Nvidia's $250B Backstop for OpenAI's Ohio Data Center

The AI buildout keeps finding new ways to move eye-watering sums of money in a circle. Nvidia is in talks to provide a roughly $250 billion financial backstop for OpenAI, according to reporting from The Wall Street Journal, guaranteeing the debt behind a planned 10-gigawatt data-center campus in southern Ohio. The arrangement would function as a credit enhancement — Nvidia standing behind the borrowing rather than writing a check — and it is designed to help OpenAI secure favorable terms despite the ChatGPT maker’s lack of an investment-grade credit rating.

The report has not been independently confirmed, and the companies have not publicly detailed the structure. But the sheer size of the figures, and what they say about how AI infrastructure is now being financed, made it the market’s dominant story heading into a heavy week of tech earnings.

What Nvidia would actually be signing

The distinction between an investment and a backstop matters here. In the reported arrangement, Nvidia would guarantee financing tied to the data-center lease and the construction debt — reducing the risk lenders take on, and therefore the interest rate OpenAI would pay. It is closer to a co-signer than a shareholder.

That structure is the point. OpenAI is spending at a scale no operating cash flow can cover, and it does not carry the kind of credit rating that would let it borrow hundreds of billions on its own name at reasonable rates. A guarantee from Nvidia — one of the most valuable and cash-rich companies on the planet — transforms the economics of the loan. Lenders are effectively underwriting Nvidia’s balance sheet, not OpenAI’s.

Separately, the report says Nvidia is discussing financing for chip purchases that could total another roughly $350 billion — the GPUs that would actually fill the Ohio campus. Stack the two together and Nvidia is potentially standing behind well over half a trillion dollars of spending, much of which flows back to Nvidia as revenue.

The Ohio campus

The site itself is enormous. The campus is planned for Piketon, in southern Ohio, and could ultimately cost more than $500 billion once chips, construction, power, and labor are counted. At 10 gigawatts, it would rank among the largest AI-focused facilities ever announced — roughly the output of ten large power plants dedicated to a single company’s compute.

The project is being developed by SoftBank Group’s energy subsidiary, and the first phase is expected to deliver about 800 megawatts by 2028. That phasing is a reminder that even with financing lined up, capacity of this size arrives in slices over years, gated by the availability of power, transformers, and the physical build.

OpenAI is not the only interested tenant. Microsoft, Google, and Anthropic have also reportedly expressed interest in the site — a sign that developable, power-connected land at this scale has become one of the scarcest assets in the industry. This buildout sits alongside OpenAI’s other mega-projects, including its Georgia data-center campus and the broader 10-gigawatt partnership with Nvidia unveiled earlier this year.

”Around and around we go”

The reaction that traveled fastest came from investor Michael Burry, who reacted on X: “Around and around we go. Nvidia to guarantee $200 billion of ChatGPT’s spending on $NVDA chips.” His point is the one now shadowing every big AI-infrastructure headline: circularity.

The concern runs like this. Nvidia’s customers need capital to buy Nvidia’s chips. Nvidia helps supply or guarantee that capital. The capital is spent on Nvidia’s chips, which shows up as Nvidia’s revenue, which supports Nvidia’s valuation and its ability to extend more financing. Critics argue that when a chip vendor becomes the backstop for its own demand, reported growth can outrun genuine, independent end-demand — and the risk that would normally sit with diversified lenders concentrates back onto the vendor.

Supporters counter that vendor financing is old and ordinary — aircraft makers, telecom-equipment suppliers, and enterprise-hardware giants have all done it for decades — and that guaranteeing debt for a customer whose product is in obvious demand is a rational way to accelerate a market. The truth is that both can be right at once: the demand can be real and the financing structure can amplify fragility if the demand ever softens.

The market backdrop

The timing sharpens the debate. The report landed during a jittery stretch for AI-exposed stocks. The PHLX Semiconductor Index (SOX) has fallen roughly 18% in July, on track for its worst month in years, after nearly doubling in the first half of 2026. Investors are openly questioning whether the roughly $725 billion in combined AI capital spending planned by the largest tech companies this year will generate returns fast enough to justify it — a question that will be front and center as Microsoft, Meta, Apple, and Amazon report earnings this week.

Not every stock fell on the news. Oracle shares jumped in overnight trading, as investors read a giant new OpenAI campus as more demand for the cloud and infrastructure partners tied to it. That split reaction captures the moment: the same headline reads as opportunity to some and as a warning sign to others.

This is also of a piece with how AI’s largest players are now funding themselves. OpenAI has separately mapped out compute spending approaching $750 billion by 2030, and rivals have leaned on exotic financing of their own — from Amazon’s $25 billion AI bond sale to the surging valuations captured in Anthropic’s secondary-market pricing. The capital structure of the AI boom has become as important a story as the models themselves.

What it means

A $250 billion guarantee, if it closes as reported, would be one of the most consequential financing moves of the entire AI cycle — and a template others will copy.

Who wins. OpenAI wins access to cheaper, larger capital than its own credit could command, letting it lock up scarce power and land ahead of rivals. Nvidia wins a mechanism to convert its balance-sheet strength into durable demand for its chips, pulling forward revenue that might otherwise wait on customers’ ability to borrow. SoftBank and infrastructure partners like Oracle win as the tenants and builders of the physical campus.

Who carries the risk. The risk migrates onto Nvidia. A guarantee is a contingent liability: it costs little unless something goes wrong, but if AI demand disappoints or OpenAI’s economics falter, Nvidia could be on the hook for debt tied to a half-built campus. That is precisely why circular-financing critiques land — the entity underwriting the loans is the same entity selling the product the loans pay for.

What to watch next. First, confirmation and structure — whether the deal is announced, and whether the guarantee is a hard obligation or a softer letter of support. Second, the accounting: how Nvidia discloses the contingent liability, and how the chip-financing piece is booked as revenue versus vendor credit. Third, the ratings agencies and lenders — how they price debt backed by a chipmaker’s guarantee rather than the borrower’s own cash flows. And fourth, the read-through to every other AI lab: if a vendor backstop becomes the standard way to fund gigawatt-scale campuses, the industry’s growth and its fragility will increasingly rest on the same few balance sheets. For now, the deal is reported, not signed — but the direction of travel is unmistakable.