Infrastructure

Nvidia May Guarantee 250 Billion of OpenAI's Rent

OpenAI holds no investment-grade rating, so Nvidia is reportedly guaranteeing the lease on a 10 gigawatt Ohio campus. A chipmaker's credit is now a line in your compute budget.

InfrastructureBy Servola Tech Desk2026-07-274 min read

AI-assisted, edited by humans. | Editor-in-Chief: Leon Soliman | Editorial standards

Nvidia May Guarantee 250 Billion of OpenAI's Rent

Key takeaways

Nvidia is in talks to guarantee roughly 250 billion dollars of financing so OpenAI can lease a single 10 gigawatt campus, which means the credit standing behind the largest AI build in the world belongs to the chip supplier rather than to the tenant.
The first phase of that campus is about 800 megawatts and is expected in early 2028, so any 2027 budget built on cheaper inference is built on capacity that has not been financed yet.
Ask your cloud or AI vendor which legal entity actually carries the obligation if a compute commitment fails, because the answer is increasingly a guarantor rather than the company named on your invoice.
Read a guarantee this large as a supply signal rather than a confidence signal, since lenders demanded it precisely because the tenant could not raise the money on its own name.

A tenant that cannot borrow on its own name

In March 2026, on 3,700 acres of a decommissioned Cold War uranium enrichment site at Piketon in southern Ohio, SoftBank's energy unit broke ground alongside the United States Department of Energy. The site is called the PORTS Technology Campus and is designed to reach 10 gigawatts at full build, which would make it the largest single concentration of AI computing capacity yet planned. On 27 July the Wall Street Journal reported that Nvidia is in talks to guarantee about 250 billion dollars of the financing so that OpenAI can lease it.

Why it matters: the guarantee exists because OpenAI does not hold an investment-grade credit rating. A lease of this size is financed against the creditworthiness of the tenant, and the tenant does not have enough of it, so Nvidia's balance sheet is being substituted for OpenAI's. A separate arrangement, reported at up to 350 billion dollars, would support OpenAI's purchase of chips for the same project. The supplier would therefore be guaranteeing both the building its customer sits in and the hardware its customer buys from it.

The full campus could cost more than 500 billion dollars and is to be served by a natural gas plant reported at 33.3 billion dollars. Microsoft, Google and Anthropic have all been reported as having shown interest in the site. None of that changes the structural point. The largest planned block of AI compute on earth is being underwritten by the company that sells the chips going into it.

What 800 megawatts in 2028 says about 2027 prices

The first phase of the campus delivers roughly 800 megawatts and is expected to be complete by early 2028, at an initial cost of 30 to 40 billion dollars. That is about eight percent of the headline capacity, arriving almost two years from now, on a site where ground was broken in March. The 10 gigawatt figure is a destination, not a delivery date.

The number that matters: almost every AI procurement conversation in Europe this year has been conducted against an assumption that inference prices keep falling because supply keeps arriving. The supply that would justify that assumption is largely this kind of project, and this kind of project now needs a third-party guarantee before lenders will fund even its first phase. A buyer planning 2027 spend on the expectation of a glut is planning against capacity that is still being negotiated.

Yes, but: capacity is genuinely being added elsewhere, and model efficiency has cut the cost of a given task faster than raw compute prices have moved. Both are real. Neither changes what the guarantee reveals, which is that the people lending the money price this build as risky enough to require someone else's name on it.

Three questions to put to your cloud vendor this quarter

Ask which legal entity carries the obligation behind your committed capacity. Contracts for reserved AI compute are frequently signed with a regional subsidiary or a reseller, while the capacity itself depends on a chain of leases and guarantees several steps removed. You are entitled to know whose covenant you are actually relying on, and the answer belongs in your supplier risk register rather than in a footnote.

Ask what happens to your price if a phase slips. Most enterprise AI commitments now run two to four years, which places their back half inside the window when this Ohio capacity is meant to arrive. Establish in writing whether a delay upstream releases you, reprices you, or simply leaves you paying the committed rate for capacity delivered late.

Ask for the exit. A guarantee structure of this scale signals that the financing market treats concentration risk as material, and your own board should treat it the same way. Set a ceiling on the share of your AI workload that may depend on any single campus, model provider or chip vendor, and set it before the next renewal rather than after the next headline.

Frequently asked questions

Does a guarantee like this make AI compute cheaper or more expensive for us?

Cheaper in the short run and less predictable in the long run. A guarantee lowers the borrowing cost of the project, which supports lower headline compute prices when the capacity arrives. It also concentrates the risk, because a single guarantor now stands behind a very large share of planned supply, and concentrated risk tends to be repriced abruptly rather than gradually.

We buy through a European provider. Why does an Ohio lease matter to us?

Because reserved capacity is traded internationally. European providers frequently resell or backstop their own commitments through the same small set of hyperscale campuses and chip allocations, so a delay or repricing at the top of that chain reaches a European contract within a quarter or two. The invoice is local. The dependency usually is not.

Should we delay signing a multi-year AI commitment until this settles?

Delay is rarely the right answer, because capacity is genuinely scarce and waiting can cost you an allocation. Shortening the term is the better move. A two-year commitment with a defined renegotiation trigger gives you most of the volume discount and keeps you out of the part of the timeline where this financing question is still open.

A chip supplier guaranteeing its customer's rent is not a scandal and it is not fraud. It is a disclosure. It tells you that the people closest to this build, the ones lending against it, do not think the tenant can carry it alone. Price your own commitments on that information rather than on the headline capacity number.

NvidiaOpenAISB EnergySoftBankdata centresAI computecounterparty risk

Nvidia Stopped Renting Its CPU Core From Arm

Nvidia's new Vera server CPU runs Olympus, its first in-house core, not licensed Arm cores. The three percent benchmark win is not the story - the lock-in is.

5 min read

Infrastructure

Anthropic's Fix for Claude Outages Arrives in 2027

AMD will invest up to $5bn in Anthropic and supply 2 gigawatts of MI450 chips. The relief for Claude's rate limits is real, but the first gigawatt is dated to 2027, and your chip supplier now owns a stake in its own customer.

4 min read

Law & Regulation

The EA Buyout Now Hinges on Brussels

The largest take-private in history has cleared EA shareholders and US antitrust. The last real gate is the EU - and not classic merger review, but the Foreign Subsidies Regulation, facing its first blockbuster test.

3 min read

← All articles