The U.S. Department of Energy is rationing a specialized nuclear fuel called high-assay low-enriched uranium, or HALEU, to a small group of companies at the exact moment technology giants are racing to lock in nuclear power for artificial intelligence data centers. Domestic production of this fuel remains tiny, and the workforce needed to build and operate new reactors and the facilities that support them is not growing fast enough to match demand. Together, these two bottlenecks threaten to slow what many in the energy sector had expected to be a rapid nuclear buildout.
A Federal Fuel Program That Signals Scarcity
Rather than letting the market sort out supply, the DOE is directly selecting companies and sequencing deliveries of HALEU through a federal availability program, with multiple recipients requiring delivery in 2025. The very existence of a government rationing mechanism tells a story the industry’s optimistic press releases often skip: there is not enough fuel to go around.
HALEU is enriched to between 5 and 20 percent uranium-235, a level higher than the fuel in conventional reactors but below weapons grade. Most advanced reactor designs from companies courting Big Tech contracts depend on it. Yet the DOE’s own environmental impact statement for its HALEU program caps the upper-bound acquisition target at up to 290 metric tons, a figure that reflects both ambition and the physical limits of what the supply chain can deliver. The agency has stated its objective is to build a domestic HALEU market, detailing the full front-end fuel cycle from mining through conversion, enrichment, deconversion, storage, and transport.
The scale of actual production so far is sobering. The HALEU Demonstration Project at the Centrus Energy facility in Piketon, Ohio, began enrichment operations with a target of just 20 kilograms by the end of 2023. That is roughly the weight of a large suitcase, set against a future need measured in hundreds of metric tons. Centrus, operating as American Centrifuge Operating, LLC, holds a license from the Nuclear Regulatory Commission, which received a December 19, 2024, license amendment application for continuing HALEU operations into Phase III, according to NRC project records. Until that expansion clears regulatory review and physical buildout, output will remain minimal.
Russia Ban Tightens an Already Thin Pipeline
Congress added pressure to the supply picture by enacting the Prohibiting Russian Uranium Imports Act, signed into law as Public Law 118-62. The legislation includes explicit calendar-year quantitative import limits and enforcement provisions designed to phase out reliance on Russian enrichment services, which had supplied a significant share of U.S. reactor fuel for decades.
The strategic logic is straightforward: the United States does not want its nuclear energy ambitions tethered to a geopolitical adversary. But the practical consequence is that a major source of enriched uranium is being curtailed before domestic alternatives can fill the gap. DOE’s rationing program and its push to stand up facilities like the Piketon demonstration plant are direct responses to that squeeze. The timeline, however, does not favor speed. Building enrichment capacity requires years of construction, licensing, and workforce training, and every month of delay widens the gap between what AI-hungry data centers need and what the fuel cycle can provide.
Electricians, Welders, and a Workforce Too Small
Even if fuel supply catches up, the nuclear and data center buildout faces a separate constraint: not enough skilled workers. The Bureau of Labor Statistics projects that the economy will need to fill tens of thousands of annual openings for electricians through the early 2030s, driven by retirements, career changes, and new demand. Electricians are essential for both data center wiring and nuclear plant construction and refurbishment, meaning the two sectors are competing for the same limited labor pool.
The DOE’s 2025 U.S. Energy and Employment Report provides the most recent federal snapshot of nuclear sector employment levels and trend direction. The report is commonly cited for nuclear workforce counts and occupational breakdowns, and it allows analysts to assess whether the nuclear labor pool is expanding fast enough to meet the combined demand from plant restarts, new builds, and data center projects. The short answer, based on the trend data, is that growth has been modest relative to the scale of announced projects.
This creates a feedback loop that most coverage of Big Tech’s nuclear deals ignores. When Google, Microsoft, or Amazon sign power purchase agreements with nuclear developers, they are implicitly betting that thousands of additional electricians, pipefitters, and radiation protection technicians will materialize on schedule. If those workers do not appear, construction timelines stretch, costs rise, and the clean-power promise behind the deals erodes. The Department of Labor tracks broader workforce trends, but no federal agency has published a projection specifically modeling how AI data center competition affects nuclear hiring rates or wage inflation in the trades.
Why the Usual Optimism May Be Premature
Much of the public conversation around nuclear energy and AI treats the pairing as inevitable. Tech companies announce reactor deals; startup founders describe modular reactors as the answer to data center power needs; stock prices of uranium miners tick upward. What gets less attention is the industrial reality underneath those announcements.
The fuel cycle is a chain with many links, each requiring its own infrastructure, regulatory approval, and skilled labor. The DOE has mapped these steps in detail, from uranium mining through enrichment and eventual waste management, and the HALEU environmental review makes clear how many separate facilities and transportation routes must function smoothly to support even a modest fleet of advanced reactors. Any weak link, whether a shortage of conversion capacity, a delay in deconversion plants, or a gap in specialized transport casks, can slow the entire system.
Financing adds another layer of complexity. Utilities and data center operators are not just buying electrons; they are underwriting a long-lived industrial ecosystem. To help them understand the risks, DOE has begun publishing more granular information on project schedules, permitting status, and supply-chain needs through tools such as the Genesis project portal, which aggregates data on clean energy deployments. That information can inform investor decisions, but it also underscores how much has to go right, in the right order, for nuclear projects to deliver power on the timelines AI companies are advertising.
Research and development pipelines are similarly stretched. Advanced reactor vendors depend on a steady flow of government-backed experiments, test data, and materials science breakthroughs. Much of that technical backbone is documented in public reports accessible through the DOE’s scientific information repository, which shows how many designs are still evolving and how many safety and performance questions remain under study. Enthusiastic corporate announcements rarely mention that the underlying science is still being refined.
Can Policy Close the Gap?
Policymakers are not blind to these constraints. In addition to HALEU procurement and Russia-related import restrictions, the federal government is experimenting with new ways to coordinate capital and infrastructure. The DOE’s Infrastructure Exchange platform centralizes information on funding opportunities for energy projects, including nuclear and grid upgrades that data centers will depend on. By giving developers a clearer view of available grants, loans, and tax incentives, the agency hopes to accelerate projects that might otherwise stall for lack of financing.
On the innovation side, programs run by the Advanced Research Projects Agency, Energy, are targeting some of the technical bottlenecks that slow nuclear deployment. ARPA‑E’s portfolio, summarized in its program overview, includes efforts to improve reactor materials, thermal storage, and grid integration—areas that could, over time, make reactors more compatible with the fast-ramping, high-availability needs of AI workloads. Still, these are multiyear research bets, not near-term fixes for today’s fuel and labor shortages.
For now, the mismatch between rhetoric and reality persists. Data center operators can sign long-term contracts for nuclear power, but they cannot conjure HALEU enrichment cascades or licensed welders out of thin air. Federal programs can prioritize which projects get scarce fuel first, but that does not increase the total number of kilograms available this decade. Workforce initiatives can expand training slots, yet they must contend with retirements and competition from other booming sectors, from renewables to conventional construction.
That does not mean nuclear energy has no role in decarbonizing AI. It does mean that timelines should be treated with skepticism, especially when they assume a smooth scale-up from demonstration projects to gigawatt-scale deployment. Analysts and policymakers who want a clearer picture will need to look past corporate announcements and into the technical reports, workforce statistics, and project trackers that quietly document the system’s real pace. Those sources, from federal employment surveys to DOE data portals, consistently point to the same conclusion: without a dramatic acceleration in fuel production and skilled hiring, the nuclear buildout for AI will move slower than its most optimistic boosters claim.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
