01 — MethodologyHow to read these numbersThe benchmarks below come from named public sources current to September 2026: Georgia Power and Westinghouse disclosures plus press coverage of the final Vogtle 3 and 4 cost; the EIA capital-cost study prepared by Sargent & Lundy (December 2023, 2023 dollars, feeding AEO2025); DOE’s Pathways to Commercial Liftoff: Advanced Nuclear reports (2023 and the September 2024 update); Lazard’s Levelized Cost of Energy+ (Version 19.0, July 2026); World Nuclear Association country cost comparisons; EDF and the French Cour des comptes on Flamanville 3; Ontario Power Generation’s Darlington SMR budget; and DOE, Constellation, and Holtec filings on the restart deals.Nuclear cost data has a wider spread than any other generation technology, and the spread is the story. Engineering estimates (EIA’s $7,861/kW for a follow-on AP1000 pair, DOE’s ~$6,200/kW for a well-executed first-of-a-kind project) sit at roughly half of demonstrated US actuals near $15,000/kW, while South Korea and China build for $2,000-4,200/kW. Where sources disagree, we show who says what rather than averaging a false midpoint.Two accounting caveats matter everywhere in this guide. Overnight cost excludes financing; all-in program cost includes interest during construction, which on a decade build can approach half the total (Flamanville’s €13.2B overnight became €23.7B with interest, per the Cour des comptes, January 2025). And per-kW figures scale directly to per-MW: $10,000/kW equals $10M per MW. For context on how nuclear compares to gas, solar, and storage capex, see our power plant cost per MW guide.What these figures are — and are notThese are market benchmarks compiled from named public sources: regulatory filings, government cost studies, national-lab and investment-bank analyses, and utility disclosures. They are not iRecruit placement data, not proprietary bid intelligence, and not an estimate for any specific project. Nuclear figures mix dollar-year vintages and overnight-versus-all-in bases, so treat everything here as planning-grade context, retrieved and cross-checked in September 2026.02 — At a glanceThe benchmarks at a glanceThe whole guide in one screen. Each row jumps to the full section.01AP1000 large reactor (US actuals)Vogtle 3 and 4: about $35B all-in for 2,234 MW, roughly $14,000-15,700/kW versus a $14B pre-construction certification~$15.7k/kW02Next-unit US estimatesEIA/Sargent & Lundy $7,861/kW overnight; DOE Liftoff ~$6,200/kW FOAK target falling toward ~$3,600/kW nth-of-a-kind$6-8k/kW03International buildsSouth Korea ~$2,300/kW domestic overnight, ~$3,800/kW turnkey at Barakah; Chinese units reported near $2,000-2,600/kW$2-4k/kW04SMR first moversDarlington BWRX-300 program C$20.9B for 1,200 MW; cancelled NuScale CFPP implied ~$20,000/kW; vendor targets $4,000-6,000/kW$17k+/kW05Restarts and upratesCrane (TMI-1) restart $1.6B for 835 MW (~$1,900/kW); Palisades backed by a $1.52B DOE loan; uprates ~$5,900/kW at Braidwood/Byron~$2k/kWAnchor numbers: a new US gigawatt-class plant has demonstrated costs near $15M per MW all-in, official estimates for the next one cluster at $6-8M per MW overnight, restarting a shut-down reactor runs about $2M per MW, and the best international builders deliver new units for $2-4M per MW.
03 — Benchmarks2026 nuclear construction cost benchmarksEach figure below is capital cost per kW of net capacity; multiply by 1,000 for cost per MW. Bases are labeled: overnight (excluding financing) versus all-in (including interest during construction). The gap between demonstrated US actuals and everything else is the defining feature of this market.$14,000-15,700/kWAP1000 all-in (Vogtle actuals)Vogtle 3 and 4 cost their owners about $31B, or ~$35B including the $3.7B Westinghouse bankruptcy settlement, for 2,234 MW (Georgia Power filings, AP). That is roughly 2.5x the $14B certified estimate, delivered seven years late. Lazard’s July 2026 study assumes $12,300-17,570/kW for new US nuclear, bracketing the Vogtle experience.$6,200-7,861/kWOfficial next-unit overnight estimatesEIA’s Sargent & Lundy study prices a follow-on AP1000 pair at $7,861/kW overnight (2023$). DOE’s Advanced Nuclear Liftoff report puts a well-executed first-of-a-kind project near $6,200/kW, declining toward ~$3,600/kW after 10-20 deployments. Both assume execution and repetition the US has not yet demonstrated.