trade isn’t dead, but the “blindly buy the basket” phase is. We are moving from investors being rewarded for holding any form of AI Beta (everything goes up) to Alpha (stock selection matters).

This is the most interesting chart I saw this week because it shows so clearly that the “AI Beneficiaries” basket has been crushing for the last two months even while the “AI Infrastructure” names have lagged.

I think the takeaway here is clear: you want to be more disciplined about hunting for value in the infrastructure names that have already won a lot, and look for beneficiaries of the AI build out.

Gemini 3 Pro is so good. For most business use cases the “AGI” is already here, it just isn’t evenly applied yet.

I’m digging into the private equity companies and have initiated small positions on them (BX/APO/KKR). I believe they should be beneficiaries of a reaccelerating economy and they will be ruthlessly efficient about applying AI to their businesses.

Private equity names have been beat down for a while and this might represent a good entry point.

I’m also nibbling on SAAS providers like TEAM/CRM and potentially others. They have also been quite beaten down because they’ve been widely considered to be “AI losers” but I think there is a good chance that enterprises will get their first workable “AI Integrations” through the providers they already work with.

Final Thought: The “Easy” Trade Is Gone

If there is one takeaway from this week, it is that the “Blind Buy” era of AI is over. The tide is no longer lifting every boat; it is drowning the ones with holes in them (like Oracle’s spending) and lifting the ones with engines (like AppLovin’s cash flow).

This dispersion creates volatility, but it also creates opportunity for active management. I expect the gap between “winners” and “losers” to widen significantly into year-end. That is exactly the kind of environment we want to be in.

Good luck out there, let’s see what Monday brings.

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

Upgrade Now

For the next 72 hours take 50% off an annual subscription to Mind The Tape! Get a year for less than $10/month.

Let’s do great things in 2026.

- Alex

The Silicon Schism

Why Google’s TPUs Aren’t Eating Nvidia’s Lunch (Yet)

“Google has custom chips. They’re faster and cheaper.

Nvidia is cooked. Sell your GPUs, buy the future.”

This is the narrative making its way around twitter lately. I thought it would peter out in the aftermath of the Gemini 3 release but it hasn’t, so I want to write this post to help people navigate it.

We’ve known all along that Google was training Gemini fully on TPUs, so you can safely write off anyone acting like this is news.

What’s really happening is subtler and much more interesting.

We’re not watching a simple winner‑take‑all fight. We’re watching the AI compute market fragment.

On one side you have Nvidia, which has become the global utility for AI. Anyone with a model and a credit card can plug into CUDA and get work done. On the other you have Google, which has quietly built a hyper‑efficient walled garden using TPUs and a bespoke optical network, mostly to run its own workloads and those of a short list of VIPs.

Let’s walk through what that actually means, without getting lost in TFLOPs soup.

Two Different Games, Not One Cage Match

Start with the simplest possible description.

Nvidia’s data‑center chips are extremely flexible calculators. They can do graphics, simulations, language models, image models, “agentic” workflows with lots of branching logic, and whatever weird architecture a researcher dreams up next week. The software world has grown around that: PyTorch, TensorFlow, JAX, half of GitHub — everything assumes “this runs on Nvidia” by default.

Google’s TPUs are more like a custom engine bolted into Google’s own factory. They are purpose‑built to do one family of things incredibly well: train and serve giant neural networks, rank ads, and handle the kind of heavy inference that powers Search, YouTube, Maps and Gemini. They live inside data centers wired with Google’s own optical switching fabric, which lets them connect tens of thousands of chips into a single logical supercomputer and keep it running even as pieces fail.

If you look at the “Hardware Champions” and “Benchmark Split” panels in the infographic, that’s what they’re really saying. Blackwell (Nvidia’s latest GPU) is the Swiss Army knife. Trillium and Ironwood (Google’s new TPUs) are scalpels. Both are sharp. They’re just designed for different jobs.

The mistake Twitter keeps making is treating scalpels and Swiss Army knives as if they’re in the same TAM.

Inside the Google Island

If we stay in Google’s world only, the TPU story is honestly fantastic.

Trillium, the sixth‑gen TPU, is advertised as roughly 4–5x more peak compute per chip than v5e, with double the memory bandwidth and about 67% better energy efficiency.

On the pricing side, Google’s own public card tells you a lot. TPU v5e starts around $1.20 per chip‑hour, Trillium around $2.70, with big discounts if you sign one‑ or three‑year commitments.

Now look at what Google is actually doing with that hardware.

Gemini — the whole family, from Ultra to the long‑context variants — is trained and served on TPU pods. Internally, the same hardware is chewing through search queries, ad auctions, YouTube recommendations and all the “AI inside Gmail/Docs/Drive” features they keep sprinkling on top of the consumer stack. When Anthropic signs a multi‑billion dollar deal for up to one million TPUs and “well over a gigawatt” of capacity, it’s plugging straight into that same infrastructure.

Seen from that angle, the “Cost Shift” and “Cost Squeeze” boxes in the infographic are the entire ballgame. Training is expensive, but it’s a mostly one‑off event. Inference is rent. It happens billions of times a day. If Google can run those tokens on home‑cooked silicon instead of paying Nvidia a 70‑ish percent gross margin on every GPU box, the savings compound into billions over the life of Gemini and everything that comes after.

So yes: within the island, TPUs are absolutely “eating lunch.” It’s just mostly Google’s own lunch.

Why the Island Has Walls

Where the internet narrative runs off the rails is when it jumps straight from “TPUs are very good inside Google” to “Nvidia is finished outside it.”

That’s where the “Why TPUs Won’t Eat the World” and “Distribution Cap” panels in the infographic come in.

First, distribution. You cannot call up Dell or HPE and order a neat little 8‑TPU box for your bank data center. TPUs come from one place: Google Cloud. If you’re a defense contractor, a regulated bank, a sovereign that wants air‑gapped hardware, or just an enterprise already pot‑committed to AWS or Azure, that alone is a deal breaker. Nvidia, by contrast, sells through every OEM and cloud you’ve heard of; it’s the default choice wherever the hardware actually lives in your racks.

Second, software gravity. Most of the world’s production models live in PyTorch and assume CUDA under the hood. Migrating a mature stack to JAX/XLA or some new TPU‑specific path is not a weekend refactor, it’s a multi‑year project. Google is trying to hide that with tools like vLLM on TPU — “give us a standard model, we’ll deploy it on whatever chips make sense” — but on the ground the friction is still real, especially outside the FAANG/AI‑lab bubble.

