The fleet got a third bigger
without building anything.

The World Fleet

The construction picture is the part most often described as a global renaissance and is better described as a Chinese one. China alone accounts for 37 of the 79 units under construction worldwide, about 47 percent by unit count. Add India, Russia, South Korea, Turkey and Egypt and the top six reach 64 of 79, about 81 percent. Everything else on Earth — all of Europe, both Americas, the rest of Asia and Africa combined — accounts for fifteen units. In the United States the figure is zero, and has been since Vogtle 4 finished.

The US Fleet

The mechanical cause of the EIA-versus-NRC gap is that EIA's electric power surveys collect on Form EIA-860 at the generator level while the NRC licenses reactor units. A wrinkle specific to 2026 makes it worse: Palisades now appears on the NRC's operating list while producing no electricity, so any count including it overstates generating capability by 789 MWe, and any capacity factor computed with it in the denominator is depressed.

The Improvement Nobody Talks About

Between 1980 and 2000 the US nuclear fleet went from producing barely more than half of what it was capable of to producing nearly ninety percent of it. No new reactors were needed for that, and after about 1990 essentially none were built. It is the largest single source of US nuclear output growth in fifty years and it is almost never discussed.

How it was done is not mysterious and it is not a single trick: outage duration management, moving from twelve-month to eighteen- and twenty-four-month fuel cycles, condition-based maintenance, equipment reliability programmes, and the operational discipline that followed Three Mile Island. A fleet running at 56 percent and one running at 92 percent differ by two thirds of their own output. The United States built almost no new reactors after 1990 and still produced far more nuclear electricity in 2020 than in 1990.

Forty Years, Sixty, Eighty

US operating licenses were originally issued for 40 years — a term set by antitrust and amortisation considerations rather than by any engineering limit on plant life. Renewal adds 20 years at a time, subject to an aging management review that has to demonstrate the license renewal commitments are met for the extended period.

1. 1
   Initial license renewal, 40 to 60 years

   Now routine. The great majority of the operating fleet has been through it, and the review is a well-trodden process rather than a novel undertaking.
2. 2
   Subsequent license renewal, 60 to 80 years

   As of August 2026 the NRC had approved 13 applications covering 25 reactor units at 13 sites — Turkey Point, Peach Bottom, Surry, North Anna, Point Beach, Oconee, St. Lucie, Monticello, Summer and others — with seven of those approvals issued in the eighteen months to June 2026. Note that at least one widely cited secondary source gives a substantially lower number; the NRC's own table is the one to use.
3. 3
   Beyond eighty years

   No US plant has been licensed past 80 years. The technical questions are concrete rather than philosophical: reactor pressure vessel embrittlement, concrete containment degradation, cable qualification, and irradiation-assisted stress corrosion cracking in core internals. These are being studied; they are not settled.

This matters commercially as much as technically. A plant that has just been licensed to 80 years is a plant with decades of amortisation left, which is the background condition for every power purchase agreement being signed with a technology company — and it is why the recent capital expenditure cycle shows up in the fleet's cost data.

Shutdowns, and the Restarts That Have Not Happened

Between 2013 and 2022 a wave of US reactors closed. The dominant cause was economic: sustained low natural gas prices after 2008, combined with energy-only merchant market designs that paid nothing for firm capacity or for zero-carbon attributes. Single-unit merchant plants in restructured markets were the vulnerable class, being sub-scale on fixed operations and maintenance cost per megawatt-hour with no rate base to fall back on.

• Economics — Kewaunee, Vermont Yankee, Oyster Creek, Pilgrim, Three Mile Island 1, Duane Arnold, Palisades.
• Technical events with economic outcomes — Crystal River 3, where containment delaminated during a steam generator replacement and repair cost exceeded plant value, and San Onofre 2 and 3, where replacement steam generator tube wear ended the plants.
• Political and regulatory — Indian Point 2 and 3, closed under a negotiated settlement with New York State rather than for want of a license or a market.

Then it stopped. No US reactor has permanently shut down since Palisades in May 2022. The 45U production tax credit, state zero-emission credit programmes and data centre load growth between them removed the economic case for further closures — and three of the recent closures are now being reversed.

What It Costs

Vogtle 3 and 4 are the American data point for new build, and they are sobering. Construction began in 2009 with Unit 3 planned for 2016 and Unit 4 for 2017, at an estimated $14 billion. Unit 3 entered commercial operation in July 2023 and Unit 4 in April 2024, at a project cost EIA puts at more than $30 billion — about 2.2 times the estimate, roughly seven years late, and fifteen years of construction for a project planned at seven. Figures of $35 and $36 billion also circulate; those generally fold in financing, owner's costs, or the Westinghouse bankruptcy settlement, which sat outside the construction budget.