$2,000-4,200/kWInternational comparison (South Korea, China)Reported overnight costs for Korean APR-1400s at home run near $2,300/kW, with the Barakah turnkey export at ~$3,800/kW (OECD-NEA, World Nuclear Association). Chinese units under construction are reported at roughly $2,000-2,600/kW. Serial construction, stable design, and a standing craft workforce explain most of the 2-6x gap with US actuals — not looser scope.$17,000-20,000/kWSMR first-of-a-kind (projected)OPG’s Darlington BWRX-300 program is budgeted at C$20.9B for four units (1,200 MW), with the first unit at C$7.7B including shared infrastructure (OPG, 2024$, includes interest). The cancelled NuScale/UAMPS CFPP implied ~$20,000/kW ($9.3B for 462 MW) at its January 2023 update. Vendor nth-of-a-kind targets of $4,000-6,000/kW remain undemonstrated — full detail in our SMR cost per kW guide.$1,900-5,900/kWRestarts and uprates (the cheap megawatts)Constellation is spending $1.6B to restart the 835 MW Crane Clean Energy Center (TMI-1) — about $1,900/kW — backed by a $1B DOE loan (Utility Dive, ENR). Holtec’s Palisades restart carries a $1.52B DOE loan guarantee for ~800 MW. Turbine-driven uprates at Braidwood and Byron run ~$800M for ~135 MW, near $5,900/kW (Constellation).~30-45%Interest during construction shareOn decade-long builds, financing rivals concrete. The Cour des comptes put Flamanville 3 at €23.7B including interest against EDF’s €13.2B construction estimate — financing is roughly 44% of the all-in figure. Vogtle ratepayers similarly paid billions in financing costs during construction through Georgia’s NCCR rider before the units generated a single MWh.Sources: Georgia Power / Georgia PSC filings and AP coverage of Vogtle 3 & 4; EIA / Sargent & Lundy, Capital Cost and Performance Characteristic Estimates (Dec 2023, 2023$); DOE, Pathways to Commercial Liftoff: Advanced Nuclear (2023, updated Sept 2024); Lazard LCOE+ v19.0 (July 2026); World Nuclear Association and OECD-NEA country comparisons; Cour des comptes on Flamanville 3 (Jan 2025); EDF on Hinkley Point C; OPG on Darlington New Nuclear; NuScale/UAMPS disclosures; DOE LPO, Constellation, and Holtec on restart financing. Retrieved September 2026.Why the estimates and the actuals disagree by 2xEvery official US estimate assumes the project executes to plan; the two data points we actually have did not. Vogtle’s certified $14B became ~$35B, and Flamanville’s €3.3B became €13.2B before financing. The estimate-to-actual gap is not random noise — it concentrates in first-of-a-kind engineering, incomplete design at construction start, and craft productivity on nuclear-grade QA work. DOE’s own Liftoff analysis treats recent US actuals above $10,000/kW as the risk case and ~$6,200/kW as achievable only with a completed design, an intact supply chain, and a retained workforce. Budget accordingly: the actuals are the base case until a US project proves otherwise. 04 — Cost driversWhat moves the numberSix forces separate a $2,300/kW Korean build from a $15,700/kW American one, and they compound. Ranked roughly by impact on a US gigawatt-class project:~2.5xFirst-of-a-kind premiumVogtle finished at ~2.5x its certified cost largely because design, licensing, and construction ran concurrently. DOE’s Liftoff report projects costs falling from ~$6,200/kW FOAK toward ~$3,600/kW only after 10-20 consistent deployments.10-15 yrsLicensing and scheduleVogtle ran 15 years from license application to commercial operation; every added year compounds escalation and interest. NRC combined licenses (COL) de-risk scope but US builds have yet to demonstrate the 5-6 year construction cycles Korea and China achieve.9,000+Craft labor peakVogtle peaked above 9,000 workers on one site (DOE/Bechtel) — welders, pipefitters, electricians, ironworkers — assembled from a country that had not built a reactor in 30 years. Nuclear-qualified craft scarcity drove rework and overtime premiums throughout.1 vendorSupply chain and large forgingsUltra-large reactor pressure vessel and steam generator forgings come from a handful of global suppliers (Japan, Korea, France); the US has no domestic heavy-forging capacity at AP1000 scale. Long-lead orders lock in years before concrete and reprice with each delay.