Nvidia Lends OpenAI the Money to Buy Nvidia Chips

Michael Burry saw the structure and posted three words: 'Around and around we go.'

That is the cleanest summary of what Nvidia is reportedly negotiating with OpenAI. According to the Wall Street Journal, Nvidia is in talks to guarantee as much as $250 billion in financing to help OpenAI lease a 10-gigawatt data center campus on federal land in southern Ohio. The project is expected to cost more than $500 billion in total, including the chips inside the data center. The proposed guarantee would help OpenAI secure favorable financing despite the ChatGPT maker lacking an investment-grade credit rating.

Read that again. OpenAI has no investment-grade rating. The conventional debt market said no. So the chip supplier stepped up to backstop the lease so the customer can afford to buy the chip supplier's chips.

Translation: we will keep selling GPUs until the customer runs out of money, and we will lend them money so they do not.

The Loop Has a Name

The $250 billion guarantee covers the data center lease and debt financing, but not the Nvidia chips inside the center — Nvidia was also discussing financing OpenAI's chip purchases worth up to $350 billion. That is not a vendor relationship. That is a bank that only lends to one customer, who can only spend the money at the bank.

The structure is circular: Nvidia guarantees OpenAI's ability to pay for Nvidia's own hardware, exposing a financing gap that conventional debt markets have not filled on their own. Moody's flagged this dynamic three days ago. Moody's pointed to structural circularity within the AI boom, noting that multibillion-dollar backlogs reported by hyperscalers stem from strategic deals with pre-IPO AI labs including OpenAI and Anthropic.

The credit market already knows how this unwinds. On July 9, 2026, S&P Global Ratings lowered Oracle's long-term issuer credit rating from BBB to BBB- with a stable outlook. S&P characterized OpenAI as a 'key credit risk' for Oracle, noting that of Oracle's total contractual obligations amounting to $638 billion, OpenAI alone accounts for approximately half. Oracle got punished for concentration risk to a customer that has never turned a profit. Nvidia is now proposing to take on a structurally identical position, but larger and without even the pretense of being a separate company.

What Piketon Actually Is

The U.S. Department of Energy announced a public-private partnership to develop a major data center on a decommissioned uranium enrichment plant in southern Ohio. The Portsmouth Gaseous Diffusion Plant in Pike County — branded as the 'PORTS Technology Campus' — is expected to include a 10-gigawatt data center and up to 10 gigawatts of new power generation, including 9.2 gigawatts of natural gas generation.

The power is controlled by the U.S. government and funded separately by Japan under a recent trade deal. Commerce Secretary Howard Lutnick is involved in deciding who gets access. Power allocation at this scale is a political decision, not a market one. OpenAI is in advanced talks to lease the site and shows the strongest interest, while Anthropic, Microsoft and Google have also spoken to Lutnick in recent weeks. When the Commerce Secretary personally controls who gets the electrons, the deal depends on a person, not a contract.

The reported deal is structured as a 20-year lease, with OpenAI controlling the equipment at the facility and payments beginning only once operations start. The first phase is expected to deliver 800 megawatts of capacity in 2028. Full buildout takes at least a decade. Nvidia is being asked to guarantee a commitment whose revenue test does not arrive until the early 2030s, against a customer whose monetization at scale remains unproven.

Negotiations are in their early stages and could collapse or financing terms may change. That caveat belongs in the headline, not the footnote.

What You Need to Decide Before Tomorrow

If this structure closes, every AI lab asks its primary chip supplier for the same terms. Nvidia faces a choice between extending more guarantees to preserve market share or watching AMD and custom silicon close the gap on accounts it declines to finance.

For Nvidia shareholders, the question is when a $250 billion contingent guarantee shows up as a liability in an SEC filing. That is not hypothetical. That is a line on a balance sheet that does not exist yet in public disclosure.

For OpenAI competitors, this deal forecloses the company as a customer for any rival silicon for the duration of a 20-year lease. You do not switch chip vendors when your landlord's guarantee is the chip vendor's balance sheet.

For operators with board exposure to AI infrastructure capex: the credit market has already priced the downside on Oracle's OpenAI concentration. Nvidia is now proposing to take on a structurally identical position, one order of magnitude larger.

The loop is simple. The number that matters is not $500 billion. It is whether Nvidia discloses this as a contingent liability before its next earnings call.

What to watch: (1) Whether Nvidia's next 10-Q or 8-K discloses this guarantee as a contingent liability and how it is sized. (2) Whether Lutnick awards power access to OpenAI or a competing bidder, and on what timeline. (3) Whether AMD, Intel, or a custom silicon provider offers competing financing to another AI lab, starting a vendor-financing arms race. (4) OpenAI's next public financials — any signal of monetization stall changes the risk math on this guarantee immediately.