OpenAI's 10 GW Stargate, in reactors

What the math actually says

A 1,117 MW AP1000 reactor running at the U.S. fleet-average 93% capacity factor produces ~9.1 TWh of electricity per year. To cover 88 TWh of continuous demand, you need 9.7 reactors — round up to 10. Footprint: about 832 acres per reactor site, so the entire 10-reactor build-out occupies under 10 square miles.

The same 88 TWh from West Texas solar (where capacity factor is about 27% — among the best in the U.S.) requires roughly 37 GW of nameplate solar capacity — about 370 utility-scale 100 MW farms on a raw-energy basis. But solar produces zero electricity from dusk to dawn, so to give a 24/7 AI workload firm power equivalent to nuclear, NREL Standard Scenarios suggest 2.5–4× more capacity plus storage. Use the midpoint and you get about 1,100 solar farms totaling roughly 1,000 square miles of land — larger than the entire built footprint of the city of Houston (~600 sq mi).

The wind comparison goes a similar direction: West Texas wind has excellent capacity factors (around 42%) but requires ~2.5× the nameplate to firm up, which works out to roughly 16,000 turbines spread across 2,500 sq mi — larger than the state of Delaware.

Per-site translation

Stargate's currently announced sites and their nuclear equivalents:

Per-site capacities approximate, based on partner press releases and DCD reporting through October 2025. AP1000-equivalents assume 1,117 MW × 93% CF and AI-workload near-baseload utilization.

Why this matters for the grid debate

A single AP1000 reactor (~9 TWh/yr) covers roughly two of these campuses outright. Ten reactors covers the entire $500 billion build. That's a tractable number — equal to about 10% of the existing U.S. nuclear fleet. By contrast, the firm-equivalent solar build needed at the same scale would be roughly equal to the entire utility-scale solar capacity currently operating in the United States.

Most Stargate partners are not actually proposing nuclear — they plan to lean heavily on a combination of grid-tied power, on-site gas turbines (with attendant air-quality concerns), and tax-abated land. But the order-of-magnitude comparison is what makes the argument for nuclear-for-AI viscerally clear: ten reactors on ten square miles, versus the city of Houston covered in solar panels plus battery storage.

Try it yourself

Use the Compare app to plug different metros and power sources together. The Stargate flagship sits inside the Texas grid, but the math holds for any of the 25 metros we model — and the per-metro solar and wind capacity factors are pre-loaded from NREL regional data, so a Seattle comparison correctly reflects that solar capacity factor there is only about 16% (versus 27% in West Texas).