Data centers worldwide consumed roughly 415 terawatt-hours of electricity in 2024, equal to about 1.5 percent of global demand. That single-year total already rivals the annual electricity use of entire midsize nations. With consumption projected to more than double by the end of the decade, grid operators, utilities, and policymakers face a widening gap between where new power-hungry facilities are being built and where clean generation and transmission capacity can keep up.
Grid strain from data center growth is arriving faster than new supply
The scale of the problem becomes clearer when the numbers are broken into national pieces. In the United States, data centers used 176 TWh in 2023, accounting for roughly 4.4 percent of the country’s total electricity, according to a Department of Energy report drawing on Lawrence Berkeley National Laboratory analysis. That figure had already tripled from 58 TWh in 2014. The DOE projects U.S. data center demand will reach 325 to 580 TWh by 2028, which would represent 6.7 to 12 percent of national electricity consumption.
Those ranges matter because the gap between the low and high estimates reflects deep uncertainty about how fast artificial intelligence workloads will scale. The low end assumes a steadier pace of conventional cloud computing growth. The high end accounts for a surge in AI training and inference clusters, each of which draws far more power per rack than traditional servers. Either way, the trajectory points to a level of demand growth that U.S. utilities have not had to accommodate in decades.
Regions where data center construction is concentrated, particularly Northern Virginia, central Texas, and parts of the Midwest, face the sharpest pressure. New facilities often connect to the grid years before planned renewable generation or transmission upgrades come online. In the interim, the most readily available source of dispatchable power is natural gas. Even under moderate scenarios for renewable additions, the timing mismatch between load growth and clean energy delivery suggests that natural-gas-fired generation share in these corridors will rise through at least 2028. Transmission planning cycles tracked by grid operators and federal agencies have not kept pace with the speed at which hyperscale campuses secure permits and begin drawing power.
Local effects can be stark. In some fast-growing hubs, individual campuses request hundreds of megawatts of capacity, enough to match or exceed the demand of nearby cities. If those loads materialize before new lines or substations are completed, utilities may have to curtail other large customers, delay new industrial connections, or lean more heavily on peaker plants. That risk is especially acute in regions where permitting and siting fights have already slowed high-voltage transmission projects.
Developers are experimenting with on-site generation, including large battery banks, rooftop solar, and in some cases behind-the-meter gas turbines, to hedge against grid constraints. Yet these measures typically cover only a fraction of a modern facility’s round-the-clock needs. Without parallel investment in regional transmission and utility-scale renewables, the net effect is still a substantial increase in grid demand and, in many markets, higher emissions intensity for each incremental terawatt-hour consumed.
IEA and DOE projections anchor the country-scale comparison
The headline comparison to a midsize country rests on two primary datasets. The International Energy Agency estimated that global data center electricity use reached about 415 TWh in 2024 and projects that figure will climb to roughly 945 TWh by 2030. The agency explicitly frames that 2030 projection as “slightly more than Japan’s total electricity consumption today,” placing data centers on track to rival one of the world’s largest economies in power demand within five years.
On the domestic side, the DOE’s report to Congress provides the granular U.S. breakdown. The tripling of consumption from 58 TWh to 176 TWh between 2014 and 2023 already outstripped most utility forecasts from that period. The projected jump to as much as 580 TWh by 2028 would, at the upper bound, put U.S. data centers alone in the same consumption class as many European nations. The wide 325-to-580 TWh range reflects competing assumptions about AI adoption rates, chip efficiency gains, and the pace at which new campuses reach full power draw.
Both agencies treat data centers as a distinct and fast-growing category of electricity demand, separate from broader industrial or commercial loads. The IEA’s decision to devote dedicated tracking to AI-related energy use signals that this is no longer a niche concern for grid planners but a primary driver of demand growth through at least the middle of the decade. In parallel, the DOE highlights data centers as a central factor in its national assessment of future load, alongside electrification of transport and heating.
These projections also underscore how quickly past assumptions can become outdated. Just a few years ago, many long-term resource plans anticipated flat or only modestly rising electricity demand in mature economies. The rapid ascent of cloud computing and AI has reversed that narrative, particularly in the United States, where the combination of cheap land, favorable tax treatment, and existing fiber routes has drawn a disproportionate share of new construction.
Key gaps in the data and what to watch next
Several important questions remain open. The IEA’s global figures stop at 2024 estimates, and no primary source in this reporting block has yet published a confirmed 2025 consumption total. The headline framing of “this year” relies on trajectory rather than a measured result. Readers should expect updated figures from the IEA and DOE later in 2025 or early 2026 that will clarify whether actual consumption is tracking closer to the low or high end of current projections.
A second gap involves the split between AI and conventional cloud workloads. The DOE report identifies the overall growth curve but does not publish a precise breakdown of how much of the 2023-to-2028 increase is attributable to AI training runs versus standard enterprise computing. That distinction matters for policy because AI workloads tend to cluster in a small number of very large facilities, concentrating grid stress in specific locations rather than spreading it across the country. Without clearer data on that split, grid operators are planning around wide uncertainty bands and may either overbuild or underbuild in key regions.
Third, while the IEA uses Japan as its benchmark for the 2030 projection, no primary source in the current reporting block provides a country-by-country mapping of the 2024 figure of 415 TWh against specific midsize nations. The comparison is directionally sound, as 415 TWh falls in the range of several advanced economies, but readers should note that the analogy is approximate rather than a one-to-one match with any single country’s latest verified total. Future statistical releases that pair sectoral demand with national baselines would make this framing more concrete.
Finally, much of the public discussion still focuses on annual energy use, while many of the hardest challenges for grids arise from hourly and seasonal patterns. Data centers typically run near constant load, which can be helpful for absorbing steady baseload generation but problematic when it locks in high demand during periods of scarcity. How operators respond-through demand response programs, flexible scheduling of non-urgent computing tasks, or deeper integration with local renewable portfolios-will shape whether the coming wave of facilities amplifies or alleviates stress on power systems.
As new numbers arrive over the next few years, three indicators will be particularly important: whether global data center demand continues to outpace overall electricity growth; how closely U.S. consumption tracks the DOE’s high-end scenario; and whether AI-heavy regions see a measurable shift in their generation mix toward gas. Together, those data points will determine whether the sector’s rapid expansion can be squared with decarbonization goals-or whether the world’s digital infrastructure becomes one of the main obstacles to meeting them.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