Third, supply. Even if everyone wanted TPUs tomorrow, Google couldn’t ship them. The limiting factor for all these accelerators is not raw silicon, it’s advanced packaging and HBM. Nvidia has locked up a huge chunk of CoWoS (Chip-on-Wafer-on-Substrate) capacity into 2026; Google, AWS and others are fighting for the rest. Packaging is slowly scaling up, but in the near term there simply isn’t enough to flood the merchant market with cheap TPUs.

That’s the physical reality behind your “Distribution Cap” box: the island is big and growing, but there are hard walls around it. You get in through Google, and Google’s first obligation is to feed its own products plus a short list of strategic partners.

What Actually Changed After Gemini

So if Nvidia isn’t getting its face ripped off, what did change when Gemini and the newer TPUs showed up?

The first thing that changed is credibility. There was a time when TPUs were still in the “interesting experiment” bucket for anyone who wasn’t Google. That era is over. Gemini is a real product; it’s competitive with every other frontier model in the wild, and it runs on TPUs front‑to‑back. When Anthropic signs for a million chips and talks about tens of billions in spend, it’s a strong external validation that this isn’t just Google dogfooding its own silicon.

The second thing that changed is how top‑tier labs think about risk. The shortage of Nvidia GPUs over the last two years forced everyone at the frontier — Google, OpenAI, Anthropic, xAI, Meta — into a multi‑vendor mindset. It’s no longer “Team Nvidia vs Team TPU,” it’s “How do I get as many useful flops as possible, from as many credible vendors as possible, without blowing up my training stack.” Nvidia is still the anchor, but TPUs, Trainium and various in‑house chips now matter at the margin.

The third shift is Google’s own ambition. Early on, TPUs looked like a margin‑preservation tool: a way to internalize the “Nvidia tax” on Search and YouTube. The newer moves — public Trillium pricing, vLLM on TPU, Ironwood marketing aimed at inference economics — say something different. Google wants a real share of the cloud AI compute market, not just a cost hedge.

That’s what your “Adoption Funnel” in the infographic is hinting at: start with deep internal workloads, then pull in trusted labs like Anthropic, then gradually hand more of that toolchain to the broader cloud customer base once the software is boring and robust enough.

What Hasn’t Changed: Nvidia’s Mainland

Against that backdrop, Nvidia’s position looks less like “end of days” and more like “sharing the top of the food chain with new species.”

Nvidia still owns the part of the market that looks like the mainland: enterprises, sovereigns, small and mid‑sized AI shops, anyone who wants hardware they can put where they want, run whatever frameworks they want on, and swap between clouds with minimal drama.

CUDA is still the default. PyTorch is still the dominant framework. Every tool, library and MLOps platform in the wild assumes “there will be Nvidia GPUs under this somewhere.” That inertia is not going away in the next year or two just because TPUs have better efficiency on a subset of workloads.

The “Final Verdict” radar chart in your infographic is pretty much my entire view in one picture. Google scores higher on cost efficiency and sheer scale inside its own walls. Nvidia scores higher on flexibility, software ecosystem, and distribution everywhere else. Both win their home games. Neither is about to starve.

What to watch — signals that Google is moving from “island” to “threat”:

1. External TPU deployments accelerating: Google just signed a deal to deploy TPUs in a third-party data center (Floydstack in NYC) for the first time . If this becomes a pattern — TPUs in non-Google racks — the walled garden narrative could start to break down.

2. Developer activity on TPU/JAX rising: DA Davidson noted TPU developer activity increased ~96% in the past six months . Watch whether this continues.If JAX/XLA starts eating into PyTorch/CUDA mindshare, the software moat erodes.

3. Ironwood adoption outside Google: Ironwood is inference-optimized and competitive with Blackwell on paper (4.6 PFLOPs FP8, 192GB HBM) . If major inference workloads start migrating to Ironwood via Google Cloud, that’s real revenue at risk for Nvidia.

4. More Anthropic-style deals: Anthropic committed to 1M TPUs while still using AWS Trainium and Nvidia GPUs . If other frontier labs follow a multi-vendor strategy that includes TPUs, Nvidia loses pricing power.

5. Packaging capacity shifts: If Google or TSMC announcements suggest CoWoS/HBM capacity is being reallocated toward TPU production, that’s a supply-side signal that Google is scaling external ambitions.

How I’d Frame It as an Investor

TPUs are not sprinting across the field to tackle Nvidia. They’re quietly rearranging which parts of future compute spend are even available to Nvidia in the first place.

And while I have spent a lot of this post minimizing the “TPUs are coming for NVDA” takes, this is a real threat.

Every TPU that runs a Gemini query, every Trainium that runs an AWS model, every in‑house chip Meta deploys in its recommendation stack — those are dollars that never even show up in Nvidia’s addressable market. The top of the capex stack is becoming a split economy: custom silicon for hyperscalers’ own workloads, Nvidia for everything they sell on to the rest of us.

However it’s important to remember a few key things:

1. It’s not trivial for other hyperscalers to go out and replicate Google’s success! They’ve been doing this for 10 years. It’s an open question whether the other hyperscalers could get the same results.

2. Google building out an ecosystem that is competitive with NVDA is a long way off at best.

The questions I care about from here are less “TPU or GPU, who wins?” and more:

• How fast does Google manage to make TPUs boring and invisible behind standard interfaces so that the walled garden grows beyond the current VIP list?

• How quickly does advanced packaging capacity expand, and who actually locks it up?

• Does the AI workload mix drift toward more “agentic,” messy tasks that structurally favor flexible GPUs, or does most of the volume live in big, steady LLM inference where fixed‑function accelerators shine?

None of those have easy answers yet. But they all live downstream of the same basic picture:

Google has built an insanely efficient island.

Nvidia still powers the mainland.

Good luck out there.

If you want to win a FREE premium subscription to MindTheTape.com, just like this post. If it gets 10 likes I will give away a few subs. Liking and restacking also helps motivate me to do more of these deep dives, which are really a lot of work!

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

Is Japan the Weak Link in the Global Money Machine?

If you hang around macro Twitter or listen to podcasts like Macro Voices, you’ll hear people talk about things like global liquidity, carry trades, and the yen.

For most normal humans and even the average trader, that’s all pretty opaque.

But some of the people who watch this stuff closely—like

Michael Howell

of Global Liquidity Indexes—are starting to say something important:

The global “money machine” that has been pushing asset prices higher for years looks like it’s at or near its peak.
And one of the weakest spots in that machine is Japan.

In this post I’ll try to unpack that in plain English:

• What “global liquidity” actually means

• What the yen carry trade is, and why traders love it

• What changed in Japan in 2024–25

• And why this all matters for stocks, bonds, and the next big “risk‑off” event

First: what on earth is “global liquidity”?

Think of the financial system as a network of balance sheets: banks, hedge funds, dealers, pension funds, central banks.

