Data centers consumed 1.5% of global electricity in 2024, and AI is driving concentrated buildout in grid-constrained clusters where one in five projects face interconnection delays.
Data centers consumed approximately 415 terawatt-hours of electricity in 2024—roughly 1.5 percent of global electricity use, according to the International Energy Agency's 2025 Energy and AI report. That figure, while appearing manageable as a percentage, obscures the underlying challenge: the problem is not electricity supply itself, but the infrastructure to deliver power where new servers are clustering.
AI's computational demands concentrate in buildings filled with accelerators, networking equipment, cooling systems and electrical hardware. The United States accounts for 45 percent of global data-center electricity use, China for 25 percent, and Europe for 15 percent. Within the United States, nearly half of existing data-center capacity concentrates in five regional clusters, and half of data centers under development are planned in existing large clusters.
A typical AI-focused data center consumes as much electricity as 100,000 households. The largest facilities under construction are expected to use approximately 20 times that amount. Adding one very large load to one local grid differs fundamentally from distributing the same annual electricity use across millions of homes and thousands of substations.
The IEA projects global data-center electricity consumption to reach approximately 945 terawatt-hours by 2030—slightly more than Japan consumes today—with AI as the largest driver of growth. For 2035, agency scenarios range from 700 to 1,700 terawatt-hours depending on AI uptake, hardware efficiency, and how quickly energy bottlenecks are resolved. A separate Lawrence Berkeley National Laboratory report estimated that US data centers used 176 terawatt-hours in 2023, or 4.4 percent of the country's electricity. Its 2028 projection ranges from 325 to 580 terawatt-hours—6.7 to 12 percent of US consumption—a gap too large for grid planners to assume recent efficiency gains will quietly absorb the growth.
The real bottleneck is infrastructure. The IEA estimated that around 20 percent of planned data-center projects could face delays unless connection risks are addressed. In advanced economies, constructing new transmission lines takes four to eight years. Wait times for transformers and cables have doubled in three years. A developer can order servers and construct buildings, but supplying them may require a new substation, upgraded transformers, additional generation, transmission studies, permits and equipment with its own manufacturing queue.
Electricity is not uniformly available. A country may have sufficient generation while one county lacks the substation, transformers and transmission capacity to support another large campus. A facility with a contract for sufficient renewable electricity over a year may still draw from a constrained local grid during particular hours—annual matching and physical delivery answer different questions.
Some responses are emerging. Google researchers described a production system that forecast electricity carbon intensity and delayed flexible computing work to lower-carbon hours, shifting capacity while preserving total daily computing capacity. However, this cannot be universal: search, live model responses and other latency-sensitive services cannot pause when electricity is scarce. Long training runs are also costly to interrupt, especially when thousands of accelerators must remain synchronized.
An AI-focused data center can be ten times more capital-intensive than an aluminum smelter. Owners have invested heavily in keeping expensive chips busy, making requests to reduce output costly. Without clear rules around payment, faster connection, and protection of customer service and equipment, spare server capacity is not a dependable grid resource.
Siting matters. Placing new campuses where transmission and generation are available avoids bottlenecks before they exist. Yet clustering occurs for concrete reasons: fiber connections, skilled workers, land availability, tax policy and proximity to existing infrastructure.
Data-center demand grows quickly, but forecasts remain wide. Efficiency will improve, but cheaper and more capable AI may also increase use. New generation can be built, but a power plant hundreds of kilometers away does not help until the network can deliver its output.
The decisive questions are specific, not universal: Which grid will serve the facility? What must be upgraded? Who pays? How much workload can move? What happens during the system's hardest hour? AI arrives on a screen in seconds. The infrastructure beneath it still moves at the speed of substations, transformers and public decisions.