Patient Zero

---

There is no point pretending to be coy about the conclusion, so let us begin with the fact that organizes everything else: Nvidia is the most valuable company that has ever existed, worth a little over $5 trillion, and it is, by a wide margin, the most important single stock on the planet. Not the most important technology stock, or the most important chip stock — the most important stock, full stop. It is the largest constituent of the S&P 500 and of the MSCI World. It sits, directly or through index funds, inside a substantial fraction of every pension, every 401(k), every sovereign-wealth portfolio on earth. When Nvidia moves, the market moves; on a single bad day this spring, it shed more than $320 billion of market value — an entire Coca-Cola, gone between lunch and the closing bell — and dragged the indices down with it. No company has ever been this large, and no company this large has ever been this central to the fortunes of this many people who have never knowingly bought a share of it.

That centrality is precisely why it deserves the most skeptical reading in the market, not the least. The bullish case for Nvidia is so obvious, so well-rehearsed, and so visibly correct in the rear-view mirror that it has become almost impolite to interrogate. The numbers are staggering. The technology is genuine. The dominance is real. And yet underneath the triumph sit two facts that, examined honestly, describe a degree of fragility that the $5 trillion valuation does not acknowledge — and because Nvidia is the keystone of the entire edifice, that fragility does not belong to Nvidia alone. It belongs to everyone standing on the structure it holds up.

This is a piece about those two facts, and about why, if the AI boom ever sickens, the first symptoms will appear here, in the patient from whom everyone else caught the trade.

Giving the giant its due

Begin with the case for the defense, because it is overwhelming and it is true, and any honest account has to sit inside it before pushing against it.

Nvidia's business is, by almost any measure, the most spectacular in the history of corporate enterprise. In its fiscal 2026 the company booked $215.9 billion in revenue, up 65% in a single year, at gross margins of roughly 71% — that is, it keeps seventy-one cents of gross profit on every dollar of sales, a figure that belongs to a luxury-goods house or a software monopoly, not a maker of physical hardware that must be fabricated in silicon and shipped in crates. Then it accelerated. In the first quarter of its fiscal 2027, reported in the late spring of 2026, revenue reached a record $81.6 billion, up 85% from a year earlier and 20% from the prior quarter — the third consecutive quarter in which the rate of year-over-year growth went up rather than down, a near-impossible feat for a company already this enormous. Data-center revenue alone was $75.2 billion, up 92%. Management guided the next quarter to around $91 billion, blowing past Wall Street's estimates, and waved the flag of confidence with an $80 billion share-buyback authorization and a dividend increase.

This is not the financial profile of a bubble stock in the pejorative sense. Bubble stocks do not earn 71% gross margins on two hundred billion dollars of real revenue. Nvidia makes the best AI accelerators in the world, surrounds them with a software ecosystem — CUDA — that has taken rivals a decade to fail to dislodge, and sells every unit it can manufacture into a wave of demand that is, for now, genuinely insatiable. The company is a marvel. Its founder, Jensen Huang, has built something close to a monopoly on the most strategically important commodity of the age. If you are looking for a story in which Nvidia is secretly worthless, you will not find it here, because it is not true.

The question is not whether Nvidia is a great business. It is whether a great business resting on the two foundations we are about to examine should be the single largest holding of the entire financial system — and whether the people who own it, mostly without choosing to, understand what they are actually leaning on.

The first fact: three customers

Here is the first foundation, and it is the one Nvidia's own filings disclose in language so dry that almost no one reads it.

A company with $215 billion of revenue sounds diversified by definition; surely sales that vast must come from thousands of buyers. They do not. Nvidia's revenue is among the most concentrated of any large company in the world, and the concentration is getting worse, not better, as the boom matures. A year and a half ago, the company disclosed that two unnamed customers together accounted for about 39% of its revenue. By the third quarter of fiscal 2026, four direct customers each exceeded 10% of sales — 22%, 15%, 13% and 11% respectively — meaning 61% of revenue came from just four buyers. By the first quarter of fiscal 2027, the most recent reporting indicates three customers accounted for roughly 64% of revenue. Read that progression slowly. As Nvidia has grown, its dependence on a tiny handful of customers has intensified: from two-fifths of revenue concentrated in two names, to nearly two-thirds concentrated in three.

Who are these customers? Nvidia does not name them, but they are not a mystery: they are the hyperscalers and the AI labs — Microsoft, Amazon, Alphabet, Meta, Oracle, OpenAI, and the cloud builders like CoreWeave — the same small club of giants that recurs in every chapter of this boom. And the thing those customers have in common is the thing that should worry a Nvidia shareholder: they are not buying chips because chips generate profit for them. They are buying chips on the strength of a belief that AI will eventually generate profit, financing those purchases increasingly with debt, and racing one another to spend before the demand they are betting on has fully arrived. Nvidia's revenue, in other words, is not a diversified stream of customers meeting present need. It is a concentrated bet, by a few players, on a future — and Nvidia has booked that bet as present-tense sales.

