IEA / IMF / global energy industry · 2025 · 04 · 17 · Impact · ~3 min read
AI's data centers reopened fossil fuel power plants
What's actually new
- Data center electricity at scale. ~415-460 TWh globally in 2024 — bigger than the electricity use of many mid-sized countries.
- US data centers: 4% of national electricity in 2024, on track for 12% by 2028. Single-site data centers in some areas now demand more power than entire cities.
- Fossil-fuel reopenings. Mothballed gas and coal plants were brought back online in the US, China, and elsewhere to meet AI demand. The 'dash for gas' became real.
- Renewables can't keep up — yet. Wind and solar are growing, but not at the pace AI infrastructure is being built. The IEA estimates ~50% of new demand will be met by renewables through 2030; the other half is fossil fuels.
If you want more
Worth knowing
- Tech companies' renewable PPAs are real but not always additional. Many 'powered by 100% renewables' claims rely on certificates that don't necessarily mean new clean energy was built for that data center.
- Efficiency gains are real but losing the race. Data centers got more efficient; AI demand grew faster than efficiency improved.
- 'AI ends climate change' headlines are vendor talk. AI may help research clean energy; in the meantime it's accelerating the climate problem.
Who should care
Anyone tracking climate change. Anyone who lives near a planned data center. Energy policy people. Investors in clean energy or fossil fuel infrastructure. Parents thinking about what their kids' planet will look like in 2050. Anyone who casually uses AI thousands of times a day without thinking about the electricity behind it.
What to do about it
If your company is buying AI services, ask vendors about their data centre energy mix — and whether their renewables are 'additional' or just certificate offsets. If you're in an area where data centers are being built, ask local officials about the grid load and emissions impact. As an everyday user, no individual chat moves the needle — but the cumulative effect is real, and the right pressure point is policy and procurement, not personal abstinence.
Honest take
AI data center electricity demand is the AI story most likely to make headlines for the wrong reasons in 2026-2030. The technology that's supposed to help solve climate change is, right now, making it harder to solve climate change — by reopening fossil-fuel plants and outpacing renewable growth. The honest answer isn't 'don't use AI' (the demand is structural, not individual). It's 'pressure the energy procurement of the companies running the AI', and bet on nuclear and grid-scale storage being part of the actual answer through the rest of this decade.
Sources
Last verified · 2026 · 05 · 05 · Found a fact wrong? corrections@aguidetocloud.com