A single firm just agreed to shoulder an unprecedented slice of the financial risk needed to build a U.S. AI mega-campus, making a project that once looked speculative far easier to finance.
NVIDIA is offering contracts that promise to cover the leftover value of OpenAI's leased data-center infrastructure, with an aggregate obligation capped at $105 billion. Separately, NVIDIA is investing $1.5 billion in SB Energy to help fund the PORTS-Pike campus in Ohio.
Those guarantees initially underwrite about 4.25 gigawatts of IT capacity, and NVIDIA holds an option to expand support toward roughly 8 gigawatts if exercised. The campus is being developed for OpenAI, putting it on track to become one of the country’s largest dedicated AI compute sites.
Why this matters: building huge, specialized data centers has always been a financing problem. Lenders and developers worry that custom hardware can lose value if demand shifts. By promising to cover that residual value, a major chip supplier is effectively acting as a backstop, lowering the financial hurdle for much larger builds.
How it works in plain terms: think of the guarantee as an insurance policy for banks and builders. If the specialized site loses value, NVIDIA would step in up to the contract cap. That’s not the same as handing over $105 billion today, but it makes lenders and developers far more willing to commit capital.
Practical reality: this does not change who can run frontier models on a laptop. It changes who can afford to build the enormous, power-hungry facilities those models require.
What changes now: projects that once required cautious, incremental steps can scale faster because financing risk has shifted. Startups still won’t own this hardware, but the path to industrial-scale compute is now easier to underwrite by companies tied to the supply chain.
What to watch next: will other vendors copy this mix of guarantees and equity, and how will regulators view a chipmaker underwriting critical infrastructure? The strategic question is whether this becomes the new normal for scaling AI compute, or a one-off tied to a handful of customers.