Gartner's latest forecast puts worldwide data center power demand at 132 gigawatts by the end of 2026, up 27% from 104 gigawatts in 2025, and projects it will more than double again to roughly 290 gigawatts by 2030. AI-optimized servers alone are expected to account for about 31% of all data center power consumption this year — a category that barely existed at meaningful scale five years ago.
This isn't just a hyperscaler problem, even though the headlines mostly focus on the handful of companies building the largest campuses. Grid operators are already revising their own forecasts upward mid-cycle — ERCOT, the Texas grid operator, more than doubled its 2030 data center demand estimate within about a year, and utility capital spending plans tracked across 51 U.S. providers grew more than 21% year-over-year, with more than 30 of those utilities citing data centers specifically as a growth driver.
The part that eventually reaches an ordinary business's budget is electricity pricing and grid capacity in the regions where data centers cluster. When a local grid gets tight, the cost of that tightness doesn't stay contained to the companies that caused it — it shows up in regional power rates, and eventually in the pricing hyperscalers pass through for cloud compute in capacity-constrained regions.
None of this demands an immediate response from a typical IT team. But it's a reasonable input into a couple of decisions that do come up regularly: whether a new cloud workload should sit in a less power-constrained region, and whether a colocation contract renewal is a good moment to ask the provider directly about power availability commitments, not just uptime SLAs. Power is becoming a capacity conversation the way bandwidth was fifteen years ago.