AI could need 315 GW by 2033 — an 1,100% jump from today

The figure making the rounds: 315 GW of AI-driven electricity demand by 2033, up more than 1,100% from 2025. That math only works if you're counting AI compute alone — call it ~26 GW today — and not the whole data center fleet.

For scale, Gartner puts total global data center power at 105 GW in 2025, 133 GW in 2026, and 291 GW by 2030. AI is the entire story behind that curve: AI-optimized servers jump from 95 TWh in 2025 to 175 TWh in 2026 (+84%), overtake conventional servers by 2027, and reach roughly half of all data center draw by 2030. Conventional servers grew 1.2% over the same stretch.

The IEA is more conservative on absolute numbers — 945 TWh globally by 2030, about 3% of world electricity — but agrees on the shape: accelerated servers compound ~30% a year and account for nearly half the net increase. Neither forecast thinks silicon is the constraint. Gartner's blunt version: grid supply will be insufficient to meet the demands of future data center construction — and that hits every data center tenant, not just the AI ones.

Treat 315 GW as a directional bet, not a settled number. Long-range power forecasts have a habit of being revised in both directions.

Why it matters: compute is now an energy business, and the grid — not the fab — is what caps how fast AI can scale.