Nvidia is in talks to provide roughly $250 billion in financing guarantees for OpenAI to lease a massive data center in Ohio, according to reporting from Bloomberg and the Wall Street Journal on July 26. The deal, still in early stages, would mark a turning point for AI infrastructure: OpenAI gaining independence from cloud giants like Microsoft and Amazon.

The project is expected to cost more than $500 billion total. SoftBank’s energy subsidiary is developing a 10-gigawatt facility in southern Ohio. Nvidia’s guarantee would cover the data center lease and debt financing, but not the chips themselves—though the chipmaker is separately discussing up to $350 billion more to finance OpenAI’s chip purchases.

What This Deal Means for OpenAI

OpenAI has relied on Microsoft’s infrastructure since the companies deepened ties in late 2023. Amazon and Oracle have also hosted OpenAI workloads. Owning dedicated infrastructure gives OpenAI control over capacity, latency, and—critically—the ability to manage its own cost structure as the company scales.

The first phase is expected to finish in 2028 with around 800 megawatts of power. For context, that’s roughly enough to power a small city. The facility would give OpenAI the dedicated compute it needs to train, fine-tune, and serve increasingly large models without throttling from third-party vendors.

The Nvidia Angle

For Nvidia, the deal guarantees demand for its chips for years to come. OpenAI consumes cutting-edge GPUs at scale. A commitment to finance OpenAI’s chip purchases locks in revenue and also signals to the market that Nvidia believes AI infrastructure buildout is accelerating, not peaking.

Negotiations are early. Deal terms could change or the agreement could collapse. Nvidia has made similar infrastructure plays before, but this scale is unprecedented: a quarter-trillion-dollar bet on OpenAI’s trajectory.

Timing and Leverage

Why now? OpenAI is burning through capital to train larger models and expand its enterprise business. Microsoft’s infrastructure is proven but constrained by Microsoft’s own compute needs. SoftBank has been aggressive about energy projects in the US, particularly in the Midwest where land and power are cheaper.

The deal also reflects a broader shift. Hyperscalers are no longer the default architects of frontier AI. Companies like OpenAI and Anthropic want their own compute, not rented access that can be throttled or priced dynamically.

If the deal closes, it reshapes how frontier AI labs build their infrastructure—and who profits from that build.

FYI

How much money are we talking about here?

$250 billion for the data center lease and debt, plus up to $350 billion for chips, totals roughly $600 billion in Nvidia exposure. For scale: that’s about 20% of Nvidia’s current market cap, hypothetically, as a guarantee, not an immediate outlay.