The most important nuclear development of the past 24 hours is not another SMR design milestone. It is the United States putting up to $4.4 billion behind the restart of three already-built nuclear plants, because the nuclear revival has finally reached the only test that matters: can it return reliable megawatts to the grid at a price customers will pay?

The plants are Palisades in Michigan, the Crane Clean Energy Center at Pennsylvania’s former Three Mile Island site, and Duane Arnold in Iowa. Together, they represent roughly 2.25 gigawatts of potential capacity, enough to supply about two million homes and businesses. Palisades is targeting a March 2027 return, while Duane Arnold is tied to a partnership between NextEra Energy and Google that targets early 2029. Microsoft has separately agreed to buy power from the restored Crane facility.

This is real progress, not because a government loan makes nuclear economic by definition, but because these projects begin with the asset that has defeated most nuclear proposals: a nuclear island, a licensed site, grid interconnection, cooling infrastructure, an operating history, and a workforce that can be rebuilt. Restarting a shut plant is still technically difficult. Fuel, inspections, equipment qualification, licensing, and deferred maintenance do not disappear because a data center wants clean power. But the risk profile is fundamentally different from building a first-of-a-kind reactor on an empty site.

The hype is that this proves every retired reactor should reopen. It does not. The fear is that public financing means the plants cannot compete. That is also too simple. A restart loan can be a bridge across a timing and capital problem, not a permanent subsidy. The serious question is whether restored plants can deliver high-capacity-factor power before new transmission, gas supply, or renewable overbuild can do so. The answer will decide whether nuclear’s next growth phase is built around expensive new reactors or around preserving and reactivating the fleet that already works.

The economics of a restart are less about headline levelized cost and more about avoiding replacement cost. A gigawatt-scale nuclear plant operating at a 90 percent capacity factor produces about 7.9 terawatt-hours annually. Replacing that output with intermittent generation requires some combination of overbuilding wind and solar, storage, firm gas capacity, transmission, or curtailment. A plant restart therefore competes not against the marginal operating cost of a solar project, but against the full system cost of obtaining dependable electricity around the clock.

That distinction matters enormously for Microsoft and Google. Their demand is not merely “clean energy” on an annual accounting basis. Artificial intelligence data centers require power at a quality and continuity standard that intermittent generation cannot provide by itself. A power purchase agreement with a nuclear plant is valuable because it reduces the amount of balancing infrastructure the customer must procure elsewhere. The buyer is paying for firmness, not just carbon attributes.

Still, nuclear operators should not pretend these assets are cheap. Palisades was shut in 2022 after more than five decades of operation. Crane is a restart of a retired plant at a site that hosted the Three Mile Island Unit 1 reactor, and Duane Arnold closed in 2020. Each project carries refurbishment risk, regulatory risk, and schedule risk. A loan guarantee lowers the cost of capital and can make the project financeable, but it cannot repair a steam generator, qualify a component, or solve an aging-cable problem.

The economic advantage is that restart projects avoid the worst cost trap in new nuclear: first-of-a-kind construction. Vogtle 3 and 4 demonstrated that an AP1000 can operate, but they also demonstrated how design immaturity, supply-chain gaps, and construction learning curves can turn a nominally standardized project into a multibillion-dollar ordeal. The Nuclear Regulatory Commission’s renewal and revision of Westinghouse’s AP1000 design certification, incorporating lessons from Vogtle, is useful precisely because it converts painful experience into a more deployable reference design.

That is a regulatory milestone, not a completed project. Investors should treat it as design-risk reduction, not evidence that the next AP1000 will automatically be cheap. The same discipline applies to advanced reactors. The Department of Energy says Kairos Power completed criticality testing for its eVinci microreactor, while a reported $643 million funding package is intended to support Hermes 2 and expanded manufacturing. Hermes 2 is planned as a 50 MWe demonstration reactor with a target around 2030, and the company’s Google agreement contemplates up to 500 MW by 2035. Those are meaningful steps, but none is commercial fleet generation today.

The industry keeps confusing three different achievements: raising money, obtaining permission, and producing electricity. The restarts are closer to the third category than most SMR announcements. That is why they deserve more attention.

The immediate winners are Constellation, Holtec, NextEra, and the suppliers that can execute nuclear refurbishment without treating every outage as a science project. Palisades benefits from being furthest along, but its success will depend on disciplined maintenance execution and NRC oversight after the fuel-loading incident that required an emergency approval for removal of a tilted assembly. The lesson is not that the project is failing. The lesson is that restart work produces real nuclear problems, not presentation-grade milestones.

Constellation’s Crane project has a strategic advantage in Microsoft as an anchor customer. That structure matters because a long-term buyer can support financing and reduce merchant-market exposure. But a corporate offtake agreement does not eliminate the need for competitive pricing. If the delivered power price is too high, other data-center customers will eventually demand concessions, and the burden shifts back to ratepayers or taxpayers.

Duane Arnold is the clearest test of whether a closed plant can become a platform for the next wave of nuclear demand. NextEra and Google are attempting to align a utility asset with hyperscale load growth. If they can return the plant on schedule, the template becomes powerful: identify retired or retiring reactors near constrained load pockets, secure an anchor customer, use public finance to reduce construction risk, and preserve the existing nuclear workforce. That is far more credible than promising a fleet of reactors before the first unit has cleared commissioning.

The restarts also improve the position of large reactors relative to SMRs. That sounds counterintuitive, but it is strategically important. Every successfully restored gigawatt strengthens the case that the fastest route to more clean firm power is a portfolio: extend existing reactors, restart viable retired units, build proven large reactors where scale justifies them, and develop SMRs in parallel for locations that cannot support a conventional plant.

Poland’s agreement involving SGE, ZE PAK, and up to six BWRX-300 units shows where SMR momentum is heading, toward repeatable deployment rather than isolated demonstrations. But the BWRX-300 still has to prove construction, licensing, fuel, and operating economics at commercial scale. The companies that win will not be the ones with the most sophisticated reactor animation. They will be the ones that turn a design into a supply chain, a site, a license, and a revenue-generating unit.

The signal from Washington is therefore more important than the dollar amount. Governments and customers are beginning to pay for nuclear’s actual product, dependable electricity, rather than rewarding nuclear companies merely for promising future technology. That shifts capital toward operators with assets, licenses, customers, and execution records.

Watch the first restart schedule, the final refurbishment cost, and the power price paid by the anchor customers. If Palisades returns in 2027 and Crane and Duane Arnold follow without major cost escalation, the nuclear revival will have a repeatable near-term growth engine. If they slip badly, the industry will learn that even existing nuclear assets require more than political support.

The nuclear comeback is moving from narrative to infrastructure. That is real progress. It is also where the easy optimism ends.