Nvidia reported fourth-quarter revenue of $68.13 billion on February 26 — up 73% year-over-year, above estimates, with Data Center revenue of $62.3 billion. The stock fell more than 5%. Analysts attributed the decline to "investor concerns around the AI infrastructure boom's sustainability." But the concerns are more specific than that. Buried in Nvidia's filing was a disclosure: the company has $3.5 billion in guarantees to companies leasing land, power, and data center capacity for GPU infrastructure. Nvidia is no longer just selling the chips. It is guaranteeing the ecosystem that buys them.
Six Steps
The transformation happened in stages, each one pulling Nvidia deeper into the financial architecture of the AI economy. Trace it from the beginning.
Step 1: Seller. In 2023, Nvidia sold GPUs to CoreWeave, a cloud startup that offered Nvidia hardware on demand. A standard vendor relationship — chips for cash.
Step 2: Investor. In December 2023, Nvidia took a minority stake in CoreWeave. The vendor now had equity in its own customer. CoreWeave's success became Nvidia's return — not just through chip sales, but through ownership.
Step 3: Collateral. By November 2024, Blackstone, Pimco, Carlyle, and BlackRock had loaned $11 billion to CoreWeave and other data center operators — with Nvidia GPUs as collateral. The chips weren't just computing devices anymore. They were financial instruments, backing billions in debt the way real estate backs mortgages.
Step 4: Income. In August 2025, Nvidia's CFO revealed that Q2 "net other income" was $2.2 billion, "driven by" the GPU leasing ecosystem. Nvidia was now earning not just from selling hardware but from the financial infrastructure built on top of it.
Step 5: Lessor. In November 2025, an Nvidia filing showed the company planned to rent $26 billion worth of servers over the next several years. The chipmaker was becoming a landlord — leasing back the infrastructure its products powered.
Step 6: Guarantor. On February 26, 2026, Nvidia disclosed $3.5 billion in guarantees to GPU lessors. If those companies — the ones leasing land and power for AI data centers — can't meet their obligations, Nvidia pays. The chipmaker is now the backstop.
Each step was individually rational. Investing in a growing customer, collateralizing a scarce asset, earning from a new revenue stream, leasing infrastructure, backstopping partners. No single step was reckless. But the cumulative effect is that Nvidia now sits on both sides of the ledger: it is the supplier of the AI economy's critical input and the guarantor of the financial obligations built on top of that input.
The Circle
Follow the money through one cycle.
On February 26 — the same day as the Nvidia filing — The Information reported that Amazon plans to invest up to $50 billion in OpenAI. As much as $35 billion of that could take the form of Azure cloud credits. Azure runs on Nvidia GPUs. So Amazon's investment in OpenAI flows to Microsoft, which flows to Nvidia, which flows back to companies like CoreWeave, which borrow against Nvidia GPUs, which Nvidia guarantees.
CoreWeave itself is seeking an $8.5 billion loan from banks, backed by its customer contracts. Six days earlier, AMD agreed to backstop a $300 million loan for a GPU lessor. The chipmakers are now guaranteeing the loans that finance the purchases of their own products.
The Financial Times reported on February 26 that GPU-backed debt has become a widespread financing model, pioneered by CoreWeave and now adopted across the industry. Tech companies raised $157 billion in US debt markets in 2025. An increasing share of that debt is collateralized by GPU infrastructure whose value depends on sustained AI demand.
The Pattern
Michael Burry — the investor who shorted the housing market before the 2008 financial crisis — flagged the "troubling" jump in Nvidia's supply commitments after the February 26 filing. The parallel he sees is structural, not necessarily predictive: a market where the supplier of the underlying asset is also guaranteeing the debt built on top of it.
In 2006, banks sold mortgages, packaged them into securities, and guaranteed tranches of those securities through credit default swaps. The originators were also the backstops. When the underlying assets lost value, the guarantees amplified the losses rather than containing them.
Nvidia's situation is not identical. GPUs are physical assets with genuine productive value, not synthetic derivatives of consumer debt. The demand for AI compute is real — $68 billion in quarterly revenue proves that. But the structure rhymes: Nvidia sells the GPUs, earns income from the leasing ecosystem, invests equity in its largest GPU customers, and now guarantees the lease obligations of the infrastructure operators. If AI compute demand contracts, Nvidia's exposure isn't limited to lost chip sales. It extends to the guarantees, the equity stakes, the leases, and the $26 billion in planned server rentals.
The market isn't doubting Nvidia's revenue. It's pricing the distance between what Nvidia earns and what Nvidia has guaranteed.
Jensen's Answer
Nvidia CEO Jensen Huang's response to the sustainability question was direct: "Compute equals revenues." The logic is that every dollar a company spends on Nvidia GPUs generates multiple dollars in AI-enabled revenue, making the demand self-sustaining. This is the same argument every infrastructure boom makes — and sometimes it's correct. The railroads did generate more economic value than they cost. The fiber-optic cables laid in the 1990s did eventually carry the traffic they were built for.
But the railroads and the fiber-optic companies didn't also guarantee the loans taken out by the freight companies and the ISPs. They were suppliers. Nvidia has become something else — supplier, investor, lessor, and guarantor, all at once. The Financial Times headline on the same day's market action: "Tech stocks slide as AI spending fears return."
On February 16, Bloomberg reported that credit derivatives trading on specific tech companies continues to rise — a sign that sophisticated investors are beginning to hedge against the possibility that the AI infrastructure build-out produces the kind of concentrated financial risk that the 2026 numbers suggest.
The Disclosure
Nvidia's $3.5 billion in guarantees is small relative to its $68 billion quarter. The number matters not for its size but for what it reveals about Nvidia's role. A company that sells products does not need to guarantee its customers' real estate leases. A company that underwrites an ecosystem does.
Three years ago, Nvidia sold chips. Today it sells chips, invests in the companies that buy them, earns income from the financing structures built around them, plans to lease back the servers they power, and guarantees the infrastructure obligations of the companies that operate them. The stock fell on record earnings because the market has started to read the filing the way Burry reads a filing — not for what the revenue shows, but for what the guarantees imply.
The revenue says $68 billion. The guarantee says Nvidia knows that number depends on an ecosystem it can no longer merely supply. It has to sustain.