AP1000 Fleet Economics

Westinghouse released a PwC-commissioned study outlining the economic impact of deploying 10 AP1000 reactors in the United States, representing roughly 11,500 MWe of new nuclear capacity. The construction window runs from 2026 to 2038, with an assumed 80-year operating life.

The report was politically and industrially important because it framed large-scale nuclear deployment as a national growth engine. But it was also conspicuous for what it did not emphasize: hard comparisons on capital cost, levelized cost of electricity, and how an AP1000 fleet actually competes with gas and renewables when the end customer is a power-hungry AI and data-center buildout.

That omission is where the real investment question begins. The AP1000 story is not just a nuclear story. It is a firm-power-for-AI story. If hyperscalers and grid planners need long-duration, carbon-free baseload with very high availability, then large nuclear becomes economically relevant again even if its headline LCOE is not the absolute lowest number on a simple spreadsheet.

The economics of the AP1000 depend on whether investors believe the U.S. can move from disastrous first-of-a-kind execution toward repeatable nth-of-a-kind performance.

Vogtle is the cautionary tale. It came after a roughly 30-year pause in U.S. large-reactor construction, with degraded supply chains, atrophied labor capacity, and a project-delivery ecosystem that had lost recent muscle memory. The original estimate was around $14B; final cost came in near $35B.

That is why the fleet thesis cannot be underwritten on FOAK history alone. It has to rest on NOAK learning curves, standardization, workforce reuse, modularization, and the idea that the U.S. can eventually build more like a repeating industrial system and less like a bespoke megaproject every time.

Headline LCOE comparisons are useful, but for AI and data-center demand they can be deeply misleading because they treat all megawatt-hours as economically interchangeable.

For a data center, the meaningful question is not “what is the cheapest average MWh?” It is “what delivers 24/7/365 firm power at extremely high reliability, without weather dependence, fuel volatility, or future carbon liability?”

Under that lens:

• Nuclear offers high capacity factor, long asset life, and carbon-free firm generation.
• Solar plus storage can look competitive in narrow storage-duration assumptions, but costs rise sharply once the system is sized for true round-the-clock firmness rather than intraday smoothing.
• Natural gas remains economically strong on a simple basis today, but carries fuel-price risk, emissions exposure, and policy uncertainty.

This matters because AI cluster economics are unusually tolerant of higher power costs if the power is reliable. If inference revenue and compute scarcity dominate the economics, then paying a premium for firm clean baseload can still be rational. The wrong conclusion is that the lowest nominal LCOE always wins. The right conclusion is that quality of power matters, and for hyperscalers it may matter a great deal.

The investable spread in this thesis is the distance between optimistic modeled NOAK costs and the most recent Western real-world builds.

Independent cost estimates from the source material illustrate the range:

• MIT CANES (2024): NOAK AP1000 at $4,750/kW, including owner’s costs; by the 10th unit, as low as $2,900/kW in 2018 dollars
• INL Meta-Analysis (2024): median estimate of $7,468/kW for AP1000-class large reactors
• PwC implied figure: about $3,400/kW based on direct construction GDP across the proposed 10-unit fleet

At MIT’s NOAK case, the implied capital requirement for the full fleet is:

• 11,500 MW × $4,750/kW = roughly $54.6B
• or about $5.5B per reactor

That is a world away from Vogtle’s realized cost structure, and materially below Poland’s currently indicated economics. The thesis works if the U.S. can industrialize large-reactor delivery. It struggles if every build still behaves like a semi-custom political infrastructure project.

China’s AP1000 deployments show that costs in the $3,500-4,000/kW range are possible in practice. But China is not the United States. Regulatory cadence, labor markets, supply chains, industrial policy, and state coordination are fundamentally different. Investors should treat China as proof of engineering possibility, not proof of U.S. replicability.

The policy backdrop is more favorable to nuclear than it has been in decades.

• Trump-era executive action in May 2025 targeted expedited uprates and support for 10 new large reactors by 2030
• The broader U.S. ambition is to move from roughly 100 GW of nuclear capacity toward 400 GW by 2050
• Nuclear now enjoys unusually broad bipartisan support, especially when framed through energy security, industrial strategy, grid resilience, and AI competitiveness
• Westinghouse’s ownership by Brookfield and Cameco creates a politically and financially connected anchor for supply-chain mobilization

These tailwinds do not eliminate construction risk, but they do improve the odds of permitting support, federal alignment, and long-duration policy attention. In capital-intensive infrastructure, that matters. Large nuclear projects fail as often from political and institutional incoherence as from engineering weakness.

Westinghouse itself is private, so public-market exposure comes through the surrounding supply chain, fuel cycle, and counterparties likely to sign long-term power arrangements.

Key listed exposure points include:

• Cameco (CCJ) — uranium and fuel-cycle exposure, plus partial ownership of Westinghouse
• BWX Technologies (BWXT) — reactor components and deep nuclear manufacturing capability
• Curtiss-Wright (CW) — pumps, controls, and other critical reactor systems
• Utilities likely to anchor nuclear PPAs or projects — including names like Constellation, Southern Company, and Duke depending on project geography and structure
• Fuel and enrichment plays — including Centrus Energy (LEU) and multinational enrichment providers where accessible

The investment map is broader than reactor vendors alone. If the AP1000 fleet thesis gains traction, value should also accrue to the firms controlling specialized components, nuclear-qualified manufacturing, enrichment, fuel supply, and grid-side offtake relationships.

The nuclear-for-AI baseload thesis is structurally compelling. Large AI deployments need firm power, grid reliability, and increasingly low-carbon energy. The AP1000 is one of the few designs that is both modern enough to matter and mature enough to be credibly built at scale.

But this is still a mid-2030s story, not a near-term fix for the immediate AI power crunch. Westinghouse’s own framing implies first meaningful fleet effects only toward the end of the 2030s build window. The thesis therefore depends on investors being willing to look through a long construction timeline in exchange for potentially strategic infrastructure scarcity later.

The core judgment is straightforward:

• Strategically sound: yes
• Economically proven in Western fleet conditions: not yet
• Worth tracking closely: absolutely

What would most strengthen the case from here is tangible evidence that the first post-Vogtle U.S. AP1000 can be contracted with credible pricing discipline, backed by real offtake demand tied to data centers or AI infrastructure. That is the bridge between an attractive macro narrative and an investable industrial reality.