Stargate's $500 billion question: who actually pays for the power?

Half a trillion dollars. Ten gigawatts. The largest private infrastructure pledge anyone has ever made. Stargate is a number so big it stops meaning anything, which is exactly the problem. Strip away the steel and the silicon, and one quieter line item is left standing, the one that doesn't get bought once but paid every single hour for a decade. Who pays for the power?

The pledge, in numbers

Announced in January 2025 by OpenAI, SoftBank and Oracle, The Stargate Project committed to invest $500 billion over four years in US AI infrastructure, with $100 billion deployed immediately, targeting 10 gigawatts of capacity (OpenAI).

By September 2025 it had a flagship site live in Abilene, Texas, and five more announced across Texas, New Mexico, Ohio and the Midwest, bringing planned capacity to roughly 7 gigawatts and committed investment above $400 billion (OpenAI, CNBC). Oracle's slice alone has been reported in the hundreds of billions for several gigawatts of capacity.

The number behind the number

Here's the trap in a headline like "$500 billion." It reads as a one-time cost, a giant cheque written once to buy chips and pour concrete. But a data center is not a building you finish. It's an appetite you feed. Ten gigawatts running flat out is roughly the continuous output of ten large power stations, consumed every hour, every day, for the life of the hardware.

So the capex headline, vast as it is, may be the smaller story. Over a decade, the cost of the electricity to run 10 GW can rival the cost of building the thing in the first place. The pledge everyone quotes is the down payment. The power bill is the mortgage, and nobody puts the mortgage in the press release.

Why the power price is the whole ballgame

If the electricity bill is the part that compounds, then the single most important term in the entire project is the price per kilowatt-hour, multiplied by every hour the machines run. A few cents of difference, applied across gigawatts and years, is the gap between a profitable cluster and a stranded one. This is why the frontier is suddenly signing power deals with the urgency of people who've realised the grid won't simply hand them what they need.

The lesson scales down without losing its edge. You don't need 10 gigawatts to feel it. At any size, the operator who secured cheap, dependable power has a structural advantage that no amount of clever software erases, because their cost floor is simply lower, hour after hour, for as long as the cluster lives.