Skip to content

415 TWh today, 945 by 2030: the IEA numbers behind the data-centre panic

ecogpt.coPublished Oct 19, 2026 · 2 min read

The International Energy Agency puts global data-centre electricity at 415 TWh in 2024, about 1.5% of world demand, and projects 945 TWh by 2030. Here is what sits inside those figures and why the range around them is wide.

  • 415 TWh in 2024 is roughly 1.5% of global electricity; the IEA's base case more than doubles it by 2030.
  • The United States accounted for about 45% of data-centre electricity in 2024, China 25%, Europe 15%.
  • AI is the fastest-growing component but not yet the largest; conventional cloud and storage still dominate the stock.
  • The IEA's 2035 projection of 1,200 TWh depends on efficiency gains that are not guaranteed.

The headline

In April 2025 the International Energy Agency published its first dedicated report on energy and AI. Its central figures: data centres consumed about 415 terawatt-hours in 2024, around 1.5% of global electricity, and will consume about 945 TWh by 2030 in the base case, more than Japan uses today. By 2035 the projection reaches 1,200 TWh.

What is in the number

The 415 TWh covers all data centres: the hyperscale AI campuses in the news, but also the conventional cloud, enterprise server rooms, content delivery and crypto mining. AI-specific servers are the fastest-growing slice, and the IEA expects them to drive most of the increase to 2030, but the installed base in 2024 was still mostly doing ordinary computing.

Geographically, the United States accounted for about 45% of consumption, China about 25% and Europe about 15%. The rest of the world shared the remainder. That concentration matters for the carbon figure: the US grid averaged about 390 gCO₂/kWh and China's about 580, while much of the European load sits on far cleaner Nordic and French grids.

Why the range is wide

Projections depend on three uncertain inputs: how fast AI demand grows, how much efficiency improves per unit of compute, and how much announced capacity actually gets built. On the first, the IEA notes that AI-related capacity could grow four-fold by 2030. On the second, Google measured a 33× fall in energy per median prompt in a single year, which no forecast anticipated. On the third, grid interconnection queues and turbine lead times are already slowing projects. The IEA's own sensitivity cases span a wide band around 945 TWh.

The US picture is sharper

Berkeley Lab's December 2024 report for the Department of Energy gives finer numbers for the United States: 176 TWh in 2023, 4.4% of national electricity, up from 76 TWh in 2018, with a 2028 range of 325 to 580 TWh, or 6.7% to 12% of US consumption. The Congressional Research Service has since summarised the same figures for lawmakers. The difference between the low and high end of that range is roughly the entire electricity consumption of a mid-sized country.

How the index uses these numbers

They set the scale on the analytics page and the home page. They do not enter any company's score. The index scores companies on what they disclose and do, not on the industry's total. But the total is the reason disclosure matters: a sector heading for a thousand terawatt-hours cannot be managed on blog-post numbers.

Sources

  1. 1.Energy and AI — Executive summaryInternational Energy Agency · Apr 10, 2025
  2. 2.Berkeley Lab report evaluates increase in electricity demand from data centersLawrence Berkeley National Laboratory · Dec 20, 2024
  3. 3.Data Centers and Their Energy Consumption: Frequently Asked QuestionsCongressional Research Service · May 12, 2026

Figures labelled as estimates on the index carry roughly ±3× uncertainty; disclosed figures are the company’s own. Corrections: team@ecogpt.com.

Index Brief

Posts like this, with every score change, once a month.

One email a month. Unsubscribe any time.