Electric power consumed by data centers worldwide hit roughly 415 terawatt-hours in 2024, enough to account for about 1.5 percent of all global electricity use. By the end of 2026, that figure is on track to rival the total electricity consumption of entire industrialized nations, effectively making data centers the planet’s fifth-largest electricity user if they were ranked alongside countries. The United States alone accounted for 45 percent of that demand, with China at 25 percent and Europe at 15 percent. The speed of this growth is forcing grid operators, utilities, and regulators to confront a question with no clean answer: can new supply come online fast enough to avoid locking in fossil-fuel generation that was supposed to retire?
Accelerating demand and the grid capacity squeeze
The scale of the shift becomes clear when measured over just a single decade. U.S. data centers consumed 58 TWh in 2014. By 2023, that figure had tripled to roughly 176 TWh, according to a U.S. Department of Energy report drawing on analysis by Lawrence Berkeley National Laboratory. The DOE projects U.S. data-center consumption will reach 325 to 580 TWh by 2028, a range whose upper bound would represent more than a threefold increase from 2023 levels in just five years.
That trajectory has direct consequences for power plants that utilities had planned to shut down. When load grows faster than new generation can be built, grid operators face a binary choice: keep aging coal and gas plants running or risk reliability shortfalls. The upper end of the DOE range implies the kind of demand shock that would force at least several gigawatts of fossil capacity to stay online past previously announced retirement dates in major grid territories such as PJM, which covers much of the eastern United States, and ERCOT, which operates the Texas grid. The U.S. Energy Information Administration has flagged this dynamic directly, noting that fossil generation could rise if data-center power demand grows faster than expected and new capacity remains constrained.
The International Energy Agency’s Electricity 2026 outlook reinforces this picture at the national level, stating that U.S. electricity demand growth is accelerating compared with the prior decade, with data-center expansions cited as a major driver. For consumers and businesses connected to these grids, the practical effect is straightforward: higher wholesale power prices, longer interconnection queues for new projects, and growing uncertainty about whether clean-energy targets can be met on schedule.
Global fuel mix still leans heavily on coal
The environmental stakes extend well beyond the United States. Generation serving data centers worldwide reached approximately 460 TWh in 2024 and is projected to exceed 1,000 TWh by 2030 under the IEA’s Base Case scenario, according to the agency’s energy supply analysis. That projection means the electricity needed just for data centers would more than double in six years.
The fuel mix powering that generation complicates any claim that the AI boom can be squared with climate goals. Coal accounted for roughly 30 percent of the global fuel mix serving data centers, while renewables supplied about 27 percent, gas about 26 percent, and nuclear about 15 percent. The coal share is driven largely by grids in Asia where data-center construction is expanding rapidly. Even as major technology companies sign corporate power-purchase agreements for wind and solar, the aggregate grid-level reality is that nearly a third of the electricity feeding their servers still comes from the most carbon-intensive fuel source available.
The IEA’s forward projection of 945 TWh of data-center consumption by 2030 is described in its executive summary as slightly more than Japan’s total electricity use today. That comparison puts the scale in sharp relief: a single category of industrial load, one that barely registered on national energy statistics a decade ago, is set to consume more power than one of the world’s largest economies.
Unanswered questions about 2026 demand and retirement delays
Several gaps in the available data make it difficult to pin down the exact ranking of data centers against individual countries by the end of this year. No primary source in the current reporting provides a 2025 or 2026 global consumption figure specific enough to confirm the “fifth-biggest” threshold with precision. The IEA’s 2024 baseline of 415 TWh and its 2030 projection of 945 TWh bracket the period, but the year-by-year path between those markers depends on how quickly facilities under construction come online and how aggressively hyperscale operators such as Microsoft, Google, and Amazon expand capacity.
The fuel-mix shares reported by the IEA are global aggregates, not broken out by region or tied to actual dispatch data from specific grid operators. That distinction matters because the carbon intensity of a kilowatt-hour consumed in a Virginia data center connected to a coal-heavy PJM subregion is very different from the carbon intensity of a kilowatt-hour used in a Nordic facility drawing mostly on hydropower. Without granular, location-specific data, policymakers are left to infer climate impacts from averages that may obscure the dirtiest pockets of growth.
Those blind spots complicate decisions about whether to delay planned coal and gas retirements. Utilities and regulators often rely on integrated resource plans that assume relatively modest load growth. If AI-related demand ends up near the high end of the DOE and IEA ranges, many of those plans will prove too conservative, forcing last-minute revisions that keep fossil units running longer. Yet overestimating the surge could also lead to overbuilding new gas capacity that risks becoming a stranded asset if efficiency improvements or policy changes slow demand later in the decade.
Another unresolved question is how much of the incremental load will be met by dedicated clean-energy projects developed specifically for data centers. Corporate power-purchase agreements have helped bring large volumes of wind and solar onto the grid, and some operators are now exploring on-site generation, long-duration storage, and even small modular reactors. But these initiatives start from a relatively small base compared with the scale of projected consumption. Unless they accelerate dramatically, the marginal megawatt-hour serving new AI clusters is still more likely to come from existing fossil plants than from bespoke zero-carbon resources.
Regulators are beginning to respond. Some U.S. states are considering requirements that new data centers procure a minimum share of their electricity from renewable sources or contribute to grid-upgrade costs in proportion to their demand. Grid operators are revisiting interconnection rules to prioritize projects that relieve congestion and support reliability in areas where data-center clusters are forming. In Europe and parts of Asia, energy-efficiency standards and waste-heat recovery mandates are being discussed as tools to limit the net impact on national power systems.
Even with these efforts, the core dilemma remains unresolved. The AI and cloud-computing boom is colliding with decarbonization timelines that were drafted under very different load assumptions. Without clearer, more granular data on where and how fast data-center demand is growing-and how that demand is being met-governments risk either underestimating the threat to climate goals or overreacting in ways that slow digital innovation. The next two to three years, as 2026 consumption figures come into focus, will be critical for determining whether data centers become a manageable new pillar of the electricity system or a disruptive force that locks in fossil generation for another generation.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
