Nvidia and OpenAI discuss a $250B AI backstop

OpenAI is discussing a $250 billion Nvidia backstop to finance a huge Ohio data center campus.

OpenAI and Nvidia are in talks over a backstop worth up to $250 billion, a number that tells you how expensive frontier AI has become. The plan would help OpenAI finance a 10-gigawatt data center campus in Pike County, Ohio, and the site alone could push total costs past $500 billion.

The size of the bet is the story

This is not a normal vendor-financing arrangement. A $250 billion backstop tied to a single AI customer says the compute race has moved far beyond buying servers and renting cloud capacity. OpenAI wants enough power to keep scaling models, training runs, and inference demand without being boxed in by existing infrastructure.

The Ohio project is especially striking because it sits on a former uranium-enrichment plant site. That detail matters: the buildout is not just about servers and chips, but about land, power, permitting, and long-term industrial reuse. SoftBank and SoftBank are already involved in the campus through SB Energy, alongside the U.S. Department of Energy.

• The backstop would support lease and construction debt, not the Nvidia chips themselves.
• The talks are still in progress and can change.
• The site is in Pike County, Ohio.
• The campus target is 10 gigawatts, a scale closer to utility planning than typical enterprise IT.

Why Nvidia matters here

Nvidia’s role is the interesting part. The company already dominates the AI accelerator market, but a credit backstop pushes it deeper into the financial plumbing behind AI infrastructure. If OpenAI can borrow against Nvidia’s balance sheet strength, it can move faster on a giant campus without waiting for every dollar of equity to arrive first.

That also explains why the chip and the financing are being discussed separately. The reported backstop covers the building and lease obligations, while the chips inside the data center would be handled in another deal. In other words, OpenAI is trying to secure the roof, the power feed, and the financing stack before the hardware arrives.

“If we really want to build the AI infrastructure, we should build the AI infrastructure.” — Jensen Huang, Nvidia CEO, on CNBC’s Squawk Box

That quote from Jensen Huang fits the moment almost too well. Nvidia has spent years talking like a chip company, but this deal shows how much the business now depends on infrastructure decisions that look more like utility finance than classic semiconductor sales.

OpenAI is spending like a company preparing for the public market

OpenAI’s timing is also worth watching. The company confidentially filed for an IPO with the Securities and Exchange Commission in June, though it has not given a public timeline. That means investors are already being asked to think about OpenAI as a business that needs durable financing, not just another fast-growing AI lab.

The company’s valuation now sits near $1 trillion in private markets, according to the reporting, and that number depends on a simple idea: OpenAI can keep its lead while building a business model that survives intense competition. That is getting harder as Anthropic, Google, Amazon, and Meta spend aggressively on their own AI stacks.

• OpenAI launched ChatGPT in 2022 and has been racing to secure more compute since then.
• Nvidia said in September it would invest up to $100 billion in OpenAI, but that plan never fully landed.
• Nvidia later put $30 billion into OpenAI’s March funding round.
• Huang said that investment might be Nvidia’s last before OpenAI goes public.

The real comparison is infrastructure, not software

For context, 10 gigawatts is enormous. CNBC’s analysis says that amount of power is roughly equal to the annual electricity use of 8 million U.S. households. That makes the Ohio campus feel less like a data center and more like a regional utility project with AI tenants.

It also helps explain why the capital stack could exceed $500 billion. When the power demand gets this large, the conversation stops being about one model or one product release. It becomes about grid access, land, construction debt, chip supply, and the economics of keeping GPUs busy enough to justify the spend.

Open-weight models, many coming out of China, add another layer of pressure because they can undercut pricing and shrink margins. If OpenAI has to spend at this scale while defending pricing power, the company will need either huge revenue growth or financing structures that buy it time.

That is why this news matters to anyone watching AI infrastructure. The headline number is $250 billion, but the deeper signal is that the biggest players now treat compute as a balance-sheet problem as much as a technical one. If this deal advances, expect more AI companies to seek similar financing structures for power-hungry buildouts.

What to watch next

The key question is whether the talks turn into a formal financing package or stay as a signal of intent. If the deal moves forward, it could become a template for how frontier AI firms fund giant campuses: part chip supply agreement, part debt support, part strategic equity, and part utility-scale power planning.

For now, the practical takeaway is simple. OpenAI needs more compute than normal cloud contracts can provide, and Nvidia has the credit profile and strategic incentive to help make that happen. The next update to watch is whether the Ohio campus gets a signed financing structure, or whether the numbers keep growing while the paperwork lags behind.