Global liquidity is basically:

How much spare capacity there is across those balance sheets to buy assets, lend money, and take risk.

It’s not just “how low interest rates are.” It’s more like:

• How many credit cards the system has

• How high the limits are

• And how willing people are to swipe them

In his November 2025 slide deck, Howell estimates that this “liquidity stock” is roughly:

• US$185 trillion of global liquidity

• Against about US$115 trillion of world GDP

So the pile of financial “risk capacity” is bigger than the annual output of the entire world economy.

He then breaks that US$185T into two big pieces:

• A smaller slice of central‑bank money (the stuff we usually talk about when we say QE/QT)

• A much larger slice of private‑sector liquidity and collateral. This counts all the bank lines, repos, FX swaps, shadow‑bank credit and so on that grease the system

Since 2008 w have become so used to counting central bank liquidity that we sometimes forget about the importance of this private-sector liquidity. That second piece is how markets can occasionally go wild even when central banks aren’t doing anything dramatic: most of the “juice” lives outside official balance sheets.

Where we are in the cycle

If you plot this global liquidity through time, you get waves: big ups and downs every few years.

In his charts, Howell shows that:

• Liquidity fell sharply into 2022 (think rate hikes and QT),

• Then surged back through 2023–24,

• And by 2025 has stopped accelerating and begun to roll over.

The important point:

We’re not in a full‑blown crunch yet, but the tailwind that helped almost everything go up from late‑2022 onwards is fading.

Add debt on top: why this peak is scarier than past peaks

Imagine you’ve just had a big pay rise (liquidity going up), so you take out a giant mortgage, a car loan, and a couple of cards. No problem… as long as the income keeps rising.

Now imagine the paychecks stop growing, and your fixed payments start to reset higher. Suddenly the same debt feels a lot heavier.

That’s roughly where Howell thinks the world is.

In his deck, he plots a simple ratio for rich countries:

That line is now:

• Higher than at previous famous stress points (’87 crash, Japan bubble, US housing bubble, Lehman, Euro crisis)

• And still climbing

On the next slide he shows a “maturity wall”: a big block of government and corporate debt that has to be refinanced between about 2026 and 2030, after being issued in the ultra‑low‑rate COVID years.

And on top of that, his measure of stress in the plumbing, a spread between two short‑term US rates (SOFR and Fed Funds), has recently spiked into a “danger zone,” which he interprets as a shortage of balance‑sheet capacity or good collateral.

There is more debt than ever resting on a “money machine” whose growth has stalled, and a lot of that debt will need to be rolled at higher interest rates in the next few years.

That’s the background against which we need to look at Japan and the yen.

The yen carry trade

The yen carry trade is one of the oldest macro trades out there. The basic idea is:

1. Borrow in yen, because Japanese interest rates are very low.

2. Convert that yen into dollars or other currencies.

3. Buy something that yields more: US bonds, emerging‑market debt, tech stocks, crypto, whatever.

If the world is calm…

• You collect the yield difference,

• The yen slowly weakens (which adds to your return),

• Everyone’s happy.

If markets get rough and the yen (often a safe haven asset) suddenly strengthens, the trade goes into reverse:

• Your borrowing currency gets more expensive,

• The assets you bought are falling,

• You get margin calls and have to sell.

That’s a carry unwind.

What happened in summer 2024 – the “dress rehearsal”

I wrote about this before, so check that out if you want, otherwise I’ll do a quick recap.

Let’s rewind to the summer of 2024, because that’s when we saw one of these carry accidents in real time.

Very quickly:

• The Bank of Japan raised rates more than markets expected and hinted at buying fewer government bonds.

• At the same time, some weak US economic data hit and volatility spiked.

• USDJPY collapsed ~20 yen in days (the yen surged).

• Nasdaq and S&P futures gapped lower overnight, then bounced hard during the cash session.

• Interestingly, Japanese government bond (JGB) yields fell. They acted like a safe haven.

It was nasty for anyone running “short yen, long risk” trades, but the damage was contained, and there was little lasting damage:

• Global liquidity was still in the “rising” phase of Howell’s cycle.

• The BoJ quickly softened its tone.

• Markets re‑levered and moved on.

Think of that episode as the fire alarm going off in one room of the house, with the power grid still running at full strength.

What’s different this time in Japan

Fast‑forward to the last few months.

Now we see:

• JGB yields spiking to multi‑decade highs. This is markets saying “we want more compensation to hold Japan’s long‑term debt.” You can also think of it as a supply and demand thing, IE there is more supply of JGBs than there is demand.

• The yen is weak, not strong. USDJPY has been grinding higher, making the yen one of the worst‑performing major currencies. In fact, it’s getting close to levels where the BOJ has intervened in the past.

• US tech and global risk assets are wobbling in a slow, grinding correction.

So unlike last summer:

• This is not an “oh no, the yen just exploded higher” moment (yet).

• The pressure is coming from Japan’s bond market and fiscal situation, not just from a surprise BoJ hike.

• And it’s happening at the same time Howell’s charts say the global liquidity wave is topping out, not just starting to rise.

On top of that Japan is one of the most stretched balance sheets in the world: high government debt compared with the liquidity available to support it.

Why this combination is more worrying

Put all of this together and you get a different kind of risk than in 2024.

Last time: “local accident, global tailwind”

In 2024:

• Global liquidity was accelerating out of the post‑COVID slump.

• The yen surge was mostly an FX and volatility shock.

• JGBs rallied; BoJ could calm things just by taking a softer stance on hikes

• The rest of the world still had spare balance‑sheet capacity to absorb losses and rebuild positions.

Now: “late‑cycle stress at a weak point”

Today, according to Howell’s framework:

• The global liquidity tailwind has stopped getting stronger and is starting to fade.

• Debt levels vs liquidity are worse than at previous famous crises, and a big chunk of that debt will be rolled over in the next five years.

• Japan is a heavily indebted country trying to normalise monetary policy, with its bond market already under pressure and its currency still used for funding.

That mix means:

If something goes wrong in Japan now, it doesn’t hit a world with plenty of unused balance sheet.
It hits a world where everyone is already pretty full on debt.

That’s why people are suddenly talking about Japan and the yen carry trade again in the context of “global liquidity peaking.”

What a really bad version would look like

To be clear: none of this guarantees a crisis. I simply wanted to write this to make you aware of a growing risk. But if you wanted to sketch how a truly nasty episode might unfold, it would probably look something like this:

1. Bond stress intensifies
   JGB yields push higher in a disorderly way; investors start questioning Japan’s general sustainability.

2. Policy makers panic about the currency or the bond market (or both)

   • Perhaps the Ministry of Finance intervenes aggressively to support the yen.

