OpenAI in Talks to Lease 10GW Data Center with Nvidia Providing Financial Backing
Miles Bennett
OpenAI is negotiating a 20-year lease on the largest AI data center ever planned in the U.S. — 10 gigawatts, at least $500 billion to build — with Nvidia offering its balance sheet as a financial guarantee, binding the two companies from chip buyer-seller to co-underwriters of infrastructure risk.
How big is this data center?
The site is a former uranium-enrichment facility on federal land in southern Ohio. Total planned capacity: 10 gigawatts. Full build-out cost: at least $500 billion.
This means → a single campus would consume roughly one-third of New York City's residential electricity. This is not a building — it is a small power city.
Phase one — 800 megawatts — is targeted for 2028. OpenAI plans a 20-year lease. Talks are not yet finalized.
Why is Nvidia guaranteeing OpenAI's lease?
OpenAI will use Nvidia hardware at the facility. In return, Nvidia will provide financial guarantees backing OpenAI's rent payments and developer SB Energy's project financing. In plain terms = Nvidia is telling lenders: "If OpenAI can't pay, we cover it."
This is brand-new territory for Nvidia — the first time a chip seller has underwritten a customer's infrastructure at this scale. The only precedent: Google guaranteeing debt and leases for Anthropic's TPU data centers.
Nvidia already completed a $30 billion equity investment in OpenAI this year. Adding a financing guarantee shifts the relationship from supplier-client to deep mutual exposure.
What makes this site special?
SB Energy, a SoftBank subsidiary, is the developer. A February U.S.-Japan trade agreement disclosed that SB Energy will invest $33 billion to build a 9.2-gigawatt natural-gas power plant on-site.
The plant will be owned by the U.S. government and operated by SB Energy. This means → the Department of Energy owns the land, so the project can bypass local zoning approvals and sidestep the community opposition that typically delays large data centers.
In March, SoftBank founder Masayoshi Son and U.S. Energy Secretary Chris Wright attended the groundbreaking — the project has moved from planning to construction.
Why isn't renting cloud capacity enough for OpenAI?
OpenAI has already signed lease agreements worth at least $665 billion over five years with Microsoft, Amazon, Oracle, and other cloud providers — yet it keeps pursuing owned infrastructure.
In plain terms = when you rent someone else's server rooms, expansion speed and priority are not yours to decide. Signing your own land, power, and chip deals is the only way to truly control your compute.
Chips and servers typically account for 70% of a data center's total cost. Sources estimate OpenAI needs roughly $350 billion just for Nvidia AI chips. Financing arrangements are still under discussion.
What does this bill mean for the IPO?
OpenAI confidentially filed for an IPO this week. This signals the company is in the final stage of pre-listing preparation — and this massive compute commitment will be a central financial issue for public-market investors.
This means → investors must answer one core question: can OpenAI's revenue growth support tens of billions in annual infrastructure spending? If not, these long-term leases become ticking liabilities on the balance sheet.
Financing terms remain under negotiation — IPO pricing and infrastructure funding will pull on each other, making them OpenAI's two most consequential negotiations over the next year.
Content is for reference only, not financial advice.