+reworkNuclear-grade QA and documentationSafety-related work carries NQA-1 documentation on every weld and pour; Vogtle suffered early rebar and module fabrication rework that cascaded through the schedule (Georgia PSC monitor reports). QA-driven productivity on safety-related scope runs well below commercial construction norms.~44%Financing on decade buildsInterest during construction reached roughly 44% of Flamanville’s all-in cost (Cour des comptes). Cutting build time and cost of capital does more for nuclear economics than any single engineering change, which is why federal loan guarantees anchor the restart deals.The common thread is that nuclear cost is mostly an execution and repetition problem, not a physics problem. The countries that build cheaply build continuously, with a standing workforce and a frozen design. The US is attempting to restart that flywheel from zero — which is exactly why the near-term money is flowing to restarts and uprates, where the flywheel already exists. 05 — VariationProgram-by-program cost spreadThere is no single nuclear construction market in 2026 — there are six, with a 10x spread in cost per kW between them. What each program type actually costs, from demonstrated actuals to projections:AP1000 large light-water (US actuals)$14,000-15,700/kW all-inVogtle 3 and 4: ~$35B for 2,234 MW including the Westinghouse settlement, 15 years application-to-operation. Unit 4 built measurably cheaper than unit 3 — the only US learning-curve data point that exists.EPR programs (France, UK)$9,000-15,000+/kWFlamanville 3: €13.2B for 1,630 MW overnight (€23.7B with interest, Cour des comptes). Hinkley Point C: EDF guidance of £31-35B in 2015 prices — up to ~£46B current — for 3,260 MW, with first power now expected around 2030.SMR first units (US, Canada)~$17,000-20,000/kW projectedDarlington’s four-unit BWRX-300 program at C$20.9B for 1,200 MW (2024$, with interest); NuScale’s cancelled CFPP implied ~$20,100/kW. First-unit SMRs are currently pricing like large reactors without the economies of scale.Power uprates (existing fleet)~$5,900/kWConstellation’s ~$800M Braidwood/Byron turbine-replacement uprates add ~135 MW by the late 2020s. Historically the US fleet added ~8 GW through NRC-approved uprates — capacity without new licensing risk.Restarts (Palisades, Crane, Duane Arnold)~$1,900-2,000/kWCrane Clean Energy Center: $1.6B for 835 MW with a $1B DOE loan, targeting 2027. Palisades: NRC-cleared with a $1.52B DOE loan guarantee — the first US restart of a shut-down reactor. NextEra is pursuing Duane Arnold on a similar playbook.International serial builders$2,000-4,200/kWSouth Korea ~$2,300/kW domestic and ~$3,800/kW turnkey (Barakah, 5,600 MW for ~$24B); China ~$2,000-2,600/kW on standardized Hualong One units built in 5-6 years. The existence proof that the US gap is execution, not technology.For owners and developers, the spread reads as a decision tree: restarts and uprates first, then nth-of-a-kind large reactors on proven designs, with SMRs as a bet on manufacturing learning rates. Our nuclear and SMR project tracker follows which programs are actually moving from announcement to concrete. 06 — TrendAI demand is rewriting nuclear economicsThe buyer side of nuclear changed more between 2024 and 2026 than in the previous two decades. Microsoft signed a 20-year PPA with Constellation in September 2024 to restart Three Mile Island unit 1 for its data centers; Amazon expanded its Talen Energy deal to up to 1,920 MW from the Susquehanna plant through 2042 and led a $500M investment round in SMR developer X-energy; and Meta and Google have signed their own advanced-nuclear offtakes. Hyperscalers are effectively underwriting nuclear capacity the way utilities once did — with balance sheets large enough to absorb first-of-a-kind risk.Policy followed the demand. DOE’s Loan Programs Office closed a $1.52B guarantee for the Palisades restart and a $1B loan for Crane; 2025 federal executive orders set a target of quadrupling US nuclear capacity