IEA forecasts 70% rise in data-center emissions by 2030 as AI drives power demand
The agency projects sharply different power supplies: gas dominates US data centers, coal supplies half of China’s AI facilities’ electricity, and renewables lead in the EU.
The IEA’s annual outlook points to a power-supply constraint behind the AI buildout: by 2030, global data-center electricity use is expected to double, with AI responsible for three-quarters of the increase, while sector emissions rise 70%. Regional mixes diverge sharply—US data centers rely heavily on gas, China’s AI centers on coal, and EU centers mostly on renewables—so renewable-energy purchases alone do not determine the grid impact. Planned US gas capacity underscores how infrastructure choices will shape the buildout’s emissions profile.
01
The IEA estimates AI data centers will add 350 terawatt-hours of annual electricity demand by 2030, more than Great Britain uses in a year.
02
As of August 2026, about 180 gigawatts of gas-fired capacity was planned for US data centers; these are proposals, not completed plants.
03
The IEA tracked construction at 93 large AI factories and expects their combined capacity to reach about 50 gigawatts when completed.
Data-center developers have become the world’s biggest corporate buyers of renewable energy. Yet their growing demand for power is still pushing emissions upward. The International Energy Agency forecasts a 70% increase in data-center carbon emissions by 2030, as fossil fuels supply much of the electricity behind the AI buildout, according to The Guardian’s account of its annual report.
The agency expects data centers to consume twice as much electricity by the end of the decade as they do today. AI accounts for three-quarters of that growth—not all of it. The emissions forecast covers the data-center sector, rather than AI facilities alone.
For AI data centers specifically, the IEA estimates an additional 350 terawatt-hours of electricity consumption each year by 2030. That increase alone exceeds Great Britain’s annual electricity consumption. It measures the extra electricity AI facilities would use each year, not their total consumption over the four-year period.
Three regions, different power supplies
The projected emissions increase reflects where the electricity comes from. Despite developers’ renewable-energy purchases, the IEA expects their appetite for fossil fuels to grow even faster, particularly in the US and China. Its regional forecasts describe very different electricity mixes rather than one uniform global path.
In China, renewables would supply most of the electricity remaining after coal, with smaller contributions from nuclear power and very little gas. In the EU, nuclear would follow renewables, ahead of gas and a very small coal contribution. Those forecasts put the two regions on distinctly different supply paths.
A US gas pipeline nearly as large as the EU’s installed fleet
The US gas forecast has a substantial development pipeline behind it. As of August 2026, the IEA identified plans for about 180 gigawatts of new gas-fired power plants linked to US data centers. That proposed capacity is almost as large as all the gas-fired generating capacity currently installed across the European Union.
These are plans, not an inventory of completed plants. Their geographic concentration is striking: the agency identified 11 gigawatts of planned gas generation for data centers in Canada, and just 4 gigawatts proposed by AI developers across the rest of the world. Renewables are expected to play a greater supply role outside the US and China.
Tracking the buildout from shipments and space
To estimate growth in an industry with limited transparency, the IEA combined evidence about equipment sales with observations of construction. The Guardian describes two inputs to the agency’s forecasts:
AI server shipments: annual global shipments have increased eighteen-fold since 2020, providing a measure of the equipment flowing into the sector.
Satellite imagery: the agency tracked construction at 93 large “AI factories” worldwide. It expects their combined capacity to reach about 50 gigawatts when completed.
The existing global data-center fleet, including facilities with AI-optimized servers, stands at an estimated 120 gigawatts. China, the US and the EU hold 80% of that capacity. Under the IEA’s central scenario, another 120 gigawatts would be built by 2030, with AI data centers accounting for three-quarters of the addition.
Sources
theguardian.comDatacentres to create 70% more carbon emissions by 2030, IEA report warns
Reader comments
Newest comments first. Replies stay oldest first.