AI’s Power Crunch Puts Nuclear Back in Focus

The market is waking up to a simple bottleneck in artificial intelligence: power. Training and inference workloads are concentrating in clusters that need 24/7 electricity, not only renewable megawatt-hours but reliable capacity when the grid is tight. That is why nuclear is back in the conversation, alongside batteries and gas peakers, as hyperscalers and utilities plan the next wave of data center buildouts.

Recent policy and corporate disclosures point to a meaningful shift. The International Energy Agency estimates data centers will require over 320 gigawatts of additional electricity generation capacity between 2024 and 2035, most of it in the United States and China, with low-emissions sources carrying a growing share after 2030. The U.S. Energy Information Administration also notes computing electricity use in commercial buildings could rival space cooling and ventilation, underscoring a structural load story tied to AI rather than a one-off spike.

For investors, the question is which parts of the nuclear value chain have catalysts that could matter over the next few years. Below we outline where large regulated utilities, integrated competitive generators, microreactor manufacturers, and fuel suppliers fit as the market weighs how to power AI’s next leg.

Southern Company: new U.S. baseload is here

The Southern Company ( $SO The Southern Company $95.91 ) delivered a tangible proof point. Georgia Power said Vogtle Unit 4 entered commercial operation on April 29, 2024, making Plant Vogtle the largest single source of clean electricity in the country as Units 3 and 4 join the existing reactors. The facility is expected to produce more than 30 million megawatt-hours annually, a multi-decade baseload resource for the Southeast. Recent figures show the shares are up about 7% year to date, suggesting the market is starting to price in durable earnings from completed assets even as new megaproject risk fades.

Constellation Energy: tech PPAs pull nuclear into AI

Constellation Energy ( $CEG Constellation Energy Corporation $268.69 ) operates the largest U.S. fleet of nuclear plants and is increasingly linking that output to corporate demand for hourly carbon-free power. Microsoft disclosed a power purchase agreement with Constellation intended to enable the restart of an 835 megawatt nuclear facility in Pennsylvania that was retired in 2019, adding net-new reliable, carbon-free supply to the PJM grid that hosts major data center clusters.

If more of these 24/7 contracts land, nuclear could become a bigger part of hyperscalers’ decarbonization and capacity strategies. These deals can be structured to match hourly consumption, which matters for data centers that are measured on both energy and capacity metrics. It also gives owners of existing reactors a pathway to monetize reliability and carbon attributes in a way that wind and solar sometimes cannot without firming.

The market is still sorting out the earnings cadence, with recent data showing Constellation shares down about 22% year to date. Investors may want to monitor how PPA structures, plant uprates, and any restart timelines flow through production volumes and cash generation.

Vistra: nuclear-plus-retail can serve AI load

Vistra Corp. ( $VST Vistra Corp. $167.77 ) has been knitting together a zero-carbon generation platform with a large retail footprint, a model that could match data center needs as grid interconnections tighten. The company closed its Energy Harbor deal on March 1, 2024, adding approximately 4,000 megawatts of nuclear capacity and about 1 million retail customers, bringing Vistra’s total nuclear capacity to more than 6,400 megawatts across PJM (the Mid-Atlantic grid operator) and ERCOT (the Texas grid operator). The portfolio sits inside Vistra Vision alongside solar and storage, integrated with dispatchable gas in its legacy fleet. Momentum has improved, with the stock up about 20% over the past month in recent trading. The market will look for evidence that nuclear-plus-retail can translate into steadier margins as data center demand scales and as PJM and ERCOT capacity constructs evolve.

BWX Technologies: microreactors inch toward deployment

BWX Technologies ( $BWXT BWX Technologies, Inc. $204.77 ) is the most visible publicly traded name tied to defense-backed microreactors that could eventually support remote or campus-scale power, a format some data center operators are exploring. The company has been advancing the Department of Defense’s Project Pele demonstration and disclosed progress toward fabricating the reactor core, with TRISO fuel production completed and shipment to Idaho National Lab planned. TRISO is a robust, particle-based uranium fuel used in many microreactor designs to enhance safety margins at high temperatures.

While commercial data center deployments would require further regulatory and siting progress, BWXT’s core government work gives it engineering credibility and a potential bridge to civil markets. Recent figures show the stock up roughly 46% over the past year, but execution risk remains around timelines, licensing, and the pace of any non-defense orders.

Centrus Energy: fuel becomes a gating item

Centrus Energy ( $LEU Centrus Energy Corp. $170.38 ) has emerged as a domestic supplier of high-assay low-enriched uranium, or HALEU, a fuel type many advanced reactors will require. The company reported a key milestone by delivering 900 kilograms of HALEU to the U.S. Department of Energy and secured a production extension into 2026 for its Piketon, Ohio cascade, reinforcing the view that fuel availability may set the cadence for new reactor deployments.

The stock can be volatile around contract and policy headlines. In the latest session, shares rose about 12%, although recent data show they remain down roughly 21% year to date. Investors may want to watch for multi-year offtake agreements and how federal procurement shapes the economics of first-of-a-kind reactors that target data center and industrial loads.

Where the case can break

Nuclear solves for reliability and carbon, but it does not erase cost or timeline risks. The EIA’s Annual Energy Outlook reference case does not assume extraordinary AI-specific load beyond what is embedded in commercial demand, a reminder that forecasts could overshoot if efficiency gains or workload placement change the curve. Even with the IEA’s view that low-emissions sources pick up post-2030, project finance, transmission, siting, and public acceptance are real constraints.

For large reactors, the lesson from Vogtle is that the payoff is measured in decades, but so are the schedules. For microreactors, regulatory pathways and standards are still taking shape, and many designs depend on HALEU supply that only recently started to scale in the United States. On the buyer side, corporate PPAs that enable restarts or uprates need to be durable through cycles. The near-term read-through is that utilities and integrated power producers with existing nuclear fleets may have the cleaner line of sight, while advanced reactor and fuel names remain higher-beta research candidates tied to policy and milestone risk.