Nvidia is reportedly guaranteeing $250 billion of OpenAI's debt. That's the question to ask before bidding the next AI data center.

Nvidia is reportedly in talks to guarantee roughly $250 billion in lease and construction debt for a planned 10-gigawatt OpenAI data center campus in southern Ohio — because OpenAI itself doesn't carry an investment-grade credit rating. For any GC or trade sub sizing up the next AI data center award, that's the tell: on projects this size, the tenant's name on the lease increasingly isn't who's actually good for the money.

The Wall Street Journal first reported the talks on July 26, with Bloomberg and Reuters following. The project would sit on the site of the former Portsmouth Gaseous Diffusion Plant in Pike County, Ohio, developed by SoftBank's energy subsidiary. At 10 gigawatts and a reported total cost north of $500 billion — construction, power, chips, and labor combined — it would be the largest data center project announced to date. The first phase, about 800 megawatts, is targeted for 2028.

Why does a chipmaker need to guarantee a construction loan?

Because the tenant can't clear the bar on its own. OpenAI doesn't have an investment-grade credit rating, and lenders financing a $250 billion construction and lease package want more assurance than a fast-growing but unrated company's promise to pay rent for a decade. Nvidia's guarantee gives those lenders a stronger name to underwrite against, which is what lets the developer borrow on favorable terms instead of getting quoted junk-bond-level rates or walking away from the deal entirely.

Notably, the guarantee only covers the lease and construction debt — not the Nvidia chips that will fill the buildings. That's a separate financing arrangement being discussed in parallel, reportedly worth another $350 billion. Two different pots of money, two different guarantees, one project.

What does this actually change for a GC bidding this kind of work?

On a conventional commercial building, a GC's payment-risk homework starts and ends with the owner of record: their balance sheet, their bonding capacity, their pay history. Mega data-center work built for AI labs breaks that model. The entity signing the lease (OpenAI) may not be the entity whose credit the money is actually resting on (Nvidia, in this case, alongside the developer, SoftBank). That has practical downstream effects:

None of this changes lien rights or the mechanics of getting paid on a pay app. It changes the due diligence a GC or major sub should do before committing crews, long-lead procurement, and bonding capacity to one of these awards — ask who's standing behind the money before you ask about the schedule.

Does this mean the AI data center boom is on shakier ground?

It's a fair question, and not a new one for this newsletter. We flagged in July that Wall Street had started openly doubting whether hyperscaler AI capex pays off, with data centers now more than a fifth of all nonresidential building starts and increasingly funded by debt rather than cash. A guarantee structure like this one is a symptom of the same underlying issue: the companies driving the AI buildout — OpenAI chief among them — often don't have the balance sheets to finance it on their own credit, so the financing gets engineered around that gap instead. Investor Michael Burry, among others, has flagged the circularity here: Nvidia backstops the debt that lets OpenAI lease the data center that will run on Nvidia chips OpenAI is separately financing. If that circular arrangement holds, the buildout continues. If AI spending cools and any link in that chain wobbles, it's the same kind of mid-construction project risk this newsletter has already tracked on other AI data center sites this year.

Before bidding or subcontracting on the next AI mega-campus, ask your business development team one question that didn't used to matter on commercial work: who is actually guaranteeing this project's debt, and are they still standing behind it if AI spending slows down?

We wrote up Wall Street's growing doubts about AI capex two weeks ago — this guarantee structure is the clearest sign yet of why lenders needed convincing in the first place.

Forward this to whoever on your team is qualifying the next AI data-center opportunity — the credit question belongs in that first conversation, not after award.