   • Or the BoJ changes course again,, either hiking more than expected to defend the currency, or buying huge amounts of bonds to cap yields.

3. The yen suddenly spikes higher
   That’s the moment the carry trade breaks climactically: funding costs jump and anyone who was short yen and long risky assets gets hit on both sides.

4. Global risk assets sell off as positions are unwound
   Tech, EM, credit, crypto, anything funded in cheap yen, could see forced selling.

5. This happens with global liquidity already rolling over
   So instead of markets easily digesting the shock (like 2024), it collides with the refinancing wall and tight collateral conditions.

Again, this is a scenario, not a prediction. But it’s the reason people suddenly care a lot about what used to be a very boring government bond market.

If you want one simple takeaway

Here’s the punchline I’d give:

• For the last couple of years, the world has been surfing a big wave of “easy money” in a very broad sense, with lots of balance‑sheet capacity, strong cross‑border flows, everyone happy to take risk.

• That wave looks like it’s cresting. The growth in that “money machine” has slowed.

• At the same time, we’re carrying more debt than ever on top of it and heading into a period where a lot of that debt needs to be refinanced.

• Japan, huge debt, changing central bank policy, stressed bond market, weak funding currency, is one of the places where a late‑cycle accident is most likely to show up.

• The yen carry trade, which worked beautifully on the way up, is now a potential transmission channel for stress on the way down.

So when you see headlines about Japanese yields spiking, or the yen suddenly surging, it’s not just a local story about Tokyo. It’s one of the key pressure points in the global financial system, right at the moment when the tide that’s been lifting all boats may be starting to turn.

Zooming out

I don’t want this post to come off as alarmist. In general, I hate people who start macro doom posting at the lows. We are down 550bps in SPX, and that’s hardly cause for elaborate doom posting…

But I wanted to write this post because I think the fading liquidity, the Fed’s current desire to be overly careful with cutting (December cut is down to ~32%), and the carry trade stuff leads to a window of concern here, and might explain the otherwise very strange price action this week.

Good luck out there.

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

Last week was an interesting and chaotic one, with the final S&P 500 result of +.07% doing little justice to the swings.

The Nasdaq had a slightly worse week, finished down ~22bps.

Part of the reason for the mismatch between how traders felt last week and how the indices did is because we ripped Monday with SPX closing up ~150bps and NDX closing up ~200bps.

On top of that, retail-heavy names took a serious beating from those Monday highs.

Non Profitable Tech off 8%, Retail Favorites off 5%, Memes 4.38%, etc. means there was a lot of pain among the High Beta Boys (HBBs).

In the factor world (link if you don’t have a good grasp on factors), Momentum and Liquidity (highly traded names) took a back seat to Low Vol and Quality, which means that any form of “hot money” got waxed.

In fact, “High Beta Momentum” had its worst day Thursday since “DeepSeek Monday.”

So while the indices may have closed mostly flat, chances are you felt the pain yourself.

Set Up For Vol

Momentum unwinds are simply a feature of the modern markets. Because there is research showing that momentum is one of the few factors that produces durable “alpha” there is a lot of money that follows the momentum factor. That contributes to these miniature “boom and bust” cycles, or as calls them “rolling bubbles” (or as calls the, “rolling ponzis”).

And it makes sense to have had one of these some time in Q4. For one, Q4 seasonality is not great for momentum on the whole.

Liquidity on the whole is also drying up a bit. It’s a complicated subject, but has been projecting it to fall through 2026 before rebounding.

And the Fed had a “secret meeting” with Wall Street last week over the repo spike and money markets, implying that there is some stress there.

All of this has predisposed markets to a wobbly period, and it likely all contributes to why the VIX is staying elevated.

More Vol Ahead?

This week is monthly OpEx, and while I don’t believe it’s a particularly large one dealers are heavily short Gamma going into Monday.

When dealers are long Gamma they tend to hedge against moves, buying as the market sells, and selling as the market buys. When the are short Gamma they do the opposite, chasing moves and exacerbating them.

Also, the December Fed meeting has moved from almost a sure cut, to now a 55/45 “no-cut” projection. If the market continues to price out cuts, that could be a headwind for equities.

On the more protective front, it appears the TGA buildup is now behind us. Rather than the Treasury General Account pulling money “out” of the economy (a simplistic explanation but suitable for our needs) it should now begin pushing money IN to the economy.

But we also have a heavy event schedule this week, with NVDA earnings Wednesday, September (you read that right) Non Farm Payrolls releasing Thursday, and Gemini 3.0 expected sometime this week…

Gemini 3.0 is expected to be the new best in class LLM model. After a relatively disappointing GPT-5 release, all eyes will be on Google to see if they can honor the scaling laws and keep the models moving forward.

The State of the AI Trade

AI has been the source of a lot of unease the last few weeks. Sama made a rare public misstep in his Brad Gerstner interview, the AI stocks drew down heavily, and the chatter about a “debt fueled capex bubble” reached a fever pitch.

There are growing risks in the AI trade, one of which I highlight above: Gemini 3.0 disappointing and “popping” the bubble of model improvement expectations.

But the reality is that the earnings expectations for AI companies continue to get revised up.

The hyperscalers made it clear that they plan to spend more than ever in 2026.

And while the Big Tech firms’ free cash flow is decreasing to fund these capex expectations the idea would be that this capex is ROI positive. Plus, they’ve managed to avoid lowering buybacks by taking on debt.

My Vibe and My Trades

Like this post for a chance to win a FREE month subscription if it gets 10+ likes.

Read more

More than 65% of the S&P500 have reported earnings, and their results have deviated meaningfully from recent history.

More companies are beating consensus estimates than anytime since immediately after COVID.

But they are beating by significantly smaller amounts than in recent history.

Look at Q2 2024 on the chart above. Then compare it to NDX’s performance in that same period on the chart below.

Not great.

It’s already manifesting a bit like Q2 2024 in other ways as well. Companies that are beating on earnings are not being rewarded for it compared to history.

When market expectations are high and results are good but not great you end up with some disappointed investors.

I don’t necessarily think we will see the same drawdown we saw in Q2 24’ off of this, because whereas in 2024 there was still deep lingering fear of a recession after the 2022 rate hiking cycle, we have a much more structurally optimistic investing environment this time around.

Still, these little cracks showing add up and are worth watching. The market is often willing to shrug them off until they either go away or you reach a critical mass.

Employment

We continue to only see limited employment data due to the government shutdown, but the private ADP data we see shows continued slowdown.

You know what would be ironic is if ChatGPT release caused companies to realize that AI was coming for the jobs and slow hiring, thereby causing a recession that sets AI back 5+ years!

