IEA Projects 945 TWh in Global Data-Center Power by 2030 as U.S. Drives Half of Last Year's Growth
Energy Institute data show data-center electricity demand nearly doubled in five years, putting the IEA's 2030 target on a credible but grid-constrained path.
Global data-center electricity consumption reached roughly 787 terawatt-hours in 2025, nearly double the 410.8 TWh recorded in 2020, according to the Energy Institute's 2026 Statistical Review of World Energy — the first time in the report's 75-year history that data-center demand appeared as a discrete category. Sector consumption grew by approximately 129.6 TWh year on year, a rate of nearly 20%.5,6
The IEA's base-case forecast puts global data-center consumption at 945 TWh by 2030, roughly equivalent to Japan's total current electricity use. Getting there requires sustaining an average annual growth rate of 3.6% across the broader global economy through the decade, with data centers, electric vehicles, and air conditioning as primary demand drivers, the IEA says.1,3
But the U.S. is setting the pace. American data centers consumed 312.6 TWh in 2025, up from 249.0 TWh in 2024, a gain of 63.5 TWh, according to Energy Institute data cited by OilPrice.com. That single country accounted for roughly 49% of the entire 129.6 TWh rise in global data-center demand during the year and held 39.7% of world consumption.5
China ranked second at 205.7 TWh, or 26.1% of the global total. Europe accounted for 144.6 TWh, 18.4% of the total. The three geographies combined represent about 84% of global demand, with the remainder spread across Asia-Pacific and other regions.5
The aggregate numbers mask an even sharper trend inside the facilities. Accelerated servers, the GPU-driven compute nodes that power AI training and inference, are growing at 30% annually, against 12% for data-center electricity overall, IEA data show. By 2027, a single advanced server rack in a cutting-edge facility could draw peak power equivalent to 65 households, the IEA says in its 2026 update "Key Questions on Energy and AI."1
Japan illustrates how hyperscaler capital is beginning to redraw individual grid systems. Oracle, Google, and Microsoft were selected as official cloud providers by the Japanese government, prompting US$28 billion (4 trillion yen) in committed data-center investment. Wood Mackenzie estimates Japan's data centers will consume as much electricity as 15 million to 18 million households by 2034, accounting for 60% of the country's total power demand growth.4
The grid is where the projections run into friction. The IEA calculates that annual global grid investment would need to rise roughly 50% from $400 billion to absorb expected power demand growth through 2030. That gap has not narrowed.3
Supply bottlenecks are compounding the shortfall. Grid constraints alone could delay approximately 20% of global data-center capacity planned for construction by 2030, IEA projections show. Shortages of high-bandwidth memory, the key bottleneck on GPU cluster scaling, are expected to persist through at least 2027.2
Per-task AI efficiency has improved sharply. Yet global data-center demand still grew 17% in 2025, showing that deployment volume has outrun efficiency gains each year across the five-year period in which consumption doubled. The IEA's data documents this pattern through the full period without projecting a reversal.2,1
For power developers and utilities assessing long-term capacity additions, the IEA's 945 TWh 2030 number sits roughly 160 TWh above 2025 actuals, a gap that must close across four years of permitting cycles, transformer procurement, and interconnection queues. How the 20% capacity delay estimate moves as hyperscaler commitments accumulate, and whether memory supply eases before 2027 shifts the GPU build-out pace, are the variables most likely to determine the actual trajectory.2,3