The Federal Push, and What 45U Actually Does

The Inflation Reduction Act created a zero-emission nuclear power production credit at 26 U.S.C. § 45U, for plants placed in service before 16 August 2022 — new build is excluded by design. It terminates for electricity produced after 31 December 2032. It is almost universally reported as a $15 per megawatt-hour subsidy for nuclear plants, and that description is wrong in a way that matters.

1. 1
   The base credit

   0.3 cents per kilowatt-hour, or $3.00/MWh, rising to 1.5 cents per kilowatt-hour — $15.00/MWh — where prevailing wage and apprenticeship requirements are met. Both figures are indexed for inflation from a 2023 base, so the statutory numbers are not the current numbers.
2. 2
   The clawback nobody mentions

   The credit is reduced by 16 percent of the excess of gross receipts over 2.5 cents per kilowatt-hour, which with the multiplier works out at about 80 percent of that excess — and state zero-emission credit payments count toward gross receipts.
3. 3
   So what it really is

   The credit phases out entirely once realised revenue reaches roughly 4.375 cents per kilowatt-hour. It is a revenue floor of about $43.75/MWh, not a per megawatt-hour payment. A plant selling at $50/MWh receives nothing. A plant at $30/MWh receives close to the full amount. In the high-price environment of 2024 to 2026, many merchant units have received little or none of it.

Alongside the credit, the Department of Energy has moved substantial money: an SMR funding round, a reactor pilot programme with nine selected vendors, a strategic partnership with Westinghouse aimed at AP1000 and AP300 deployment, a loan package to Southern Company described as the largest in DOE loan programme history, conditional commitments across five projects of two AP1000 units each, supply chain financing, and an uprate support package covering 345 MWe across six reactors. Several of those headline figures reach the public through secondary reporting rather than DOE primary documents; for anything consequential, verify the announcement itself.

Waste Policy

US commercial spent fuel inventory was about 86,000 tonnes at 75 sites in 33 states when GAO counted in 2021, and DOE put it near 90,000 tonnes shortly afterwards; it accumulates at roughly 2,000 tonnes a year. On that trajectory the 2026 figure should be somewhere near 95,000 to 96,000 tonnes, but no source reachable for this page publishes a current number. The 86,000 tonne figure is a 2021 number and is now roughly ten thousand tonnes low, which does not stop it being quoted as current.

The money has an odd shape too. The Nuclear Waste Policy Act fee of one mill per kilowatt-hour has been set at zero since 16 May 2014, after the D.C. Circuit ordered DOE to justify a fee it could not justify without a repository programme. Meanwhile the government is in breach of the Standard Contract for failing to take title to the fuel, and had paid reactor owners about $9 billion in damages as of GAO's 2021 count, with the figure growing annually. Those damages come out of the Treasury's Judgment Fund rather than the Nuclear Waste Fund, so the growing liability does not draw down the fund and does not appear as a nuclear programme cost anywhere obvious.

Internationally the picture is different. Finland's Onkalo is the furthest advanced deep geological repository anywhere. As of August 2026 it had completed its first full-scale trial canister emplacement in June 2026 and cleared the regulator's safety assessment in August, but the government operating licence had not been granted. It has not "opened", and reports that it has are ahead of the record.

Who Is Building, and Who Is Leaving

Of 79 reactors under construction worldwide, China has 37. It is not a global construction boom with a Chinese component; it is a Chinese construction boom with a global remainder. China has been approving around ten units a year and has been completing them close to schedule and close to budget, which is the part of the story most relevant to anyone arguing about whether nuclear construction cost is a law of nature or a consequence of having stopped doing it.

The phase-out story is no longer moving in one direction. Several countries that legislated an exit have softened, delayed or reversed it since 2022, driven by energy security after the invasion of Ukraine and by decarbonisation targets that proved hard to meet without firm low-carbon generation. Germany's exit was completed and has not been reversed. Others have been revisited. If you are citing a national phase-out policy, check its date — this is the fastest moving part of the subject and several widely repeated positions are two or three years out of date.

Where This Comes From, and What Is Missing

• US fleet, generation and cost. EIA Electric Power Annual, Monthly Energy Review, Nuclear explained and the AEO2026 levelized cost report. Where two EIA series disagree, both are given.
• Licensing. NRC's own lists of operating power reactor units, approved power uprates, and subsequent license renewals, which are more current and more reliable than any secondary compilation.
• Waste. GAO-21-603 for the inventory and the damages figure; NRC's high-level waste disposal pages and NUREG-1949 for Yucca Mountain; the Federal Register for the fee determination.
• Operating cost. NEI Nuclear Costs in Context, August 2026, using EUCG data for 2025.
• World figures. World Nuclear Association, because IAEA's Power Reactor Information System publishes through an application that cannot be read as a static page. WNA disagrees with itself on capacity under construction by about ten percent, almost certainly gross against net, unlabelled.