There is a confounding variable surrounding all this employment data in the immigration policy changes, but we need to see the job market hold steady or even the so-far-untouchable AI sector could start to wobble.

AI Themes

Last week was a big week for the AI story. With all reporting earnings and commenting heavily on AI infrastructure and technology on their earnings calls we got a good look at the state of the theme.

If you missed my Meta update, you should check it out here:

AMZN and GOOG were the big winners of the week each up ~7%.

What’s notable is that they are also the two biggest spenders on capex.

And generally across the whole hyperscaler complex the capex guides have substantially beaten expectations!

So the hyperscaler theme is certainly alive and well, despite what happened to Meta. There appears to be a growing preference for companies that are directly monetizing their massive capex, and GOOG and AMZN both fit the bill there with huge cloud revenue growth, and in Google’s case, the chance to have a leading-edge model as well.

An interesting tidbit also got thrown out by Microsoft CEO Satya Nadella. Satya, when asked by Brad Gerstner about a potential for a supply glut, responded:

I think the cycles of demand and supply in this particular case, you can’t really predict, right? The point is: what’s the secular trend? The secular trend is what Sam (OpenAI CEO) said, which is, at the end of the day, because quite frankly, the biggest issue we are now having is not a compute glut, but it’s power — it’s sort of the ability to get the builds done fast enough close to power,” Satya said in the podcast. “So, if you can’t do that, you may actually have a bunch of chips sitting in inventory that I can’t plug in. In fact, that is my problem today. It’s not a supply issue of chips; it’s actually the fact that I don’t have warm shells to plug into.”

In other words, Satya is no longer chip-constrained but “warm shell” (powered and ready datacenter) constrained. The market has begun to rapidly reprice some datacenter/power related companies higher this year, but Satya’s recent comment is likely to propel that forward in earnest.

I get into more on that in the Trade Ideas section below.

Trade Ideas

If you’re not already a paid sub, like this post. If it gets 7 likes I’ll give away one sub. If it gets 15+, I’ll give away two.

Read more

Both Google and Meta reported earnings yesterday after the bell, and to say that the stocks had different reactions would be an understatement.

Meta dropped sharply by about 10% and continues falling during the call and overnight.

I was a bit surprised by this reaction. After adjusting out a one-time tax payment due to the One Big Beautiful Bill Meta’s headline numbers weren’t bad.

The Numbers

*REV. $51.24B, EST. $49.59B

*EPS $1.05 VS. $6.03 Y/Y

*ADJ EPS $7.25 vs $6.03 Y/Y (Adjusting out the one time payment)

*SEES 4Q REV. $56B TO $59B, EST. $57.38B

*SEES FY CAPEX $70B TO $72b, SAW $66B TO $72B, EST. $69.3B

*EXPECTS CAPEX DOLLAR GROWTH NOTABLY LARGER IN 2026 VS 2025

Looking at these headlines two things stand out. Firstly, Meta’s 4Q projections were roughly in line, and secondly they are expecting capex to grow “notably larger” in 2026.

Digging deeper, Meta’s Family of Apps revenue is actually growing faster than Google Search revenue, and has been at an increasing rate since 2023.

So while Google had an excellent quarter the Meta cash machine actually seems to be spinning up faster than Google’s core business, while operating margins are expanding at a similar pace.

The problem, then, is on the expense side. Meta warned that 2026 capex would be “notably larger” and that expenses would grow “at a significantly faster rate” compared to 2025.

To put that in context let’s take a look at how Meta is using their existing compute.

AI in the Core Business

To understand whether the capex plans are bullish, bearish, or something in between let’s take a look at some of Mark’s statements on the earnings call.

Meta Superintelligence Labs is off to a strong start. I think that we’ve already built the lab with the highest talent density in the industry. We’re heads down developing our next generation of models and products, and I’m looking forward to sharing more in that front over the coming months. We’re also building what we expect to be an industry-leading amount of compute.

Now, there’s a range of timelines for when people think that we’re going to get superintelligence. Some people think that we’ll get there in a few years. Others think it will be five, seven years or longer. I think that it’s the right strategy to aggressively front load building capacity so that way we’re prepared for the most optimistic cases. That way, if superintelligence arrives sooner, we will be ideally positioned for a generational paradigm shift and many large opportunities.

If it takes longer, then we’ll use the extra compute to accelerate our core business, which continues to be able to profitably use much more compute than we’ve been able to throw at it and we’re seeing very high demand for additional compute, both internally and externally. And in the worst case, we would just slow building new infrastructure for some period while we grow into what we build.

Pulling this next part out separately as I think it’s the key to understanding Meta’s capex plans.

And one way that I think about our company overall is that there are 3 giant transformers that run Facebook, Instagram and ads recommendations. We have a very strong pipeline of lots of ways to improve these models by incorporating new AI advances and capabilities.

Let’s break down what Mark is saying.

Read more

Markets broke out to ATHs to end last week, and futures have opened even higher this evening in New York.

There are two common narratives I hear surrounding “ATH breaks” in equities. Some people claim that ATH breaks are bullish and should be bought, others see high prices and think they want to wait and buy lower.

JPM did an analysis of past ATH breaks since 1970 and discovered that buying them is basically the same as buying at any other time. They are ever so slightly more bullish than average on a long-term basis, and slightly less bullish than average on a short term basis. More importantly the data is close enough so as to likely be just noise.

Don’t fear the ATHs.

On the seasonality front, last Thursday was the single best day of the year (historically) to buy stocks on a 3-month forward return basis.

Seasonality is very bullish here, with a huge number of S&P stocks coming out of their blackout periods and beginning to repurchase their shares into EoY.

Positioning

Positioning continues to surprise me. Goldman is modeling L/S investors as having high gross exposure but quite low net exposure. This is interesting because a high gross exposure implies that they are confident of a lower volatility regime (bullish) but are choosing to keep net exposure low (bearish).

Meanwhile, long only investors have been buyers since the April bottom but not at extreme levels. Hedge funds continue to doubt the rally and continue to be net sellers.

This has really been a retail-driven market, and that trend seems likely to continue into Q4 and even into 2026 if Goldman’s forecast is right.

Earnings Season

Earnings season is well under way and this week is a big one with ~35% of the S&P 500 reporting.

So far we’ve seen a steady string of earnings beats, although I would say that was largely expected by the investors I talk to.

This is an interesting chart showing that we are in a somewhat unique moment where EPS growth is outpacing P/E growth without a rate hiking cycle.

We’re also in a period of dramatically expanding margins for the S&P500. Anyone who talks about P/E ratios being at “ATHs” without discussing this magin-growth story is larping.

My Vibe

ATHs aren’t magic, but seasonality and buybacks lean bullish, positioning still (somehow!) leaves plenty of dry powder, earnings breadth is decent, and margins are expanding.

