Gartner forecasts global data-centre electricity use will hit 565 terawatt-hours in 2026, up 26 percent from 447TWh in 2025. By 2030, the firm expects total demand to reach 290 gigawatts of capacity and more than 1,200TWh a year, roughly Japan's entire annual electricity consumption.

What's driving it: AI-optimised servers, not data centres in general. Gartner expects AI hardware to account for 31 percent of data-centre power use in 2026, up from about 20 percent in 2025, with its consumption climbing 84 percent this year to 175TWh. Conventional servers, by contrast, are barely moving: up less than 1 percent in 2025 and a projected 1.2 percent in 2026. Cooling demand is also rising fast, up 22.6 percent to 195TWh, because denser AI hardware runs hotter and needs more to keep it cool. Gartner expects AI servers to overtake conventional servers in total power draw by 2027.

Checked and confirmed: these are Gartner's own published figures, not a secondhand estimate. The firm's lead economist, Linglan Wang, put it plainly: "Surging demand for compute-intensive AI workloads is driving unprecedented data centre power growth." Her advice to industry was equally blunt: "Infrastructure and operations leaders must prioritise efficiency upgrades and secure grid access."

The line worth sitting with: Gartner's own report states that grid supply will not be sufficient for the data-centre construction pipeline under current conditions. That's not a side comment, it's the headline risk. Whoever is building or expanding a data centre right now is competing for grid connection capacity against everyone else trying to do the same thing, and against the households and industries already on that grid. This is the same dynamic Voltage Check has reported on in Texas, where ERCOT froze new hookups to work out how much of its connection queue was genuine. Expect more of that, not less.