toward 400 GW by 2050 and directed the NRC to accelerate licensing. Every operable restart candidate in the country is now spoken for or under evaluation, which is why attention is shifting to uprates, license extensions, and completing partially built sites.The constraint that remains is execution capacity: forgings, fuel (including HALEU for advanced designs), and above all people. Data-center demand can fund a nuclear build, but it cannot conjure the nuclear-qualified welders, QC inspectors, and project leadership a build consumes — and every restart that succeeds pulls experienced staff from the same shallow pool the new builds need. 07 — WorkforceThe workforce a nuclear build actually needsVogtle peaked above 9,000 craft workers on a single site — pipefitters, nuclear-certified welders, electricians, ironworkers — plus thousands of engineers, QA/QC inspectors, and project controls staff across a 15-year arc (DOE/Bechtel). DOE’s 2024 Liftoff update estimates that scaling US nuclear to its 2050 targets would require roughly 375,000 additional workers across construction and operations. Restarts are leaner but front-load the same scarce specialties: Holtec and Constellation are each rebuilding operating crews, licensed operators, and maintenance organizations plant-by-plant while the construction programs compete for the identical certifications.iRecruit is building candidate density in nuclear construction ahead of that curve — assembling its network of nuclear project managers, startup and commissioning leads, QA/QC professionals, and craft supervision from project engineer through executive level, across new-build, SMR, uprate, and restart programs. Owners staffing 2027-2032 scopes can see how we cover the sector at nuclear construction recruiters.For the hiring side of this market, see the Nuclear Construction practice. 08 — FAQFrequently asked questions How much does it cost to build a nuclear power plant?+The only recent US data point, Vogtle units 3 and 4, cost about $35 billion for 2,234 MW — roughly $15.7M per MW all-in. Official estimates for the next US build run $6,200-7,861/kW overnight (DOE, EIA), or $6-8B per gigawatt-class reactor before financing, while South Korea and China deliver units for $2-4B per GW.Why are nuclear plants so expensive to build in the US?+First-of-a-kind execution, not physics: Vogtle finished at ~2.5x its certified cost after design, licensing, and construction ran concurrently. Nuclear-grade QA labor productivity, a thin supply chain for large forgings, a craft peak above 9,000 workers, and interest during construction (~44% of Flamanville’s all-in cost) compound over decade-long schedules.What does nuclear construction cost per MW or per kW?+US actuals run about $14,000-15,700 per kW ($14-15.7M per MW) all-in. Estimates for follow-on units are $6,200-7,861/kW; restarts run ~$1,900/kW; international serial builders deliver $2,000-4,200/kW. Always check whether a figure is overnight (no financing) or all-in — the difference can approach half the total.Are small modular reactors cheaper to build?+Not yet per kW. OPG’s Darlington BWRX-300 program is budgeted at C$20.9B for 1,200 MW, and the cancelled NuScale CFPP implied ~$20,000/kW — above large-reactor actuals. The SMR bet is that factory fabrication and repetition bend the curve toward vendor targets of $4,000-6,000/kW; see our SMR cost per kW guide for the full ledger.How long does it take to build a nuclear power plant?+Vogtle ran 15 years from license application to commercial operation (about 11 years of construction); Flamanville took 17 years to first power. South Korea and China routinely construct units in 5-6 years. Restarts are the fast path — Crane targets 2027, roughly three years from announcement — which is precisely why hyperscalers bought them first. Gigawatts need people firstEvery nuclear program in the 2027-2032 pipeline will be won or lost on staffing before first concrete. iRecruit is assembling its network of nuclear construction talent — project engineers through executive leadership across new-build, SMR, uprate, and restart programs — building candidate density ahead of the demand curve.Scope a search →Candidates — register with us →