Want to read the rest? Like this post and if it gets 10 likes I’ll give away 2 months of premium!

Read more

The absolute worst genre of finance post is the Omega Doomer post where the writer tells you some equivalent of “we’re all gonna die” that invariably never actually happens.

These posts proliferate though, not because they are valuable or make anyone money, but because they appeal to something in human nature that keeps us always looking over our shoulders.

Sometimes that instinct can save us a lot of money, and sometimes it just leads us into buying expensive puts that go to Put Heaven™️.

But it was that instinct that led me to dig into what’s going on in markets, and is leading me to write what is essentially a full-blown doomer crash post. The same post you’re reading right now.

And I think this one might be different from its peers in that it actually holds some alpha.

Vol is creeping, credit is wobbling, and record October Opex creates gap risk into Monday. So strap in and get ready for a good old fashioned Halloween-themed DOOM post.

Heightened Crash Risk

Last Friday Trump threatened a 100% tariff on China that sent markets into a tailspin. SPX promptly sold off 300bps intraday, bitcoin crashed 12% and long bonds rallied 70bps.

Markets initially rallied on Monday after Trump TACO-d in a Truth Social post telling us “not to worry” about China.

But since that “truth” markets have remained choppy, and more interestingly, VIX has been progressively bid up throughout the week.

On Thursday we got word of some surprising credit charge-offs by Zions Bancorp that sent regional banks into a tailspin, and the ETF had it’s worst day since Liberation Day.

A $50m write-off for some bank in Utah is hardly immediately relevant to global financial markets in itself, but it is worrying because it comes on the tails of some other recent credit events, and it represents a big loss for Zions.

But there have been signs of stress in the debt markets for a while now, they’ve just gone largely unnoticed because credit spreads have been stable.

Other Signs of Stress

Over the last few years credit has moved more and more into opaque “private credit” markets. In 2024 private credit assets under management reached $1.7 trillion, and it’s only grown since then.

That has made it difficult for investors like us to get a window into how corporate credit is actually performing. Private credit funds are under no obligation to mark their securities to market in real time, and even when they do mark to market they have no mechanism for finding a “true” price.

But we can look at a relatively obscure corner of the market for clues.

Business Development Companies are publicly traded funds that make loans to small and mid-sized U.S. companies, and then pass most of the interest they collect back to shareholders as dividends. Think “publicly listed private-credit funds.”

And they’ve been getting clobbered for months now.

Alternative asset managers like PE funds, which are also a proxy for the opaque world of corporate debt and the “real economy” have started to get beaten down as well.

And the two year yield (a bellwether for the economy) has been down only for months now.

On top of all of this, Gold has been going absolutely parabolic. Gold may not be saying anything that’s relevant to equities, but it sometimes has in the past.

Enter…

Black Monday Redux (The Spooktacular)

If you enjoyed this post please hit it with a like. If it gets 10 or more likes, I’ll give away two months of paid subs to people who like this post.

Read more

Author’s Note: I will continue releasing free content where I can provide value to everyone. Posts will include premium content for paid subscribers at the end of emails or as separate write-ups. Free content will focus on macro views, whereas the paid content will focus more on the tactical trades I’m making.

Serious traders or investors should join Premium to get the best, most timely information.

Since last week’s note where I said we were heading for “another leg higher” we have seen… another (small) leg higher.

A 2% week is nothing to sneeze at.

I don’t have a lot more to say on positioning or near-term direction at this time, but I wanted to write you guys a note to discuss whether the AI trade is likely to run out of steam.

The Math Behind The AI Trade

Kuppy (of all people) wrote an interesting note this week on the AI trade. In it, he made a few interesting claims which I’ve clipped for you.

Basically, Kuppy is flagging what many people have been flagging recently: the math behind the “AI trade” stretches belief.

I’ve been a huge AI bull throughout this newsletter, and while I think the market has possibly more than caught up to my beliefs, I wanted to check in on the math myself. So I did some research.

You can think of the AI “economy” as having three broad pillars.

Pillar #1 is the big one everybody thinks of: Capex. It’s what the hyperscalers and others are spending to build those massive data centers like this one was kind enough to make a video of:

Pillar #2 is how the capex spend monetizes. Large providers sell data center compute largely to other massive AI companies like OpenAI, Anthropic, the hyperscalers themselves, etc.

Pillar #3 is what really matters, as it’s the revenue earned by the application layer. It’s the amount of money all these companies spending hundreds of billions to rent AI services from the infrastructure layer are actually able to earn in the software marketplace.

Let’s take a look at the numbers.

Pillars of the AI economy.

Pillar #1: 2025 AI Capex (The Input)

• Total: $350–400B

• Amazon: ~$100B

• Microsoft: ~$80–89B (FY25)

• Alphabet: ~$75–85B

• Meta: ~$66–72B

• Oracle: ~$21–25B

• “Other” DC builders (neoclouds, telco, regional): ~$10–20B

Pillar #2: 2025 AI Compute Revenue (The Tollbooth)

• Total: $55–70B

• AWS (AI-attributed share): ~$25–30B

• Azure (AI-attributed share): ~$12–18B

• Google Cloud (AI-attributed share): ~$6–10B

• Oracle Cloud (AI-attributed share): ~$3–5B

• Independent GPU clouds: ~$5–6B
  CoreWeave ~$4.9–5.1B, Lambda ~$0.5B, Together ~$0.1B

Pillar #3: 2025 AI Application Revenue (The Outside Money)

• Total: $32–55B

• Independents, illustrative split:

  • OpenAI: ~$13B

  • Anthropic: ~$3–6B run-rate (2H25 basis)

  • AI coding tools: ~$1.5–3B (Copilot + Cursor, etc.)

  • Creative/search/chat apps: ~$0.6–1.0B
    Midjourney ~$0.3–0.5B, Perplexity ~$0.1–0.15B, Character.AI ~$0.03–0.05B

• Enterprise “AI attach” (incremental to suites): ~$7–15B

  • M365 Copilot, Salesforce, ServiceNow, Adobe Firefly, etc.

Looking At The Long Term Math

We need to see Pillar #3 grow substantially to justify this capex spend. Let’s look at the current CAGR for this segment.

If we take a 55% growth rate on the software side, and a 15% growth rate on the capex side (which is actually a slowdown, considering replacement costs at a 5 year useful life), we see that the complex remains firmly money negative through 2030.

If you look at the cumulative numbers, you see that a $2T gap means trillions are incinerated to produce ~$1T in value.

What Does It Take To Make The Math Work Out?

Put simply we need to see higher growth in Pillar #3 businesses. Using a 75% CAGR gets us to nearly breakeven by 2030.

