This is part of WeGoAI's daily series on the things that actually shape our world. All figures from the IEA's "Key Questions on Energy and AI" (published this month), the IEA's Electricity 2026 outlook, and Goldman Sachs Commodities Research via Aterio facility data. Not investment advice.

“This is part of WeGoAI's daily series on the things that actually shape our world.”

The Number Everyone Will Quote

The International Energy Agency updated its energy-and-AI analysis this month, and the headline is clean: global data centre electricity consumption is projected to roughly double from 485 TWh in 2025 to about 950 TWh in 2030 — around 3% of total global electricity demand by the end of the decade.

For scale: 945 TWh is slightly more than Japan's entire electricity consumption today.

But a single number flattens what is actually a range, and the range is where the decisions get made. The IEA runs four scenarios for 2030:

Notice how tight the band is in 2030 — 833 to 1,008 TWh, a spread of only about 20% — and how wide it blows open by 2035, from 942 to 1,637 TWh. The IEA's own framing: bottlenecks across chips, grid connections and energy equipment are now the binding constraint on near-term growth, not demand. Investment is booming and project pipelines are surging, yet the central projection barely moved from last year's report. That is a supply story wearing a demand costume.

The Trajectory So Far

Here is the actual curve, with the base-case path on top of it:

Growth rates, year over year: +16% (2024), +17% (2025). The AI-focused subset grew 50% in 2025 alone — well outpacing the 3% growth of total global electricity demand. The IEA's base case implies about 15% annual growth for the sector through 2030, more than four times the pace of everything else in the economy.

Two forces are pulling in opposite directions inside that curve:

  • Efficiency is collapsing the cost per task. The IEA says energy use per AI task has been dropping "by at least an order of magnitude annually in recent years" — faster than any technology in energy history. A simple text query now takes less electricity than running a TV for the same duration.
  • Usage is exploding faster than efficiency saves. Major model providers tripled active users and grew revenue fivefold in a year. And the mix is shifting toward the heavy stuff: reasoning, video generation and agentic AI, which can consume hundreds to thousands of times more energy per query than simple text generation.

That is why AI-focused data centre power use is set to triple by 2030 while the sector overall only doubles. Efficiency gains are real; Jevons' paradox is realer.

The US Is Where the Friction Lives

The United States already accounts for 45% of global data centre electricity use (China 25%, Europe 15%). And the American buildout curve is where the numbers get genuinely steep:

Goldman Sachs Commodities Research, building on Aterio's facility-level tracking (locations, permits, construction status, satellite imagery), forecasts US data centre power demand more than doubling from 31 GW in 2025 to 66 GW in 2027. Look at the additions bar: the industry delivered 6.4 GW of new capacity in 2024 and 8.5 GW in 2025. The schedule now calls for 13.6 GW in 2026 and 36.3 GW in 2027 — a quadruple jump in a single year.

Whether that schedule survives contact with reality is the question. Grid interconnection queues stretch for years, and the IEA's satellite tracking of onsite gas projects — the industry's workaround — shows most still in early stages. The agency's blunt finding: providing reliable gas-fired power for the rapid swings of AI load requires overbuilding generation by 30% to 70% relative to demand. Gas turbine order books are already at a 25-year high, and delivery slots run past 2029.

The IEA's Electricity 2026 outlook puts it in macro terms: data centres will drive about half of all US electricity demand growth through 2030 — more than 420 TWh of new American demand, half of it ours.

How the Grid Answers Back

The supply response is the most interesting part of the new report, because it is no longer just "build more gas plants":

  • Renewables PPAs. The tech sector signed around 40% of all corporate power purchase agreements for renewables in 2025. Half of global data-centre demand growth to 2035 is projected to be met by renewables plus storage.
  • Nuclear renaissance, funded by AI. Conditional offtake agreements between data centre operators and small modular reactor projects grew from 25 GW at end-2024 to 45 GW today. The first SMRs are expected online around 2030.
  • Batteries as grid assets. Because AI load swings violently (training jobs ramp and drop in seconds), onsite battery storage is becoming standard in next-generation campuses — which, with the right incentives, turns data centres into grid support rather than grid burden.
  • AI helping the grid itself. The IEA estimates AI-based line-rating and fault-detection tools could unlock up to 175 GW of transmission capacity without building a single new line — more than the entire projected data centre load increase to 2030.

The Bottom Line

The 950 TWh number is the least interesting thing in the IEA's new report. The interesting part: demand is no longer the constraint — chips, transformers, turbines and grid connections are. Whoever clears those bottlenecks first sets the actual pace of AI, and the map of who powers the AI economy (the US at 45%, China at 25%) is being drawn right now, in permit queues and offtake agreements, not in model benchmarks.

Sources

  • IEA, Key Questions on Energy and AI (updated analysis, published October 2026): 485 TWh 2025 actual, 950 TWh 2030 projection, scenario table (833/868/945/1,008 TWh for 2030; 942/1,013/1,193/1,637 for 2035), 17% growth 2025, AI-focused +50%, $400B+ capex of five tech companies in 2025 with +75% expected in 2026, SMR offtake pipeline 25→45 GW, 30–70% gas overbuild, 15–27 GW onsite gas by 2030, 40% share of corporate renewables PPAs, 175 GW transmission unlock.
  • IEA, Energy and AI (April 2025): 2024 regional split (US 45%, China 25%, Europe 15%), 415 TWh 2024 baseline, US half-of-demand-growth framing.
  • IEA, Electricity 2026, demand chapter: 3% global demand growth 2025, 3.6% average 2026–2030, US +420 TWh over five years with data centres ~50% of growth.
  • Goldman Sachs Research (Wei, Struyven, Dart), based on Aterio facility data: US data centre demand 31 GW (2025) → 41 GW (2026) → 66 GW (2027); capacity additions 6.4/8.5/13.6/36.3 GW.
  • Industrial Info Resources via IIR Energy: ~4,130 operating data centres worldwide, ~4,500 projects starting 2026–2028 valued at ~$2.63 trillion.