For the better part of a generation, the story of American nuclear power was one of managed decline. Reactors that had run reliably for decades were closing early, undercut by cheap shale gas, punished by energy-only markets that assigned little value to firm, carbon-free capacity, and stranded by a policy environment that rewarded marginal cost over resilience. Between 2013 and 2022, roughly a dozen units went dark, and the industry’s obituary was written so often it began to read like liturgy. This week, that narrative reached its most striking inflection point yet as the Palisades plant in Michigan moved within striking distance of becoming the first permanently closed commercial reactor in United States history to return to service. Palisades is no longer an outlier. It is the vanguard of a broader realignment in which nuclear power for data centers has become a defining commercial and reliability question of the decade, and in which the atom has been recast from stranded asset to strategic scarcity.
The Vanguard at Palisades
The Palisades restart is remarkable precisely because it was, until recently, considered impossible. When the roughly 800-megawatt plant on the shore of Lake Michigan powered down in May 2022, decommissioning was assumed to be the endpoint; no operator had ever reversed that process. Holtec International, which acquired the site to dismantle it, instead reversed course, backed by a federal loan guarantee of up to $1.52 billion and a series of Department of Energy disbursements that have continued through 2026. Over the past year, the project has moved through a methodical checklist of milestones: steam generator repairs, fresh fuel delivery and loading, and the dismissal of a court challenge to the Nuclear Regulatory Commission’s exemption pathway. Chief executive Kris Singh has signaled a 2026 restart target, subject to completion of remaining work and NRC oversight, ahead of a March 2027 contractual commitment to supply Michigan’s electric cooperatives. The symbolism is difficult to overstate. A restart would show that retired or at-risk nuclear capacity is not necessarily lost, and it would give regulators, financiers, and utilities a template for treating existing reactors as recoverable infrastructure rather than sunk cost. Every kilowatt-hour Palisades sends out will be firm, around-the-clock, and carbon-free—the precise combination the modern grid is discovering it cannot easily buy on the open market.
A Second Act on the Susquehanna and Beyond
Palisades is the leading edge of a formation, not a solitary event. In Pennsylvania, Constellation Energy is working to bring Three Mile Island Unit 1 back online under its rechristened name, the Christopher M. Crane Clean Energy Center, with the roughly 835-megawatt unit potentially returning as early as 2027, ahead of the company’s original 2028 expectation. That project cleared a meaningful hurdle when the Federal Energy Regulatory Commission granted a waiver allowing Constellation to transfer capacity interconnection rights from Eddystone units to Crane, a step that could help the unit become fully operational before the end of 2030 under its twenty-year power purchase agreement. The project has come to symbolize the new logic binding generation to computation. The economics that once killed these plants have inverted. Where reactors once struggled to clear wholesale markets, they now command long-dated contracts at premium prices from counterparties that value certainty above all. Similar forces are visible in renewed interest at Duane Arnold in Iowa and in the steady stream of uprates and license extensions across the operating fleet, as owners squeeze additional megawatts from assets they had recently marked for retirement. What unites these projects is a single realization spreading through boardrooms and control rooms alike: in a grid straining under unprecedented demand, the cheapest new firm capacity is often the firm capacity that already exists and can be restored. The math is compelling. Recovering a mothballed reactor or extracting additional megawatts from an operating unit can sidestep the decade-long permitting and construction gauntlet that greenfield generation faces, while delivering power through interconnection rights established years ago. For utilities and grid operators contending with interconnection backlogs and supply-chain bottlenecks across nearly every other resource class, a reactor that can be restored on a timeline measured in years rather than a decade looks less like a relic and more like one of the fastest firm megawatts available on the system.
The Hyperscaler Compact
Driving the reversal is a wave of capital from an unexpected quarter. According to industry trackers, the largest technology companies, confronting electricity as the binding constraint on their artificial-intelligence ambitions, have committed to roughly 9.8 gigawatts of nuclear capacity across more than a dozen announced deals as of mid-2026, although only about 1.9 gigawatts is delivering power today. Amazon anchored the trend with a major arrangement tied to Talen Energy’s Susquehanna plant, active since June 2025, and layered on an investment in X-energy to develop advanced reactors in Washington State. Microsoft’s Crane agreement remains the emblematic transaction, but Meta has arguably been the most aggressive, with tracker estimates putting its nuclear-related commitments at up to roughly 6,600 megawatts across Constellation’s existing fleet and the advanced designs of TerraPower, Oklo, and Vistra. Google, for its part, contracted with Kairos Power for 500 megawatts across a series of small reactors slated for the early-to-mid 2030s. These hyperscaler nuclear power purchase agreements represent something structurally new in the electricity business. They are creditworthy, multi-decade offtake commitments from some of the most valuable enterprises on earth, and they address the financing problem that has bedeviled nuclear construction for half a century by de-risking revenue before the first shovel breaks ground. In doing so, they have quietly transferred a measure of grid-planning authority from regulators and utilities to the balance sheets of a handful of technology firms—a shift whose long-term implications for ratepayers and reliability planners are only beginning to be debated.