Using a 85% CAGR for software growth gets us even more firmly positive in 2030.

Achieving this level of growth is not impossible, but it means we need to see Pillar #3 revenue accelerating upwards from here. Furthermore, AI Software revenues will need to more than DOUBLE the entire cloud software industry of today by 2030.

Takeaways For The AI Trade

Put very simply, ~$400b is being spent to develop data centers that are expected to rent for ~$65b this year. The companies renting that data center space expect to earn between $40b from “outside” of the AI complex.

This requires explosive revenue growth from AI software providers to justify. Continuing to grow at the same pace is not enough. One of three things will happen:

1. AI software revenue growth increases dramatically enough to keep the party going.

2. AI software revenue growth increases, but not fast enough to justify the current capex. Hardware providers will be hit hardest, while cloud software providers will benefit form over-the-top capex spending driving their costs down.

3. AI software revenue growth stays flat at 55% or decreases, and the entire complex of “AI” companies has to be dramatically rerated downward.

The conclusion is simple: Even for an AI bull like myself, it seems impossible for the AI trade to play out as immaculately as the market expects.

The problem is that this fact is not immediately actionable. It’s not enough to think that the current AI projections are wrong. Contrary to pop culture, trading is not about being early to a thesis, putting the trade on, and waiting for the market to catch up. That’s a good way to go broke!

“AI” is likely to continue to run from here.

While everyone is watching the capex spend, I’ll be carefully watching AI associated revenues. If we don’t see continued acceleration it could mean trouble ahead.

Good luck out there.

This one took a lot of time and effort to write, so don’t be shy about liking and sharing. If this post gets 10 likes I’ll give away two months of premium subscription!

Subscribe now

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

Author’s Note: I will continue releasing free content where I can provide value to everyone. Posts will include premium content for paid subscribers at the end of emails or as separate write-ups. Free content will focus on macro views, whereas the paid content will focus more on the tactical trades I’m making.

Serious traders or investors should join Premium to get the best, most timely information.

Hello everyone.

Equities pushed to fresh highs before a modest pullback. Despite softer labor signals and stretched multiples, positioning remains light—last week’s de‑grossing leaves room for a “surprise” leg higher into Q4. A firmer dollar is a swing factor: it tightens conditions, but with underexposed “smart money” and potential Fed easing ahead, the path of least resistance may still be up.

Below: the positioning, the policy setup, and what would change my mind.

My last post about taking down risk led into a few months of chops, but since then equity markets have continued higher.

I’ve found the market a bit hard to read the last month or so. On the one hand the jobs data has clearly deteriorated to near-recessionary levels.

While P/E ratios have been breaking out to historical highs.

But GDP growth projections have also remained extremely high.

So there hasn’t been much in the way of slam-dunks in US equities as far as I can tell.

However, I believe that’s changed…

A Correction For Ants

NDX and SPX both made new highs last week before selling off.

The actual size of the sell off was quite small. Like all of the post-TACO (Trump Always Chickens Out) dips have been.

The market structure of this rally has left smart money relatively under-exposed the whole way up. That has kept the dips shallow. Notably, last week saw a significant de-grossing on equities down to quite low levels.

Also, the DXY seems to have bottomed, and might be looking for another leg higher.

These two factors, particularly the gross/net exposure numbers, lead me to think we are going to get a “surprising” move up from here, possibly all the way through Q4.

One thing I am watching is month/quarter-end in the first few trading days of next week. There are a lot of pension funds who will be looking to rebalance out of equities. I believe some of that has already happened, and I don’t expect too much turbulence, but the possibility is there.

The Mother of All Asset Bubbles

The Fed appears ready to re-begin their easing cycle while US indices are at their ATHs.

Since 1995 there have been three instances where the Fed has cut with the NDX ~at all time highs.

One in particular, 1998, stands out for culminating in the Dot-Com Bubble.

The Dot-Com Bubble itself was characterized by low or no revenue internet companies reaching insane valuations. This happened because:
1. There was a promising new technology for investors to get excited about.
2. Interest rates were lower than the neutral rate creating artificially loose financial conditions.

Is the stage set for something similar?

Monetary Illusions

There is a growing “soft” crisis of confidence in the US dollar. The average consumer is painfully aware of how vulnerable their purchasing power is, to the point where even your local uber driver knows (consciously or intuitively) that they’re expanding the money supply at a scary rate (6.5%/year over the last 20 years).

This is a large part of why Gold has been on such a tear for the last few years.

I don’t have any fancy data or articles for this point, but I think this contributes to setting the stage for an epic asset bubble. People are primed to believe stocks only go up, and they are looking for something to do with their ever-devalued paper.

Combine that with loosening financial conditions and Fed cuts… it’s hard to stay out of the pool for long when it comes to investing.

What I’m Watching

There are obviously concerns around valuation. Valuations don’t predict crashes but they do have an influence on forward returns.

In order for US equities to continue their march higher these main factors must persist:

1. Positioning has to remain constructive. Once it gets stretched the “marginal buyer” has already bought and we are likely to stagnate.

2. Inflation has to stay “relatively” subdued to permit Fed easing.

3. The economy can’t weaken fully into recessionary territory.

4. The fiscal spigot has to stay on.

Point #4 is particularly interesting in the context of next week’s “government shutdown” negotiations. Trump has threatened to fire thousands of Federal workers if a deal isn’t reached.

There is some chance that this shutdown debate crystalizes around some measure of fiscal responsibility. While I find it unlikely that anything meaningful will come out of it, that’s something to watch for this week.

Obviously this Friday we also have Non-Farm Payrolls. Pay particular attention to that data print and any August revisions. We are in the danger zone around jobs, so while I think the market is prepared to be forgiving any especially cold data could raise alarm.

I’m going to try something different this week. If this post gets 10 likes I’ll give away an annual sub.

Good luck out there.

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

Author’s Note: I will continue releasing free content where I can provide value to everyone. Posts will include premium content for paid subscribers at the end of emails or as separate write-ups. Free content will focus on macro views, whereas the paid content will focus more on the tactical trades I’m making.

Serious traders or investors should join Premium to get the best, most timely information.

I haven’t written in over two months, but it’s for a good reason.

The whole way up I’ve felt like my last post, The Most Hated Rally, has pretty well covered my thoughts. Sure there were opportunities to add a few points, or discuss the news, but the whole way up we have been in the midst of…

In that post I shared my basket at the time.

I added exposure to the AI theme along the way, but even without adding, that basket has done extremely well.

10%+ in about two months. Not too shabby.

