Gigawatt-Year: The Unit That Makes AI Economics Legible

Power is a rate and money is a flow, and the two only become comparable when you fix a period. A gigawatt-year is that period: one gigawatt of capacity sustained for one year, which is 8,760 gigawatt-hours, or 8.76 terawatt-hours of energy.

Why the unit exists

Build cost is a stock — $35–60 billion for a gigawatt, paid once. Rent is a flow — so many billion a year. Without a common denominator the two numbers cannot be put in the same sentence, and the industry puts them in the same sentence constantly. The gigawatt-year is what makes “does this asset pay back?” an answerable question.

The mistake it prevents

When a slide says “$110 billion per gigawatt”, the missing “per year” is doing a lot of work. Read as a stock, against a $50 billion build, the asset repays itself in six months. Read correctly, as an annual flow, the comparison becomes: how many years of rent does it take to cover the build?

The last row is the one that should look wrong. Assets that pay back in six months do not get rented and financed and litigated over for five years; they get bought. When a payback period is that short, the number above it is usually not revenue.

A sanity check anyone can run

Multiply the energy by a price. At $50 per megawatt-hour — a plausible wholesale power price — a gigawatt-year of electricity is worth about $438 million. That is not what the asset sells. It sells tokens, at prices set by model quality and scarcity, which is why the same 8.76 terawatt-hours can be worth $10 billion or $110 billion depending on assumptions nobody states.

The unit does not tell you which of those is right. It tells you which one a given slide is claiming.