When two-thirds of your revenue comes from three customers, your business is only as stable as those three customers' willingness to keep spending at the current, frantic pace. The moment any one of them — pressured by its own shareholders, its own cash flows, its own dawning doubts about AI returns — decides to slow its capital expenditure, a double-digit slice of the world's most valuable company's revenue is at risk. The concentration is not a footnote. It is the whole risk, hiding in a 10-K.

The second fact: the circle

The second foundation is stranger, and it is the one that turns a concentration risk into something closer to a hall of mirrors. Because Nvidia does not merely sell to its largest customers. It increasingly invests in them.

Consider the arrangements. In 2025, Nvidia and OpenAI announced, with enormous fanfare, a partnership under which Nvidia would invest as much as $100 billion in OpenAI to help deploy ten gigawatts of Nvidia systems — a headline number that helped cement the narrative of bottomless AI demand. Then the number quietly shrank. By early 2026 the "$100 billion" had become a $30 billion equity stake, part of a larger OpenAI funding round, and Jensen Huang was on the record clarifying that the hundred-billion figure had "never been a commitment" — a letter of intent, an aspiration, a press release. Separately, Nvidia invested $2 billion in CoreWeave at $87.20 a share, accumulating a stake of around 11% in the debt-laden cloud company, and agreed, as CoreWeave's own filings reveal, to buy back up to $6.3 billion of CoreWeave's unsold computing capacity through 2032. Nvidia has built a sprawling portfolio of investments in the very companies that are its largest customers.

Trace the dollar and the problem becomes vivid. Nvidia invests billions in OpenAI and CoreWeave. Those companies use capital — capital that, directly or indirectly, Nvidia helped provide — to buy Nvidia chips. Nvidia books that purchase as revenue, at a 71% gross margin, and the revenue growth drives its stock to $5 trillion, which makes its balance sheet strong enough to invest still more in its customers. Critics have a blunt name for arrangements in which a company funds its own customers' purchases of its own products: round-tripping, or vendor financing, the same maneuver that flattered the revenues of telecom-equipment makers like Lucent and Nortel right up until the dot-com collapse, when the customers they had financed went bankrupt and the "demand" evaporated because it had never been independent demand at all. No one is alleging Nvidia is doing anything illegal — these investments are disclosed, and the demand for AI chips is, unlike some historical analogues, substantially real. But the structure is unmistakable, and it has a specific consequence: some unknown portion of Nvidia's record revenue is demand that Nvidia itself has underwritten. The boom is, in part, buying from itself with its own money, and reporting the proceeds as growth.

The shrinking of the OpenAI pledge from $100 billion to $30 billion is the tell within the tell. The grandest demand commitment in the AI story turned out, on inspection, to be a fraction of its headline and "never a commitment" at all. If the most-publicized number in the circle was that soft, it is worth asking how firm the rest of the structure is.

Cisco's ghost

If the circular financing feels familiar, that is because one of the most successful investors of the modern era has staked more than a billion dollars on the proposition that it is.

Michael Burry — the investor who saw the subprime mortgage collapse before almost anyone and was immortalized for it in The Big Short — has built a position of bearish put options on Nvidia reported at around $1.1 billion in notional value, and he has been unusually explicit about the historical template he believes is repeating. The template is not Enron, and it is not fraud. It is Cisco Systems at the peak of the dot-com bubble. Between 1995 and 2000, Cisco's stock rose roughly 3,800%, and at the summit of the internet mania it became, exactly as Nvidia is now, the most valuable company in the world, worth around $560 billion — the indispensable "picks and shovels" supplier whose routers and switches built the infrastructure of the internet, owned by virtually every fund and adored by virtually every analyst. "Once again," Burry has written, "there is a Cisco at the center of it all, with the picks and shovels for all and the expansive vision to go with it."

What happened next is the part the analogy is built to deliver. Cisco's business was real, its technology genuine, its dominance unquestioned — and its stock still fell more than 80% when the demand it had been selling into turned out to have been financed by a wave of telecom and dot-com customers who promptly went bankrupt. A quarter of a century later, Cisco's share price has never returned to its 2000 peak. The company did not fail. The stock failed, because the price had embedded a permanence of demand that the underlying customers could not sustain once the financing stopped — and a great deal of that demand had been propped up, in Cisco's case as in Nvidia's, by vendor financing: the supplier lending its customers the money to buy its own products.

