OpenAI and NVIDIA Plan 10 Gigawatts of AI Compute

The scale of the AI buildout keeps redefining what “big” means. OpenAI and NVIDIA announced a strategic partnership to deploy at least 10 gigawatts of AI data centers built on NVIDIA systems — millions of GPUs powering OpenAI’s next generation of infrastructure. To support it, NVIDIA intends to invest up to $100 billion in OpenAI, released progressively as each gigawatt comes online. The first gigawatt is slated for the second half of 2026 on NVIDIA’s new Vera Rubin platform.

What 10 gigawatts actually means

We’re used to measuring data centers in servers or chips. This deal is measured in power — and that’s the tell. At this scale, the binding constraint on AI isn’t just buying GPUs; it’s finding the electricity and the physical sites to run them. Ten gigawatts is the output of roughly ten large power plants, dedicated to one company’s AI compute.

The physical footprint is just as striking. Separately, OpenAI is reported to be in talks to lease a 10-gigawatt data-center campus in Ohio, an undertaking estimated to cost at least $500 billion once you count chips, construction, power, and labor. Numbers like that put the AI infrastructure race on the scale of national infrastructure projects — which is part of why governments are now writing comparably sized checks.

The circular logic of the deal

There’s a dynamic here worth naming plainly: NVIDIA is investing up to $100 billion in a customer that will spend heavily on NVIDIA chips. Critics call it circular; supporters call it aligned incentives. Either way, it’s become a defining pattern of the era — suppliers funding the customers who buy from them, locking in demand and supply at once. It’s the same logic behind Micron’s investment in Anthropic, just at a larger scale.

Why it pairs with memory

A GPU is only as useful as the data you can feed it, which is why this announcement is inseparable from the memory supercycle. Each gigawatt of compute needs enormous quantities of high-bandwidth memory, and the Vera Rubin platform is designed around exactly that pairing. The chips, the memory, and the power all have to scale together — a shortfall in any one caps the whole system.

The takeaway

The OpenAI–NVIDIA partnership is less a product launch than a declaration of intent: both companies are betting that demand for AI compute will keep climbing steeply enough to justify spending on the scale of national infrastructure. If they’re right, 10 gigawatts is a down payment. If the demand curve bends, this is the kind of commitment that looks very different in hindsight. For now, it sets the pace everyone else in the industry is measured against.