The numbers in AI infrastructure keep losing touch with intuition. The latest, per a report from The Information: OpenAI is in advanced talks to lease a 10-gigawatt data-centre campus in Ohio.

What's reported

According to the report — widely syndicated via Reuters and others — OpenAI is negotiating a 20-year lease on a proposed campus on Department of Energy land in southern Ohio, developed by SB Energy, a unit of SoftBank. The build could cost at least $500 billion; OpenAI would control the equipment, with lease payments starting once it's operational and a first phase expected around 2028. Nvidia is expected to supply the hardware and provide a financial guarantee for OpenAI's lease and SB Energy's financing.

Put 10 gigawatts in perspective

10GW is not a normal data-centre number. A large data centre today might draw tens to a few hundred megawatts. 10,000 megawatts is on the order of ten nuclear reactors' worth of power dedicated to a single AI campus — which is exactly why federal land, SoftBank-scale development and Nvidia financial guarantees are even in the conversation. This is infrastructure at the scale of national energy planning.

The caveat that matters

This is a single-source report. The original came from The Information, and Reuters said it could not immediately verify it. Neither OpenAI, Nvidia nor SoftBank has confirmed the specifics. So treat the figures as reported, not official — the kind of story that's directionally important even before it's nailed down.

Why it matters either way

True or not-yet-true, the report fits a clear pattern: the frontier labs increasingly believe the binding constraint on AI isn't algorithms — it's power and compute at industrial scale. When the plans on the table are measured in gigawatts and half-trillions, the race has stopped being about models alone and become about who can physically build the most intelligence.