OpenAI has signed a 20-year lease with SoftBank-backed SB Energy for a massive data center campus in Ohio, supported by a NVIDIA financial commitment of up to $105 billion.

The project demonstrates the extraordinary scale of infrastructure now required to support AI’s growth. The campus is expected to require about 10 gigawatts of total power capacity, with roughly 8 gigawatts available for IT equipment. The first 800 megawatts are scheduled to come online in 2028.

NVIDIA will provide backing for the project’s initial phase, which accounts for about 5 gigawatts of power capacity, or approximately 4.25 gigawatts of IT capacity. In exchange, NVIDIA will be the exclusive chip provider for the first half of the project and will invest $1.5 billion in SB Energy.

Additionally, the chipmaker has the option to expand its participation as additional capacity is developed.

The financing structure is significant because NVIDIA is not simply guaranteeing OpenAI’s lease payments. Instead, its commitment protects part of the value of completed data center facilities if OpenAI eventually exits the project.

Under the arrangement, SB Energy would first attempt to find another tenant if OpenAI left. If that failed, SB Energy would seek a buyer for the facilities. NVIDIA could then be responsible for part of any shortfall in value, with its commitment capped at $105 billion for the initial phase.

This structure gives lenders additional protection while limiting NVIDIA’s exposure. It also makes it easier for SB Energy to raise the enormous amount of debt required to build the campus.

For OpenAI, the Ohio agreement secures a major block of computing capacity without requiring the company to build the entire infrastructure itself. For NVIDIA, the deal extends its role from supplying the processors that power AI to helping finance the physical infrastructure needed to deploy them.

Scaled Back Deal

The final agreement is considerably smaller than an earlier proposal. NVIDIA had discussed backing the entire 10-gigawatt project with a commitment potentially totaling about $250 billion. Investors raised concerns about the scale of that exposure, and Nvidia shares dropped 5% after news of the discussions became public.

“NVIDIA’s guarantee for OpenAI’s Ohio campus fell from $250 billion to under $120 billion in three weeks,” said Mitch Ashley, VP and Practice Lead at The Futurum Group. “It now covers the first phase, not all 10 gigawatts. Platform leads are planning against capacity nobody has committed to funding.”

Furthermore, he added, “A chip supplier underwriting its largest customer’s buildout leaves your compute plan to its shareholders. They cut that backstop once already. Teams betting a two-year AI stack on this site need contracted delivery dates and a second source now.”

For NVIDIA, the Ohio project also creates a potentially enormous source of chip sales. The company is expected to supply hundreds of billions of dollars of processors for the first phase and has discussed financing OpenAI’s purchases of its chips.

The project highlights how AI infrastructure competition has expanded well beyond access to GPUs. NVIDIA CEO Jensen Huang has described land, power and building shells as major constraints on AI expansion. Securing enough electricity and physical data center capacity has become challenging as AI companies keep building large computing systems, especially as communities resist new data center construction.

The Ohio campus partly occupies a former US Department of Energy uranium-enrichment site. Its power requirements are expected to be supported by a 9.2-gigawatt natural gas plant owned by the US government, with Japan providing financing under a trade agreement.

The vast size of the financial commitments in this deal are echoed in similar deals across the AI sector. An industry analysis estimates that nine major tech companies have roughly $3 trillion in largely AI-related obligations that do not currently appear on their balance sheets. Future leases alone account for about $1.2 trillion, four times the level of a year earlier.