The IEA expects data centers to use about 950 terawatt-hours of electricity in 2030. Getting power to the right place could be harder than the global total suggests.
A data center can be planned around a new generation of chips. The electricity system that supplies it moves on a different timetable. That mismatch is becoming one of the practical constraints on the expansion of artificial intelligence.
In its 2026 Key Questions on Energy and AI report, the International Energy Agency projects that worldwide data-center electricity use will rise from about 485 terawatt-hours in 2025 to roughly 950 TWh in 2030. That would put the sector at around 3% of global electricity consumption.
-
Lake Powell ended August at 22% capacity. Why 28 feet matters for Glen Canyon Dam
-
Residents of Caucaia Monitor Approval of R$ 5.85 Billion Gas Power Plant with 1,199 MW and Five Units Planned by 2031
-
Itaipu Operators Open Spillway After Rainfall, Increasing Flow to 3,100 Cubic Meters Per Second to Manage Reservoir Levels in Paraná
-
Consumers in Over 30 Rio Cities See Utility Complete R$ 11 Billion Recovery After Three Years of Judicial Oversight
A global share can hide a local bottleneck
The forecast covers all data centers, including facilities running workloads other than AI. The IEA separately expects demand from AI-focused facilities to triple over the period.
Even a relatively small global share can create a large challenge where facilities cluster. The agency warns that the different speeds of data-center development and energy investment, combined with uncertainty about future demand, can leave infrastructure and load out of step.
A developer needs a connection at the chosen site, on the date the facility is ready. Electricity available elsewhere cannot solve a local connection delay.
The US forecast has a wide range
The US Department of Energy cites an updated Berkeley Lab analysis estimating that data centers could account for 11.8% of American electricity use by 2030. Its scenarios range from 9.5% to 15.3%.
The US share is much larger than the IEA’s projected global share. The estimates cover different markets, reflecting how strongly data center demand is concentrated in the United States.
DOE also notes that the modeling is built around projected equipment shipments. It does not directly assess whether grid or on-site power supply will expand enough to meet that modeled demand. A forecast of electricity requirements is not proof that every proposed facility will be connected.
Cooling and power have to be planned together
Servers also need reliable cooling. In its discussion of geothermal energy and data centers, DOE treats continuous power and heat management as connected infrastructure needs. Securing electricity without a workable cooling design would leave a basic operating problem unresolved.
Storage is another part of that discussion; CPG has covered a proposed zinc-battery project aimed at data centers. Individual announcements still need to be assessed against their delivery schedules and operating requirements.
A new campus needs generation, a grid connection and cooling to arrive on compatible schedules. Delays in any one of them can leave expensive computing equipment waiting for the infrastructure that makes it usable.
