AI data-center demand will keep coal plants running longer, BNEF warns

AI data-center demand will keep coal plants running longer, BNEF warns Surging power demand from AI data centers threatens to delay coal retirements and add to global carbon emissions. BloombergNEF warned in a report published Tuesday (2026-05-19) that the data-center expansion needed to support artificial intelligence will keep fossil fuels in use for longer, working directly against the energy transition.³ The scale behind that warning is large. Global data centers consumed an estimated 415 terawatt-hours of electricity in 2024, a figure the International Energy Agency expects to more than double to 945 TWh by 2030.⁵ The generation to run those servers is not arriving fast enough. US utilities have committed to add 116 gigawatts of large-load connections, equal to roughly 15% of US peak electricity demand, according to Wood Mackenzie's tracking.⁴ Spending by the five largest hyperscalers, including Amazon, Microsoft and Google, is forecast to jump 50% to over $300 billion in 2025.⁴ Coal still supplies roughly 35% of global electricity, and more than 2,100 GW of capacity remains operational worldwide.¹ Renewables are now cheaper than coal in most markets, yet retirement timelines hinge on whether replacement capacity can be built before the new demand lands.¹ The intensity is striking. A single server rack inside an advanced data center could draw peak power equivalent to 65 households by 2027, the IEA said in its 2026 update, "Key Questions on Energy and AI".⁵ Ireland already spends a fifth of its electricity powering data centers, more than its urban homes use.⁶ That is a preview of what happens when load growth outpaces grid build-out.⁶ The US power sector remains a stubborn emitter. It contributes roughly 44% of national SO2, 11% of NOX and 30% of CO2 emissions, and is the largest stationary source of greenhouse gases at 25% of domestic output.² Those shares are hard to cut while adding gigawatts of round-the-clock load.² Demand growth is concentrated in developing markets. Wood Mackenzie's base case sees power-demand compound annual growth above 5% in the fastest-growing developing economies, places that already lean heavily on coal.⁴ Pressure to keep existing plants running there will only intensify.⁴ Global coal capacity has slipped from its peak only because retirements in wealthy nations have partly offset new construction elsewhere.¹ Nearly 90% of coal capacity already carried pollution controls such as scrubbers and selective catalytic reduction as of 2023, which helps local air quality but does nothing for carbon dioxide.² The carbon read is two-sided. Stronger power demand can lift ICE EUA Dec-rolling prices if it tightens the generation stack, but that logic weakens if new load connects faster than renewable and firm capacity can.³ The ICE EUA Dec-rolling contract sat at €78.73 on Thursday (2026-06-18), pricing in some incremental demand but not the shift a 945 TWh data-center load would represent.⁵ Developed economies have the capital to retire coal. Whether they have the time, as AI load arrives, is the one thing a single BloombergNEF report cannot settle.³