On June 23, 2026, the US Department of Energy announced a program called American Nuclear Supply Chain Loans.

At first glance, that may sound like another nuclear-revival headline. The United States has been talking about bringing nuclear power back for years.

But this one is worth reading more carefully.

The Department of Energy said the Office of Energy Dominance Financing would provide up to US$17.5 billion in loans for American nuclear supply chain projects. According to the announcement and related reporting, the money is meant to support five projects and a total of ten large reactors using Westinghouse AP1000 technology. Seven utilities or energy companies have already signed letters of intent, though the specific companies and project sites have not been publicly disclosed.

That is a large bet.

US$17.5 billion.

Ten reactors.

Five sites.

Construction targeted before 2030.

Power generation aimed for the mid-2030s.

The deeper reason is not only climate policy. It is electricity demand.

Reports cited estimates that US data centers consumed about 4% to 5% of national electricity in 2024 and could nearly triple that by 2028. In only a few years, one sector can move from being a large customer on the grid to being a force that changes power planning itself.

This is no longer about paying a bigger electricity bill. It forces power generation, transmission, substations, grid connection, and permitting to be redesigned around a new load profile.

AI Runs Into Old Infrastructure

When people talk about AI, they usually focus on models.

But the next layer is less glamorous.

Where do the models run? In data centers.

What do data centers consume? Electricity.

Who supplies it? The grid.

Where does grid power come from? Power plants, transformers, transmission lines, substations, fuel supply, permitting, and years of infrastructure work.

The internet industry is used to weekly iteration, monthly launches, quarterly fundraising, and annual expansion.

The power industry does not move like that.

Adding several gigawatts of stable electricity to a region is not a matter of working overtime.

Nuclear power makes that mismatch even clearer. A large nuclear plant cannot be approved today, built tomorrow, and connected next week. Reactor pressure vessels, steam generators, nuclear-grade pumps and valves, control systems, fuel supply, site construction, quality assurance, and safety review all move slowly.

That is why the most important use of the DOE loan may not be building entire plants immediately. It is likely to include early procurement of long-lead components.

That detail matters.

It exposes the real question: not whether anyone wants nuclear power, but whether the United States still has the industrial capacity to build large nuclear projects continuously.

A nuclear supply chain cannot be recreated like a software team. When large projects pause for many years, equipment suppliers shrink, construction teams disperse, nuclear-grade skills weaken, and project-management experience fades.

When demand finally returns, the constraint is not just money. It is industrial muscle.

Vogtle Is Both Proof and Warning

The last major modern US nuclear construction story largely comes down to two projects: Vogtle units 3 and 4, and VC Summer. VC Summer was abandoned. Vogtle 3 and 4 survived and were completed.

Vogtle's meaning is not simple success. It answered a harder question: can the modern United States still complete a new nuclear plant?

The answer is yes, but at very high cost.

The project moved from an initial cost estimate of around US$14 billion and a planned start date around 2016 to a final cost above US$30 billion and many years of delay. It experienced several typical problems of modern nuclear construction: design changes after work had started, rework, Westinghouse's bankruptcy as contractor, module-construction problems on site, and the weakness of a US nuclear construction ecosystem that had not built continuously for decades.

Vogtle therefore became a stress-test mirror.

The AP1000 technology did not fail in a simple technical sense. But in a long-cycle, high-complexity, heavily regulated engineering system, cost, schedule, and organizational capability were all amplified.

That is why Vogtle is often used as a cautionary example when explaining why new US nuclear construction is difficult. It also helps explain why the industry has been paying more attention to small modular reactors, standardized designs, and productized nuclear concepts.

Why This Loan Program Tries a Different Approach

The Department of Energy and Westinghouse appear to be trying a different model.

One project fighting alone is too hard. Multiple sites using the same design, the same reactor type, and similar supply chains may create a better chance of continuity.

In auto manufacturing, the worst case is designing every vehicle from scratch. Platformization and repeat production matter. Nuclear plants cannot be copied like cars, but the direction is similar.

The United States wants to move large nuclear construction away from "every project feels like the first one" and toward some form of repeated production.

The uncertainty remains large.

The money has been announced. Long-lead components may be ordered. But project locations have not been disclosed.

That secrecy is not surprising. Nuclear siting is never only an engineering decision. It involves local governments, residents, environmental groups, utilities, regulators, land, water, emergency planning, and long public consultation.

Even if a local government wants jobs, investment, and tax revenue, residents and other groups may not agree. If sites are named before engineering and political work is ready, the public fight can begin before the project has a chance to take shape.

That is another way nuclear is slow: even siting is difficult.

Nuclear Is Slow, Expensive, and Still Hard to Avoid

The objections to nuclear power are serious.

It is expensive. It is slow. Permitting is difficult. Construction risk is high. Public acceptance is fragile. Waste remains a long-term issue.

Critics who say wind, solar, storage, and gas can be deployed faster and cheaper are not wrong for many near-term power gaps.

If the only goal is to meet short-term demand, nuclear is clearly not the fastest option.

But when the requirements are large scale, long duration, high reliability, and low carbon at the same time, nuclear becomes harder to avoid.

AI accelerates many things. But it is now colliding with one of the least accelerated parts of the economy: physical power infrastructure.

The US$17.5 billion loan plan is not a victory lap for nuclear power. It may still face cost overruns, local opposition, licensing delays, and supply-chain shortages.

The hard questions are still ahead.

Who absorbs overruns? Will electricity prices be acceptable to data centers? Will project sites face resistance? Can regulators move fast enough? If several projects start before 2030, can construction teams, nuclear-grade equipment suppliers, project managers, and quality systems keep up?

These questions cannot be solved by a press release.

But if this nuclear revival does work, AI's power map will change.

Sources

1. U.S. Department of Energy, Department of Energy Announces American Nuclear Supply Chain Loans, 2026-06-23, https://www.energy.gov/articles/department-energy-announces-american-nuclear-supply-chain-loans 2. U.S. Department of Energy, Office of Energy Dominance Financing, https://www.energy.gov/edf 3. The Mary Sue/AP, A massive $17.5 billion bet on American nuclear power is officially underway, but industry insiders are keeping the project sites under wraps, 2026-06-23, https://www.themarysue.com/a-massive-17-5-billion-bet-on-american-nuclear-power-is-officially-underway-but-industry-insiders-are-keeping-the-project-sites-under-wraps/ 4. Federal Register, Executive Order 14302, Reinvigorating the Nuclear Industrial Base, 2025-05-29, https://www.federalregister.gov/documents/2025/05/29/2025-09801/reinvigorating-the-nuclear-industrial-base 5. Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report, https://eta-publications.lbl.gov/publications/2024-united-states-data-center-energy