It was reported this weekend that Nvidia (NASDAQ:) is discussing guaranteeing up to $250 billion of OpenAI’s lease on a 10-gigawatt, $500 billion Ohio data center targeted for 2028. Separately, Nvidia is discussing $350 billion in financing for OpenAI’s chip purchases. Layer on a $500-plus billion SK Hynix buildout announced last week, and Nvidia’s already-announced $540 billion in prior deals, and the aggregate number exceeds $750 billion in AI infrastructure commitments currently in motion.
On the surface, this is straightforwardly bullish. Nvidia providing funding means AI infrastructure spending continues without a financing hiccup. Semiconductor suppliers still benefit from massive demand at favorable pricing. Revenue and profit growth for the picks-and-shovels trade continues.
But Nvidia’s 5-year credit default swap (CDS) jumped the most on record today. The market is pricing a contingent liability — a $250 billion guarantee is exactly the kind of tail risk CDS are designed to price. And underneath that CDS move is a deeper fear: Nvidia is now financing its own biggest customer, which makes it impossible to cleanly separate organic demand from manufactured demand. Jensen Huang calls the circularity concern ’ridiculous.’ But the dollar figures just got big enough that dismissing it is no longer sufficient.
The comparison that is spooking investors is the Dot Com infrastructure buildout: Lucent and Nortel financing their own customers’ equipment purchases in the late 1990s, which inflated reported demand, then blew up spectacularly when those customers could not service the debt and there turned out to be vastly more capacity than anyone needed. Dark fiber sat idle for a decade. WorldCom declared the largest bankruptcy in U.S. history. The fear today is the same pattern at a larger scale — the world’s most important AI companies simultaneously levering up massively on a single bet that AI becomes indispensable infrastructure within five to ten years.
That fear is fair to articulate. It is not fair to accept as the base case without examining the specific differences.
Five Reasons the Circular Financing Fear Is Overblown
The Dot Com parallel is sharp in some ways and specifically wrong in the ways that matter most. Here are the five distinctions that determine whether the fear is legitimate or overblown.
First: there is no distribution constraint this time. In 1999, internet adoption was capped by physical rollout — modems, PCs, broadband infrastructure that reached only half of American homes. AI applications face none of that. When OpenAI or Anthropic ship a new model capability, it arrives on more than a billion smartphones within hours. The bottleneck that made telecom demand arrive too late for its debt simply does not exist in 2026.
Second: monetization is already catching up. Agentic AI and usage-based pricing are capturing value that flat subscription pricing cannot — the same structural shift that allowed cloud computing to out-monetize boxed software. Enterprise AI spending is demonstrably accelerating, as ’s 82% Cloud growth confirmed last week. This is not hypothetical future monetization. It is happening now.
Third: the chip layer is showing scarcity, not glut. is raising prices by up to 10% — and some reports suggest as high as 25% in certain categories — starting in 2027, with Vanguard and UMC following suit. Prices rise in conditions of supply scarcity. The dot-com disaster featured bandwidth prices collapsing as dark fiber supply overwhelmed demand. The AI chip market today is showing the exact opposite pricing dynamic.
Fourth: GPU clusters are reusable, not stranded. The dark fiber of the Dot Com era sat idle for years because no one wanted it in the near term. A GPU cluster built for OpenAI’s current workloads has instant next-best-use demand from every compute-starved frontier lab, enterprise, and sovereign AI program on earth. The asset is not at risk of being stranded.
Fifth: Nvidia is confident, not desperate. Lucent and Nortel financed their customers because they needed to move product — they were pushing against weakening demand and used vendor financing to manufacture the appearance of healthy order books. Nvidia does not need this deal to move GPUs. It is backstopping OpenAI because it believes the math works — because the company with the best visibility into AI compute demand has concluded that OpenAI’s capacity expansion will generate sufficient returns to service the financing. That is a signal from the most informed party in the chain.
The stakes are genuinely high, and the market is right to take them seriously. But when specific fear runs into a specific intact thesis, the fear usually breaks first.