Burry's specific objections map directly onto the two facts of this essay. He has compared Nvidia's roughly $95 billion of purchase commitments to the obligations that flattered Cisco's books before the fall, declaring flatly: "This is not business as usual. This is risk." And he has built an entire secondary thesis around depreciation — the contention that the AI buyers stretching the useful lives of their Nvidia chips from three years to five or six are systematically overstating their profits, by an amount he estimates could reach $176 billion across 2026 to 2028, and understating how fast the hardware actually goes obsolete. If he is right, the customers' AI investments are less profitable than they appear, which makes their continued frantic buying less sustainable than it looks, which puts Nvidia's concentrated, partly-self-financed revenue at exactly the kind of risk the Cisco ghost embodies: a demand cliff, masked until the moment it isn't.

Burry has been early before, and early has often looked, for long stretches, exactly like wrong; he was mocked for years before the housing thesis paid off, and he may be mocked for years over this one, or forever. The bulls have a strong rebuttal — that AI demand is more real, more profitable, and more durable than dot-com bandwidth ever was, and that Nvidia's margins prove it. They may be right. But the value of the Cisco comparison is not as a prophecy. It is as a reminder that "the most valuable company in the world, the indispensable picks-and-shovels supplier owned by everyone, whose customers it helps to finance" is a description that has occurred before, in almost identical language, at almost exactly this point in a technology cycle — and that the last time, the company survived and the shareholders did not.

The keystone problem

Now combine the two facts with the one we started on — that Nvidia is the single largest holding in the global financial system — and you arrive at why this company, specifically, is the one to watch.

In an ordinary stock, concentration risk and vendor financing would be the shareholders' problem, contained within the share price, a matter for those who chose to own it. Nvidia is not an ordinary stock. It is the load-bearing wall of the index. Through the mechanics of passive investing, hundreds of millions of people who have never analyzed Nvidia's customer concentration own enormous quantities of it anyway, because it is the biggest thing in the fund. Its $5 trillion valuation is not just Nvidia's valuation; it is a giant, leveraged-by-indexing chunk of the retirement savings of the Western world. And that valuation rests on the continued, accelerating, debt-financed capital spending of three customers, some of whose demand Nvidia is itself underwriting.

The arithmetic of that transmission is worth making concrete. Nvidia alone is on the order of 8% of the entire S&P 500 — a single company outweighing whole sectors. A 30% decline in Nvidia, of the kind that would be unremarkable for a volatile growth stock and which Nvidia has briefly approached more than once, would by itself knock roughly two and a half percent off the index before a single other stock moved. But no other stock would stay still: the same three or four hyperscaler customers that drive Nvidia's revenue are themselves among the largest companies in the index, their valuations underwritten by the same AI thesis, their stocks correlated to Nvidia's on every AI headline. A genuine crack in the keystone would not subtract one company from the index. It would re-rate the entire AI complex — the chipmakers, the hyperscalers, the power utilities, the credit — all at once, in the same direction, because they are all, ultimately, the same trade wearing different tickers. Concentration is not just a Nvidia problem or an index problem. It is a correlation problem, and correlations go to one precisely when you least want them to.

This is what it means to call Nvidia patient zero. The AI trade is one organism, and Nvidia is its heart. Every other story in this market — the hyperscalers borrowing $150 billion in bonds, Oracle staking its balance sheet on OpenAI, CoreWeave's founders selling stock, the credit-default-swap market exploding, SoftBank's leveraged bet, the Korean memory mania — is ultimately a story about the demand for Nvidia's chips and the belief that it will continue. If that belief holds, Nvidia grows into and beyond its valuation and this essay ages badly. But if it ever breaks — if one of the three customers blinks, if the AI returns disappoint, if the circular financing is revealed to have been propping up a slice of the demand — the contagion does not stay contained, because the patient is wired into everyone. The same concentration that has made the index soar on the way up is the mechanism by which a Nvidia stumble would be transmitted, instantly and with leverage, into the savings of people who could not tell you what a GPU is.

None of this is a forecast that Nvidia falls tomorrow, or this year, or that it falls at all. It is the best business of its generation, run by one of the great operators of the age, riding a wave of demand that is more real than any bubble sceptic wants to admit. The bull case may simply keep being right, as it has been, humiliatingly, for everyone who doubted it so far. But "the best business, at the center of everything, that everyone owns, whose revenue comes from three customers it is partly financing" is not a description of safety. It is a description of magnificent, world-spanning concentration — the kind that looks like strength in every year but one.

The market has decided that Nvidia is the surest thing it has ever seen, and has therefore made it the largest single bet humanity has ever collectively placed. Those two facts are in tension, and the tension is the whole story. When the most certain thing becomes the most owned thing, certainty stops being a comfort and becomes the risk itself. Patient zero is not sick. It has never looked healthier. That is exactly what makes the temperature worth taking — because everyone in the room is breathing the same air, and no one has thought to ask what happens if the strongest one among them ever coughs.