Nvidia guarantees up to $105 billion for OpenAI's Ohio data center

OpenAI has signed a twenty-year lease with SoftBank subsidiary SB Energy for the PORTS-Pike campus in Ohio, securing around 8 gigawatts of IT capacity there, The Decoder reports; the Wall Street Journal puts the project at 10 gigawatts gross, cooling and infrastructure included. Nvidia is guaranteeing the venture with up to $105 billion and becomes its exclusive chip supplier, US broadcaster CNBC confirmed, citing a stock exchange filing published on Monday. The site sits partly on a former uranium enrichment plant belonging to the US Department of Energy, and draws on a state-owned 9.2-gigawatt gas power station.

How the guarantee actually works

Nvidia is not guaranteeing OpenAI's rent payments. It is guaranteeing the residual value of the completed data centers of the first construction phase, 4.25 gigawatts of IT capacity. If OpenAI walks away, SB Energy must first look for a replacement tenant, then attempt a sale — only after that does Nvidia pay the difference in value, capped at $105 billion. In exchange, Nvidia becomes the exclusive chip supplier for the first half of the site and invests a further $1.5 billion directly into SB Energy.

Nvidia chief executive Jensen Huang coined an acronym for the project: LPS, for land, power and shell. Those foundations have become the real bottleneck of the AI build-out, after chips and networking equipment, Huang said according to The Decoder. Per system generation he expects around 1.5 million GPUs, or $150 billion to $200 billion in revenue. Should Nvidia exercise its option on the remaining 3.75 gigawatts in Ohio, the package grows to roughly 16 gigawatts, worth about $600 billion. The first 800 megawatts are due to come online in 2028, according to CNBC.

Three trillion dollars off the books

How large such pledges have become industry-wide is set out in a Wall Street Journal analysis cited by The Decoder: nine technology groups, among them Alphabet, Meta, Microsoft and Nvidia, carry around three trillion dollars in mostly AI-related commitments that do not appear on their balance sheets — because leases are only recognised once payments begin, and purchase commitments only on delivery. Leases not yet in effect alone total $1.2 trillion, four times as much as a year earlier. At Alphabet, purchase obligations jumped from $332 billion to $811 billion within three months. Since the contracts are barely cancellable, and both Alphabet and Amazon recently reported negative free cash flow, Morgan Stanley analysts warn that investors can hardly assess these companies' real indebtedness any more.

The figure announced is itself already a reduction. According to The Decoder, the Wall Street Journal had reported that Nvidia originally wanted to guarantee the entire project for around $250 billion, then cut the guarantee sharply after a five percent share price drop and noticeable investor pressure. The cap now sealed at $105 billion is even below the $120 billion last rumoured — and it supports only the value of the facilities, not the whole lease.

• Lease: twenty years, PORTS-Pike campus in Pike County, Ohio
• Capacity: around 8 gigawatts of IT power, first 800 megawatts from 2028
• Nvidia guarantee: up to $105 billion, capped at the residual value of phase one
• Industry-wide off-balance-sheet AI commitments, per the WSJ: around $3 trillion

What it changes for companies buying AI capacity

For any company that depends on the AI build-out as a customer, a cloud partner or an investor, this deal shows how much of that build-out runs through financing structures outside the conventional balance sheet — and how hard it is, from the outside, to judge the risk actually bundled there. Anyone weighing exposure to AI infrastructure stocks, or resting a product roadmap on OpenAI's continued compute capacity, should read these sums as conditional ceilings rather than capital already spent. And watch one specific signal: if Nvidia's shrinking guarantees, from $250 billion down to $105 billion, become a pattern under investor pressure, the pace of the build-out itself will slow — and with it the capacity available to rent.