The IEA's Energy and AI Report — Data-Centre Power Set to Double by 2030

How much electricity does AI use? The International Energy Agency answered that question with a global estimate in Energy and AI, a special report published on 10 April 2025. According to the report, data centres worldwide consumed about 415 TWh of electricity in 2024, around 1.5% of global electricity consumption. In the IEA's Base Case that more than doubles to about 945 TWh by 2030 — slightly more than Japan's entire electricity consumption today. AI is the biggest driver: electricity demand from AI-optimised data centres is projected to more than quadruple by 2030.

Announcing the report, IEA Executive Director Fatih Birol said data centres are on course to account for almost half of the growth in electricity demand to 2030 in the United States, more than half in Japan, and as much as one-fifth in Malaysia.

The report in numbers

The main figures from the report's executive summary are below. All are IEA estimates, and some are projections rather than measurements.

The report stresses that data centres are a small share of global consumption but have far larger local effects. Nearly half of US data-centre capacity sits in five regional clusters. Data centres account for around a tenth of global electricity-demand growth to 2030 — less than industrial motors, air conditioning or electric vehicles — but in advanced economies, after decades of flat demand, they account for more than 20% of growth.

Where the power comes from

Renewables and natural gas take the lead in meeting data-centre demand. Half of the growth to 2035 is met by renewables supported by storage and the wider grid. Natural gas expands by 175 TWh, notably in the United States, and nuclear contributes about the same, notably in China, Japan and the United States. The first small modular reactors come online around 2030.

One example of a technology company securing its own generation is Microsoft's agreement to restart Three Mile Island Unit 1; that story is on the Three Mile Island restart (2024) page.

The report also warned about grid constraints. Unless addressed, around 20% of planned data-centre projects could be at risk of delays. New transmission lines take four to eight years in advanced economies, and waiting times for transformers and cables have doubled in three years. Gas turbines face lead times of several years, which could push commissioning beyond 2030. If the power sector does not step up, the report says, meeting data-centre load growth could entail trade-offs with other goals such as electrification, manufacturing growth and affordability.

Uncertainty and the climate

The IEA repeatedly stresses uncertainty: how quickly AI is adopted, how capable and productive it becomes, how fast efficiency improves and whether energy bottlenecks are resolved. It therefore models a faster-adoption Lift-Off Case, a Headwinds Case and a High Efficiency Case, in which 2035 demand is 20% below the Base Case.

On climate, the report is direct. Emissions from data-centre electricity rise from 180 million tonnes today to 300 million tonnes by 2035 in the Base Case and up to 500 million tonnes in Lift-Off. Data centres are among the fastest-growing sources, but remain below 1.5% of energy-sector emissions. The IEA wrote that concerns that AI could accelerate climate change appear overstated, as do expectations that AI alone will address the issue. Broad use of existing AI applications could cut emissions by the equivalent of about 5% of energy-related emissions in 2035 — far more than data centres emit, but far less than climate goals require.

The report also covers energy security. Cyberattacks on energy utilities have tripled in four years and become more sophisticated because of AI. China accounts for around 99% of refined gallium, used in advanced chips, and data-centre demand for gallium could exceed 10% of today's supply by 2030.

A year later

On 16 April 2026 the IEA published a follow-up, Key Questions on Energy and AI. It put data-centre consumption at 485 TWh in 2025 and 950 TWh in 2030, saying the central projection remained close to the 2025 trajectory, with AI-focused data centres tripling their consumption over the period. Booming investment and swelling project pipelines might seem to point to an upward revision, but the IEA judged that real financial and physical frictions limit near-term growth, so the Base Case still offers a balanced assessment.

What it changed

AI's power problem shows up less as a global total than as a question of specific grids and electricity bills. The 2026 follow-up put this front and centre.

It says data centres accounted for 1.5% of global electricity consumption in 2025 but as much as 20–30% of demand in the states, metro areas or regions where they cluster. Large, concentrated and quickly built, they trigger new generation and grid investment; if their demand later falls short, those costs risk landing on other customers. The report notes that actual peak load from data centres is often uncertain and that, unmitigated, the mismatch could raise electricity prices in some places.

Responses have begun. According to the follow-up, regulators or utilities in 24 US states have approved or proposed tariff measures such as collateral for new connection requests, minimum billing commitments of up to 20 years and early-exit fees. Hyperscalers such as Microsoft have committed to covering the cost of the power infrastructure their data centres require, and several others signed a federal government-led pledge not to pass infrastructure costs on to US households.

Local opposition is visible in the numbers too. Citing Data Center Watch, the follow-up reports that in the second quarter of 2025 alone, 20 US data-centre projects with an estimated 98 billion dollars of investment were blocked or delayed, over concerns about the environment and electricity affordability.

The benefits of AI reach users everywhere, but the grid upgrades and the bills land on the particular places where data centres are built. The IEA's two reports put numbers on that imbalance.