01  Setting the Scene

Nvidia makes the chips nearly every AI company needs. Building the data centers that run those chips is enormously expensive. The Ohio hub OpenAI is negotiating to lease, for one example, is valued at roughly $500 billion for 10 gigawatts of capacity, far more than most AI companies actually earn. So increasingly, Nvidia is supplying the money too: taking equity stakes, guaranteeing loans, backstopping unused computing capacity. The company on the receiving end then uses that money, or the credibility it buys, to purchase Nvidia’s chips.

Vendor financing, suppliers bankrolling the customers who buy from them, is an old practice. Telecom equipment makers did it in the 1990s. Planemakers still do it. Nvidia’s version is distinguished less by the concept than by the arithmetic: the company has generated close to $200 billion in cash over the past two years and is now investor, lender, and supplier to the same tight cluster of firms, at a scale with no real precedent.

02  The Money Trail

A few of the largest recent commitments, laid out plainly. Some remain under negotiation and are not yet finalized.

Nvidia’s investment book, carrying value ($B)

03  Where the Alarm Is Coming From

Michael Burry, the investor best known for having predicted the 2008 housing crash, argues that a real share of Nvidia’s headline revenue is financed rather than earned. That worry now shows up in the price of Nvidia’s own debt: the cost of insuring it against default hit a record in late July 2026.

The concern has since jumped several pay grades. Both the Bank for International Settlements and the IMF have named this a financial-stability problem, not a Nvidia problem. Their argument: when a small cluster of companies hold stakes in each other and commit to buying from each other, nobody outside the loop can check whether the numbers are real.

04  What Tempers the Concern

That’s a serious claim, and it deserves a serious hearing. But it’s also worth being precise about what “circular” actually means here, and what it doesn’t. Every one of these deals leaves Nvidia holding something real: equity, a contract, a licensing arrangement, rather than revenue printed out of nothing. Financial Times’ Lex column has called this “back-to-back” rather than truly circular, and that’s the more accurate description. Nvidia’s balance sheet genuinely can absorb a string of bad bets without breaking.

The clearest evidence yet for that argument arrived on August 10, when Nvidia signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to build financing platforms aimed at mobilizing more than $500 billion in third-party capital for AI infrastructure. This is structurally different from the deals earlier in this piece. Instead of Nvidia’s own balance sheet standing behind its customers, outside institutional capital does. If genuinely independent money is willing to fund AI infrastructure at this scale, that’s a real signal about how the broader market is pricing the underlying demand, not just what Nvidia itself believes about it.

CoreWeave’s cost of capital, contract-backed vs. unbacked debt

The market seems to believe some of this too. Debt backed by real contracts borrows more cheaply than debt without that backing, a sign investors think there’s real value here rather than financial sleight of hand.

05  The Question That Actually Matters

Here’s what the “is it circular or isn’t it” argument distracts from: how much of Nvidia’s own capital is now betting on AI demand staying enormous, and what happens to Nvidia specifically if that bet is wrong? That’s where the two forms of support Nvidia offers stop being interchangeable.

As Nvidia moves from taking stakes toward writing bigger guarantees, it’s taking on a fundamentally different kind of exposure, one that converts “we backed a loser” into “we owe cash we don’t have a matching asset for.” The other honest thing to say here is that nobody, including Burry, the BIS, or Nvidia itself, can currently prove how much of this demand is real versus financed into existence. That’s just what uncertainty looks like this early in a buildout of this size, not a sign anyone’s being dishonest. When telecom vendor financing collapsed in the 1990s, the cause was demand assumptions turning out wrong, and the vendors who’d financed that demand were the ones left holding it, not fraud on anyone’s part.

06  The Bottom Line

“Circular financing” isn’t quite the right term for what’s happening. The revenue and assets behind these deals are real, and Nvidia’s balance sheet can absorb the current scale of commitments without a near-term solvency scare. What matters more is the direction of travel: Nvidia is gradually shifting from taking equity stakes toward writing larger credit guarantees, and those two forms of support carry genuinely different risk. An equity stake is a bounded loss if it goes wrong. A guarantee is a liability regardless of the outcome.

That shift is a structural change in how much of Nvidia’s own capital is riding on AI demand staying as strong as it currently looks. It’s worth watching as the guarantees grow, both in Nvidia’s own disclosures and in how its lenders and credit markets price the risk, well short of the kind of solvency concern or bursting-bubble scenario some of the more alarmist coverage implies.

For individual investors, this is mainly a reminder that concentrated bets on a small, interlinked group of companies carry risks that don’t always show up in a single quarter’s numbers, the kind of risk diversification exists to handle, rather than a signal to exit AI positions outright. On our side, we’re staying aware of this potential risk and will adjust client portfolios when and if we deem it appropriate.

07  Glossary