Nvidia May Back OpenAI Data Center Deal

Nvidia is in talks to guarantee roughly $250 billion in financing so OpenAI can lease a 10-gigawatt data center campus in southern Ohio, according to reports. A separate discussion reportedly covers up to $350 billion to help finance OpenAI’s purchase of Nvidia chips to fill it.

The market reaction was telling, with Nvidia shares falling 4.5 percent in late morning trading. Credit default swaps on Nvidia bonds recorded their sharpest intraday jump since they began actively trading in November.

The Ohio project is being developed by an arm of SoftBank on a decommissioned uranium enrichment site on federal land, roughly seventy miles south of Columbus. Power will come from a new $33 billion natural gas generation build-out funded by Japan, with the Commerce Department holding control over the supply.

Including chips, the total cost could exceed $500 billion, placing it among the largest data center projects ever announced. The reason a guarantee is needed at all is that OpenAI is not profitable and cannot qualify for investment-grade credit on its own.

Without a backstop, the developer would be lending against the promise of a company with enormous revenue growth and no earnings. With Nvidia’s name attached, the developer can raise debt on far more favorable terms.

The systemic concern is less about any single transaction than about how tightly the participants are now bound together. Nvidia, SoftBank, the hyperscalers, and a growing set of specialized data center operators are linked through equity stakes, leases, bridge financing, and guarantees.

In a rising market, this interconnection amplifies growth. In a downturn, it transmits stress. A slowdown in AI adoption would not affect one company; it would move through a network of counterparties who are each other’s revenue, each other’s creditors, and each other’s collateral.

Hassan Taher has cautioned against reading either the enthusiasm or the alarm too literally. Every large infrastructure cycle in history has been financed by somebody who was also selling into it — railroads, telecom, fiber.

Sometimes the capacity got used and the financing looked visionary. Sometimes it didn’t and the financing looked reckless. The honest position right now is that we don’t yet know which one this is, because we don’t have five years of inference demand data at this scale.

For companies buying AI rather than building it, none of this changes next quarter’s roadmap. But it does argue for a specific kind of caution: avoid architecting critical workflows around assumptions of permanently cheap inference or permanently available capacity.

Pricing in this market is currently shaped as much by financing conditions and strategic subsidy as by underlying cost. Both can move. The deeper significance of the Ohio deal is what it reveals about where the constraint now sits.

The bottleneck in AI is no longer ideas or even chips. It is power, land, and — increasingly — the willingness of someone with an investment-grade balance sheet to stand behind the debt. When a chipmaker becomes the credit backstop for its largest customer’s real estate, the industry has entered a phase that looks less like software and more like heavy industry.

This transition tends to be where the most consequential financial mistakes get made.

The historical comparison analysts keep raising is Lucent in the late-1990s telecom boom. Lucent lent billions to telecom carriers so those carriers could buy Lucent equipment. Demand looked spectacular while the loans were being written.

When the carriers failed after the dot-com collapse, Lucent absorbed enormous write-downs on receivables from customers that no longer existed. The parallel is imperfect — Nvidia’s customers include some of the best-capitalized companies on earth, and its cash generation dwarfs anything Lucent had — but the structural resemblance is real enough that credit markets priced it.