You know a market has gone fully unhinged when the phrase "just a quarter-trillion-dollar rental guarantee" gets typed by a financial journalist without an exclamation point. Nvidia is reportedly in talks to backstop up to $250 billion so OpenAI can lease a data center — and that's before anyone's paid for the chips going inside it.

A Decommissioned Uranium Site Becomes a Very Expensive Landlord Situation

The project comes from SB Energy, a SoftBank subsidiary, and it's a 10-gigawatt data center campus planned about 50 miles south of Columbus, Ohio, on a former uranium-enrichment site. Nvidia's guarantee would cover the lease itself, letting OpenAI commit to the space without SoftBank taking on the full financial risk alone. The chips that actually go inside — estimated at another $350 billion — are a separate conversation, one Nvidia is also reportedly circling as a potential financier.

Add it up and the total project, land to silicon, could top $500 billion, which would make it the largest data center commitment ever assembled. The campus is projected to deliver up to 800 megawatts by 2028 — enough electricity to power roughly 640,000 homes, except all of it will be spent on tokens instead of toasters.

When "Circular" Stops Being an Insult and Starts Being the Business Model

Here's the part that should raise an eyebrow: Nvidia guaranteeing financing for OpenAI to lease a building that will be filled with Nvidia chips, purchased with Nvidia's help, running workloads that justify Nvidia's next earnings call. Critics have started calling this the "circular AI" problem, where the industry's biggest players are effectively financing each other's demand rather than demand emerging organically. Nvidia's own leadership has acknowledged the optics aren't great, even as they keep signing the checks.

None of this is finalized — terms are still being hashed out, and reporting is clear the whole arrangement could still fall apart. But the fact that a quarter-trillion-dollar guarantee for real estate is a plausible headline in 2026 tells you everything about how much conviction (or how much fear of missing out) is baked into the AI infrastructure race right now.

Somewhere, a normal-sized business loan officer is having a very difficult time recalibrating what "big number" means this week.

Source: CNBC