Business & markets
Study says AI data centers plan large-scale self-built gas power stations
BloombergNEF research shows that data centers in the United States, the United Kingdom, and continental Europe are planning a combined 181GW of self-built gas power stations because of delays in grid connections. If completed, related facilities in the United States could emit about 345 million tons of carbon dioxide equivalent annually, creating a risk of long-term fossil fuel lock-in.
Research shows that AI data center developers are planning large-scale construction of on-site gas power stations to meet rapidly growing electricity demand. Disclosed planned fossil-fuel power plants in the United States, United Kingdom, and continental Europe total 181GW, nearly four times the United Kingdom's peak total electricity demand.
BloombergNEF said 90% of the planned on-site gas power stations are in the United States; if brought into operation as planned, they would emit about 345 million tons of carbon dioxide equivalent each year, equivalent to the annual greenhouse gas emissions of roughly 72.8 million fossil-fuel vehicles.
The report was released during the 31st United Nations Climate Change Conference held in Türkiye. It projects that AI data center electricity demand will approach 2,354 terawatt-hours by 2035, accounting for about 7% of global electricity demand.
Mark Daly, head of data centers and emerging technologies at BloombergNEF, said the economics of AI make “getting power quickly” a major constraint for developers, and climate outcomes will depend on the speed of grid connections and whether the power system can add clean energy at the same time.
The study noted that although developers say gas-fired power is only a temporary solution while waiting for grid connection, the economics of the stations after they begin operating may encourage continued operation, creating long-term fossil fuel lock-in. The report argues that faster grid connection is a more sustainable option for limiting the emissions impact.