The New Economics of Firmness
Understanding why capital is flowing back into a business it fled a decade ago requires understanding what changed beneath it. The energy-only markets that once starved nuclear plants priced electricity at the marginal cost of the last unit dispatched, a design that rewarded low-fuel-cost gas and zero-fuel-cost renewables while often assigning limited explicit value to the availability, inertia, and around-the-clock output that reactors uniquely supply. As variable resources have grown and demand has surged, that blind spot has become harder to ignore. The premium for firmness that hyperscalers are now willing to pay through bilateral contracts is, in effect, a private correction to a public market gap. Federal policy has amplified the shift. The Department of Energy’s loan authority has underwritten the Palisades revival, while federal tax incentives and licensing reforms have supported both the existing reactor fleet and advanced designs. Bipartisan licensing reforms enacted in recent years directed the Nuclear Regulatory Commission to modernize and accelerate its review of advanced reactors, chipping away at the regulatory uncertainty that historically inflated the cost of capital. State governments have joined in, with several legislatures moving to court reactor development or reconsider long-standing nuclear restrictions. The result is a rare alignment of market signal, federal balance sheet, and political will. For an industry accustomed to fighting the tide, the sensation of the current finally running its way is both exhilarating and, for veterans who remember past false dawns, a reason for disciplined caution.
From Blueprint to Baseload: The SMR Question
The restarts and existing-fleet contracts buy time, but the industry’s long-term bet rests on small modular reactors, and 2026 has been the year the concept began its uneven passage from blueprint toward baseload. Holtec has folded its advanced ambitions directly into the Palisades site, submitting a license application to the NRC at the end of 2025 for twin SMR-300 units, with the application entering the review process in early 2026 and the company hoping to bring the units online around 2031—an elegant reuse of an existing licensed site, transmission interconnection, and trained workforce. Elsewhere, the field is crowded and uneven. GE Vernova Hitachi’s BWRX-300 is under construction at Ontario’s Darlington station, with a first unit targeted for late 2029 or around 2030, the closest thing the West has to a commercial reference plant. TerraPower’s sodium-cooled Natrium project in Wyoming, which broke ground in 2024, is advancing toward the decade’s end, while X-energy’s high-temperature Xe-100 continues moving through the federal review process for its Washington deployment. The sobering counterpoint arrived from abroad, where China’s Linglong One was expected to become the world’s first land-based, grid-scale small modular reactor to reach commercial operation in the first half of this year. That milestone target underscored how far American developers still have to travel from paper to power. No merchant SMR is yet delivering electrons onto the U.S. grid, and the gap between announced timelines and demonstrated hardware remains the single greatest source of skepticism surrounding the enterprise.
Constraints on the Comeback
Enthusiasm should not be mistaken for inevitability, and the constraints on this comeback are as real as its momentum. Fuel is the most acute. Many advanced designs, including X-energy’s, depend on high-assay low-enriched uranium, or HALEU, a specialized fuel whose Western supply chain remains limited and is still working to replace Russian-linked enrichment capacity that once dominated the market. Without rapid domestic scale-up, the most ambitious SMR schedules risk slipping regardless of regulatory or financial progress. Cost and schedule discipline present a second hurdle; the last large reactors built in the United States, at Georgia’s Vogtle site, ran years late and billions over budget. Modular construction promises to tame that pathology through factory fabrication and standardized designs, but the promise remains unproven at commercial scale. Workforce availability is another pressure point, as the industry tries to restart, extend, and expand nuclear assets after years of contraction. Even the restart model has limits: only a finite number of recently shuttered plants remain physically recoverable, and each passing year of decommissioning narrows that window. Layered atop these operational realities is the question of who ultimately bears the risk when firm nuclear capacity is contracted privately to serve concentrated computing loads, and whether the reliability benefits and cost burdens will be shared equitably across a grid all customers depend upon. These are not reasons for pessimism so much as a reminder that nuclear baseload for AI must be built, not merely announced.
Conclusion
The convergence on display this week amounts to a structural shift in how the United States intends to power itself. A retired reactor in Michigan is preparing to test whether shutdown need not be permanent; a rebranded unit in Pennsylvania is binding its output to the cloud through a long-term agreement; the world’s largest technology companies are underwriting reactors with the long-term certainty the sector has rarely enjoyed; and a new generation of small modular reactors is inching from license application toward first concrete. Taken together, these developments signal that nuclear power has moved from the margins of the energy debate to its center, propelled less by climate ambition alone than by the unforgiving arithmetic of an AI-driven load boom that demands firm, around-the-clock electricity. The obstacles are formidable and the timelines demanding, and the coming years will reveal whether the industry can convert announcements into operating megawatts fast enough to matter. But the direction of travel is no longer in doubt. After a decade of retreat, the atom has been handed a second act, and the grid, straining under a load it never anticipated, may find that its future depends on how quickly that act can be brought to the stage.