Electric utilities across the United States are scrambling to keep pace with data center construction that is adding tens of gigawatts of new load to a grid already running near its limits. The International Energy Agency projects that global electricity consumption from data centers, artificial intelligence workloads, and cryptocurrency mining could exceed 1,000 terawatt-hours by 2026, a scale roughly equivalent to Japan’s entire national electricity use. If U.S. facilities capture the dominant share of that growth, the sector’s demand alone would rank between Japan and Russia on a global consumption table, a comparison that carries direct consequences for fuel markets, grid reliability, and carbon emissions.
Coal plants, gas shortages, and the 2026 demand collision
The speed of this load growth matters because the power plants needed to serve it do not yet exist. Data centers consumed roughly 460 TWh globally in 2022, according to the IEA’s latest summary. A doubling of that figure within four years would land squarely in a period when new natural gas combined-cycle plants are still working through permitting, interconnection queues, and construction timelines that routinely stretch beyond 36 months.
That timing gap creates a specific, testable outcome. Grid operators in the PJM Interconnection, which covers 13 states and the District of Columbia, and in the Electric Reliability Council of Texas are already managing record-setting interconnection backlogs dominated by data center requests. When demand rises faster than new generation can come online, dispatchers call on whatever capacity is available. In both regions, that means coal-fired plants that were expected to run at declining rates or retire altogether.
The U.S. Energy Information Administration has modeled this scenario directly. Its analysis warns that fossil generation could rise with faster-than-expected growth in data center power demand, a scenario built on the agency’s Short-Term Energy Outlook baseline and its generating capacity inventory of utility- and company-reported additions. If data center load outpaces the construction of new gas and renewable capacity, existing coal units would see higher capacity factors, burning more fuel and producing more emissions per megawatt-hour than the gas plants they were supposed to be replaced by.
Monthly generation data published by the EIA would show this shift clearly. A statistically detectable rise in coal-plant capacity factors in PJM and ERCOT by late 2026 would signal that data center demand is outrunning the supply response. That signal would arrive in federal data well before any new gas plants ordered today could reach commercial operation, making it an early indicator of whether the grid transition is stalling under the weight of new digital load.
IEA and EIA projections anchor the 1,000 TWh threshold
Two primary institutions supply the numbers behind this story. The IEA’s global electricity outlook contains a detailed chart tracking electricity demand from data centers, AI, and cryptocurrencies from 2019 through 2026. That chart shows the trajectory from 460 TWh in 2022 to a potential doubling by 2026, with scenario ranges and uncertainty bands that reflect different assumptions about AI adoption rates, chip efficiency gains, and cryptocurrency mining intensity.
The agency’s follow-on work in longer-term forecasts narrows the lens to the United States, identifying data centers as a major driver of domestic demand acceleration through 2030. Those projections confirm that growth pressure is not distributed evenly across the globe. American hyperscale operators, led by companies building campuses in Virginia, Texas, Georgia, and the Midwest, account for a disproportionate share of global construction activity, concentrating new load in regions where transmission and generation were planned around much slower growth assumptions.
On the domestic modeling side, the EIA’s scenario analysis draws on two specific inputs: the Short-Term Energy Outlook baseline, which captures near-term fuel price and demand assumptions, and the agency’s generating capacity inventory, which tracks every power plant addition and retirement reported by utilities and independent power producers. The combination produces a supply-demand picture where faster data center growth directly increases the call on fossil-fueled generation because renewable and gas capacity additions cannot keep pace.
Lawrence Berkeley National Laboratory has also published research on data center energy usage that feeds into broader federal assessments of how much electricity these facilities actually consume. The lab’s work helps calibrate the gap between what operators announce and what meters record, a distinction that matters when projections depend on accurate load estimates and when policymakers are trying to understand how much of the new capacity in planning queues will ultimately be used.
Missing data between global forecasts and U.S. grid outcomes
Several gaps in the public record limit how precisely anyone can track this trend. The IEA’s 1,000 TWh threshold is a global figure covering data centers, AI training runs, and cryptocurrency mining combined. No published IEA table isolates a verified U.S.-only data center figure for 2026. The headline comparison to Japan’s consumption holds at the global level, but the exact share attributable to American facilities depends on assumptions about where hyperscale operators build next and how quickly international markets absorb AI workloads.
The EIA’s scenario analysis, while valuable for identifying the fossil-generation risk, does not publish the exact terawatt-hour load it assigns to data centers in each modeled year. Instead, data centers appear as part of a broader “other sector” or commercial category in many public tables. Analysts must therefore infer the implied load from narrative descriptions, sensitivity cases, and the difference between baseline and high-demand scenarios, a process that introduces additional uncertainty.
Another gap lies between planned capacity and realized operations. Interconnection queues in PJM, ERCOT, and other regions list gigawatts of requested data center connections, but those queues do not reveal how much of that capacity will be built, how quickly it will ramp to full utilization, or how intensively AI workloads will run. A campus that is technically capable of drawing 1 GW may operate at a fraction of that level for years, depending on customer demand and the mix of training versus inference tasks.
These blind spots complicate efforts to link global forecasts to concrete U.S. grid outcomes. Policymakers debating whether to accelerate transmission build-out or tighten data center efficiency standards cannot easily see how much of the IEA’s projected 1,000 TWh will land in specific balancing areas. Utilities planning new gas plants must decide whether to trust aggressive growth scenarios that could leave them with stranded assets if AI hardware efficiency improves faster than expected.
Despite those uncertainties, the contours of the risk are clear. If data center demand grows near the top of the IEA’s range and U.S. markets capture a large share of that expansion, existing coal fleets in PJM, ERCOT, and other regions will be pulled back into heavier service. That outcome would slow or even reverse recent declines in power-sector emissions just as many states are trying to meet mid-2020s climate targets. It would also lock in higher exposure to volatile coal and gas prices, raising costs for households and businesses that never see the inside of a server hall.
Bridging the gap between global forecasts and local grid realities will require better data and faster policy responses. More granular reporting on data center load, clearer disclosure of AI workload intensity, and standardized treatment of these facilities in federal and state planning documents would all help. So would aligning interconnection reforms, transmission expansion, and clean energy incentives with the specific geography of data center construction.
Without those steps, the 1,000 TWh threshold will arrive as a blunt shock rather than a managed transition point. The warning signs-rising coal capacity factors, delayed retirements, and growing curtailment of renewable output-will be visible in public statistics. The question is whether regulators, utilities, and technology companies will act on those signals quickly enough to prevent a data-driven wave of demand from undermining the very decarbonization goals that many of those same companies publicly endorse.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