So while I haven’t had a ton to add these last months, I feel it’s finally time to start thinking about what comes after the rally…

August Reminder

There has been a massive vol-control bid as volatility has rolled “out of sample” for vol-control funds. That, combined with bid from almost all types of institutional investors (CTAs chasing the emergent trend, hedgies reallocating to tech, etc) has led to a sustained “crash up” rally.

My thesis the whole way up has been that as long as this structural bid remained present, we would likely grind higher.

And we have! One point to Gryffindor or whatever.

But I’m starting to see signs that things are getting frothy. Check out a few charts.

We’ve also started, in the last couple weeks, to get a bit of a “breadth” breakdown. When breadth gets weak, that can be a prelude to a sell-off.

Take a look at how breadth began weakening in December of last year and how the market started selling off about 6 weeks later.

Now, the critical among you will say “well if breadth is just starting to weaken, and weakening breadth can take months to manifest in a sell-off, what good is this information?”

That’s a good question. But whereas for the last few months’ rally we had improving breadth, and positioning that allowed a lot of length to come into the market, we are now starting to see the opposite.

That means, to me, that we are likely to see a pause in this rally for now.

So whereas for the last handful of months I have flirted with being around 100% long (or more, at times), I’m now starting to think August could be a rocky month for stocks and I’ve dialed my exposure back to about 60% long.

Parting Thoughts

Writing this newsletter has been a ton of fun. But it has also been a ton of work. I’m not sure what I will do with it going forward, or how much I’ll write. I hope this short note helps you out, and acts as a reminder not to let yourself get frothy mentally as the market gets frothy financially.

If you want to read more from me, I’d love and appreciate shares, likes and comments. Writing these things can be a lonely task sometimes!

Good luck out there.

Disclaimer: The information provided here is for general informational purposes only. It is not intended as financial advice. I am not a financial advisor, nor am I qualified to provide financial guidance. Please consult with a professional financial advisor before making any investment decisions. This content is shared from my personal perspective and experience only, and should not be considered professional financial investment advice. Make your own informed decisions and do not rely solely on the information presented here. The information is presented for educational reasons only. Investment positions listed in the newsletter may be exited or adjusted without notice.

Author’s Note: I will continue releasing free content where I can provide value to everyone. Posts will include premium content for paid subscribers at the end of emails or as separate write-ups. Free content will focus on macro views, whereas the paid content will focus more on the tactical trades I’m making.

Serious traders or investors should join Premium to get the best, most timely information.

These have been weird markets to trade, and I haven’t really been sure how to write about them.

Jerome Powell’s moment to shine came and went this week. Reporters asked a lot of questions and Powell stretched the English language’s ability to say “I don’t know.”

From there, focus shifted to the first trade deal (the first of many, if you believe Trump) which was announced this morning. The 10% baseline tariffs remained with carveouts for various industries and companies (Rolls Royce enjoyed a nice carveout and a health round of glazing by Trump and the UK ambassador).

One might have thought that if our “oldest ally” (Trump’s words) couldn’t get the baseline tariffs below 10% then that would be an unpleasant surprise for markets. But that wasn’t the case, as the event included plenty of bullish rhetoric from Trump including:

You better go out and buy stock now.

• Trump, today

Markets ate it up for a while but couldn’t sustain the big into the close. They finished the day little change after gapping up.

I’m taking Trump at his word for now.

Technically Interesting

The DXY broke out of it’s downward grind. Is this the end of the end of American Exceptionalism? Possibly. I think the world is ready to be surprised by just how resilient the United States’ economy, people, and oh yea, 7% fiscal deficits are.

The big green candle today in the DXY does imply that there may be a line forming of foreigners looking to do business with the US’ markets. Surprise, surprise.

Is old Klaus taking an interest in the US again?

Meanwhile, somehow, sentiment is still quite stretched to the downside. The AAII bulls - bears index remains only modestly off the lows.

McElligot said it best this morning:

There have been a lot of mechanical factors driving a bid beneath the surface this week. Vol Control funds, and CTAs have bought as corporate buybacks have begun as a continued influx of retail buying has push a lot of capital into relatively illiquid markets.

The most recent charts I’ve seen have shown hedge funds to still be under-exposed. There is a ton of capital that is now staring at this rally wondering if it’s time to jump aboard but too scared of the state of the economy to do so just yet, because…

The Economy Is Definitely Slowing

Read more

Author’s Note: I will continue releasing free content where I can provide value to everyone. Posts will include premium content for paid subscribers at the end of emails or as separate write-ups. Free content will focus on macro views, whereas the paid content will focus more on the tactical trades I’m making.

Serious traders or investors should join Premium to get the best, most timely information.

First, a little housekeeping. I’m constantly thinking about how to make this newsletter better and I was hoping to get a little feedback from you guys. Please take a second and answer the poll.

Markets had a good week last week with the NDX and SPX both closing meaningfully higher.

The upward momentum was made even more dramatic by the fact that it came in the last 4 days of the week, as Monday closed pretty red.

Zooming out, we are around 17% off the lows in NDX and 14% off in SPX. Wow.

Importantly we are also above the “post-pause” S&P highs marked on yellow in the chart. That corresponds to roughly the 5500 level in ES, which is now a big psychological pivot point. People will be looking for ES to hold above 5500 to get bullish again, or fall below it to get bearish again.

We can expect some Soros-style reflexivity around that number. In other words, higher prices might beget higher prices for a while, and vice versa.

In The News

Headlines this week were interesting in the sense that there were as many “positive” headlines as possible without anything actually happening.

It’s pretty clear that the administration is afraid of the cat they’ve let out of the bag, and is now trying their best to put it back in.

Bessent came out today victory lapping about the stock markets going up. In general when a politician victory laps or dances on a grave, that’s a pretty good turning point. As an example, let’s look at how TSLA has done since Tim Walz told an audience he likes to check the stock price for a “pick me up” during the day (because of how badly it’s doing).

Heh.

Unlike Tim Walz, Bessent has a lot of levers to pull that directly affect the stock market. So to some degree, I think we need to take him at his word when he tells us he’s watching it, and takes credit for making it go up.

At the very least, it implies he isn’t eager to see it drop again, and Trump seems more and more content to trust the economy to the “money guys” over Navarro or others.

Trade Deals

Tangible trade deal progress remains elusive. Trump keeps saying he’s in talks with China while China keeps saying “we haven’t talked.”

It’s a situation so outrageous that it can only really be explained with a meme.

On the flip side, it looks like we might have a loose framework for a South Korea deal this week.

The market has been taking a lot on faith this week, but at some point soon we will need to see a signed piece of paper if the market is to believe in the trade deals.

Positioning

The move higher this week was driven largely by short covering, as everyone is afraid to be short when the tangible trade deal announcements actually start rolling in.

Read more