Seventy-one nuclear reactors are now under construction around the world, a building pace not seen in decades, as governments and utilities scramble to find firm, low-carbon electricity for a surge in demand driven largely by artificial intelligence data centers. The International Energy Agency, the U.S. Department of Energy, and the U.S. Energy Information Administration have each published analyses tying accelerating data-center power consumption to serious grid-planning challenges. The race to close the gap between rising AI workloads and available generation capacity is reshaping energy policy on multiple continents.
Why the nuclear construction wave is tied to AI electricity demand
Data centers already consume a meaningful share of total U.S. electricity, and that share is climbing fast. The DOE’s Office of Electricity has cited an estimate from the Electric Power Research Institute for the current U.S. data-center electricity share, framing it as a resource-adequacy problem that grid operators must address now rather than in the next decade. The IEA’s Special Report on Energy and AI quantifies the global trajectory of data-center electricity consumption and models its growth path under several scenarios. Both agencies treat AI-driven load growth not as a distant forecast but as a present operational stress on power systems.
That stress explains why nuclear power, with its ability to run around the clock regardless of weather, has re-entered planning conversations that a few years ago centered almost exclusively on wind and solar. Utilities facing new data-center interconnection requests need generation that can deliver steady output for 8,000-plus hours a year. Gas plants can do that, but they carry carbon risk. Renewables paired with storage can cover part of the load, yet battery duration limits and land-use constraints make them incomplete answers for the largest facilities. Nuclear fills a specific gap: high-capacity, zero-emission baseload power with decades-long operating lifetimes.
A working hypothesis worth tracking is whether utilities in regions with the fastest data-center permitting will file at least five new nuclear license applications within 24 months of the next EIA demand revision. That threshold would signal a structural shift from talk to committed capital. So far, the public record shows growing interest but not yet a wave of formal filings at that scale.
IEA and DOE data linking data centers to grid strain
The strongest evidence connecting AI power demand to new nuclear builds comes from three government-level analyses released in the current policy cycle. The IEA’s dataset-driven modeling in its Energy and AI report projects rapid growth in global data-center electricity consumption, driven by expanding computing loads across training, inference, and cloud services. The agency’s upcoming World Energy Outlook 2025 event is expected to update those demand curves with fresh data from national grid operators.
On the U.S. side, the DOE’s Office of Electricity published an analysis of clean energy resources to meet data center electricity demand, explicitly naming nuclear among the firm generation options that planners should evaluate. The DOE references EPRI’s estimate for the U.S. data-center electricity share, grounding its policy recommendations in an independent technical assessment rather than internal projections alone.
The EIA added a scenario-based dimension to the picture. Its Today in Energy analysis described how faster-than-expected data-center growth could increase fossil fuel generation unless new clean capacity enters service quickly enough to absorb the incremental load. That finding puts a sharp edge on the nuclear question: if reactors under construction today slip their schedules, the default backup is gas-fired generation, which works against national climate targets.
Gaps in the reactor pipeline data and what to watch next
The 71-reactor figure circulating in industry discussions lacks a single, universally cited primary source that ties each project directly to AI-driven demand. The IEA, DOE, and EIA analyses confirm the demand-side pressure but do not publish a reactor-by-reactor global construction tracker that attributes individual builds to data-center load forecasts. Industry groups and the International Atomic Energy Agency maintain construction databases, yet linking specific projects to specific AI power contracts requires commercial data that is often confidential.
Several questions remain open. How many of the 71 reactors will actually reach commercial operation on schedule? Historical completion rates for nuclear projects show frequent delays measured in years, not months. Which regions will see the tightest competition between data-center operators and other large industrial loads for grid interconnection rights? And will small modular reactor designs, which promise faster construction timelines, secure enough orders to change the supply picture before the end of the decade?
For energy investors, utility planners, and technology companies siting new AI facilities, the practical next step is to monitor EIA demand revisions closely. Each upward revision to data-center load forecasts strengthens the economic case for new nuclear capacity and shortens the window in which license applications can translate into operating plants before grid shortfalls materialize. The IEA’s World Energy Outlook 2025 release will be the next major data point, offering updated global demand curves that could either validate or complicate the current construction pace.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
