Ten gigawatts is a power plant, not a purchase order
The largest number in AI this week is measured in watts, not parameters. And the entity best placed to check whether it arrives on time is a regional grid operator.
NVIDIA and OpenAI have committed to at least 10 gigawatts of NVIDIA systems, with NVIDIA intending to invest up to $100 billion as they deploy. Convert the unit before reacting to the headline.
What a gigawatt actually is
One gigawatt is roughly the continuous output of a large nuclear reactor. Ten is a national-scale generation programme. Two companies have committed to that much load for one application, on a schedule that generation and transmission projects have never been able to match.
This is why datacentre siting quietly became energy policy. The limiting resource stopped being GPU allocation some time ago. It is interconnection queues, transformer lead times, and whether a regional operator says yes — all of which run on decade clocks while model generations run on monthly ones.
The financing structure deserves a second read
NVIDIA invests as the systems deploy. The supplier is financing the customer's purchase of the supplier's product. That pattern is well established in capital-intensive industries and it is not inherently improper — but it does mean the demand signal and the funding signal are no longer independent, and independent demand signals are what an outside observer uses to judge whether a buildout is real.
The filings tell you more than the announcements
Against that, Hut 8 disclosed a second 15-year lease for 352 megawatts at $9.8 billion — in a securities filing rather than a press release. That single document does something no gigawatt headline does: it prices roughly a third of a gigawatt and states how long somebody had to promise to pay for it.
Fifteen years is the detail worth sitting with. The contract runs to 2041 and will host eight or nine hardware generations, none of them designed yet. Whoever signed it is not betting on a model. They are betting that demand for compute in general survives every architectural surprise between now and then.
What to actually watch
Not further announcements. The first gigawatt is due in the second half of 2026 on Vera Rubin — a named platform and a named half-year, which is unusually checkable for a commitment this size. Whether it energises on schedule will tell you more about the buildout than anything either company says next.
And watch how often behind-the-meter generation appears in the filings. When it becomes the default rather than the exception, that is the industry telling you the grid said no.
NVIDIA Newsroom — OpenAI and NVIDIA announce strategic partnership to deploy 10 gigawatts of NVIDIA systems → · SEC — Hut 8 Corp. — Form 8-K, FY2026 → · Bismarck Analysis — AI 2026: data centers restart growth of a stagnant U.S. electrical grid →