Jensen Huang Calls GPUs "Long-Lived." He Ships New Ones Every Year.

When Jensen Huang set out to convince Wall Street that GPU clusters deserve the same credit treatment as office towers and power plants, he chose his words with the precision of a man who knows exactly what underwriters need to hear. "These are revenue-generating assets now," he said, announcing a $500 billion financing partnership with six of the world's largest capital providers. "They're productive, they're long lived, they're fungible, they're flexible." ^[1]

These are revenue-generating assets now. They're productive, they're long lived, they're fungible, they're flexible. — Jensen Huang

Every instrument Nvidia has since constructed rests on that claim. If GPUs are durable, income-producing assets, they can collateralize loans. If they are fungible, a lender can seize and redeploy them. If they are flexible, their value persists across use cases. The claim is not rhetorical — it is the legal and financial premise of a system that now mobilizes hundreds of billions of dollars in third-party capital for AI infrastructure. That system spans a $500 billion debt syndication with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, treating compute power as the collateral standard for institutional credit. ^[2][1] A $100 billion vendor-financing deal with OpenAI — cash for chips, then that same cash reinvested as equity — drew comparisons to the telecom bubble before the ink was dry. ^[3] A $6.3 billion capacity backstop guarantees CoreWeave's GPU utilization through April 2032, making Nvidia the insurer of last resort for the largest GPU neocloud. ^[4] A revenue-sharing platform lets startups acquire GPU infrastructure without upfront capital in exchange for a cut of future product sales. And a $90 billion strategic investment portfolio, spanning 33 companies, was assembled explicitly to "tie a larger portion of the AI economy to its own technology." ^[5] Each instrument is a different mechanism for the same outcome: making GPU compute function as a credit asset, and making Nvidia the hub through which that credit flows. The market has begun accepting the standard on its own terms, independent of Nvidia's programs. In March, CoreWeave closed an $8.5 billion investment-grade GPU-backed term loan — rated A3/A-low — at rates 7.5 percentage points below its 2023 financing. ^[6] In April, Alpha Compute Corp secured a $31.9 million non-recourse loan secured entirely by pledged Dell B300 Nvidia GPUs; the lender's only remedy in default is the hardware itself. ^[7] A GPU is now, in the eyes of institutional credit markets, a thing you can borrow against. The paradox is that the same company that convinced the market to treat GPUs as durable collateral also makes them obsolete on a twelve-month clock. Nvidia's product cycle — Blackwell to Vera Rubin to the next generation — is the most aggressive in the industry's history. And the largest GPU neocloud, the one whose balance sheet most directly tests Huang's "long-lived" claim, is burning cash to keep up. CoreWeave "must perform frequent and costly hardware upgrades to remain competitive as Nvidia continues to release new products." ^[8] The company burned $8 billion in free cash flow over twelve months, spending double its revenue on capital expenditures. It carries $14 billion in total debt against $3.56 billion in revenue. ^[9] Despite 204% revenue growth, it posted a $290.5 million quarterly loss. ^[4] The financial distress is not a failure of execution. It is a feature of the asset class: the very innovation that makes Nvidia's chips indispensable is what erodes their value as a store of value. A GPU is not an office building. It does not appreciate, or even hold steady. It depreciates the moment the next generation ships, and the next generation ships every year. This is the contradiction Huang cannot resolve by assertion. He has called the idea that Nvidia's financing is circular "ridiculous." ^[10]

the idea that it is circular is, it’s ridiculous. — Nvidia

And in nearly the same breath, he has claimed Nvidia is "basically holding the planet together." ^[10]

We’re basically holding the planet together—and it’s not untrue. — Jensen Huang

The two statements sit uneasily beside each other. To deny circularity is to say the financing is incidental — that the chips would sell regardless. To claim the planet depends on Nvidia is to say the entire AI buildout cannot proceed without it. But Nvidia is not merely the chip supplier. It is the residual-value guarantor taking up to 25% of each project's risk, assessed "on a project-by-project basis" and "designed to complement — not replace — independent underwriting." ^[11] It is the capacity backstop that allows CoreWeave to borrow. It is the largest equity holder in its own primary distribution channel, with $4.3 billion concentrated in CoreWeave alone. ^[12] It is, simultaneously, "customer, supplier, and prospective shareholder" to the firms buying its chips. ^[5] The financing is not incidental to the chip business. It is the mechanism by which the chip business sustains its own demand. Whether that mechanism is virtuous or circular turns on a question the evidence splits down the middle. On one side, the demand is real and overwhelming: a 12-gigawatt global data center capacity deficit, with 8.9 GW operational against 21.1 GW of demand, and $2 trillion in cloud backlogs. ^[13] Nvidia's revenue grew 92% year over year to $75.2 billion in data center sales in a single quarter, and the company generated $49 billion in free cash flow — enough to authorize an $80 billion buyback while serving as residual-value guarantor for a $500 billion infrastructure program. ^[14][15] Cloud revenue is accelerating: Google Cloud up 82%, Azure up 43%. Old-generation GPU prices are, against all expectations, still rising. On the other side, the buyers are spending money they do not have. Alphabet posted its first-ever negative free cash flow — negative $5.9 billion — with 2026 capital expenditure guidance raised to $205 billion. Meta's free cash flow fell 91% to $784 million. Five hyperscalers have spent $412 billion on AI infrastructure, and when that capital expenditure begins depreciating on income statements, earnings will compress — potentially triggering the spending pullback that would undermine the very collateral values the financing depends on. ^[16] Michael Burry has shorted Nvidia, calling the buildout a "speculative mania comparable to the Dot-Com bubble." ^[10][15] GMO's Tom Hancock called vendor financing's history "pretty spotty" and "a feature of past bubbles." ^[3] The 12-gigawatt deficit is the fact that will resolve the fork. If the capacity shortage reflects genuine end-demand — enterprises and consumers consuming AI compute faster than the world can build data centers — then Nvidia's financial architecture is a bridge across a real chasm, and the depreciation paradox is manageable: old chips lose value, but the revenue they generate before obsolescence justifies the credit extended against them. If the deficit is instead a function of hyperscalers buying infrastructure with money Nvidia is financing — spending they could not sustain without the very credit apparatus that depends on their continued spending — then the system is circular in the precise sense Huang denies. The 12 GW number does not, on its own, distinguish between these two possibilities. It only measures the gap between supply and expressed demand. It does not tell you whether the demand is solvent. Huang's denial and his assertion of centrality are two halves of a single bet: that the demand is real, that the revenue will materialize, that the collateral will hold its value long enough for the loans to be repaid. The financial system Nvidia has built is already priced as if both halves are true. The 12-gigawatt deficit will prove one of them right and the other wrong — but the system is already running on the assumption that they can both be true at once.