A $400M reactor design that died in 2017 gets an AI-era buyer, and the same cost problem

What changed hands on June 17 is a paper trail and a set of test rigs. The mPower facilities in Lynchburg, Virginia were preserved after the programme stopped ^[7], and the design carries roughly $400 million of completed engineering ^[3], which comes to about $2.05 million per rated megawatt of prior spend ^[18]. Against originating a light-water SMR from a blank sheet, that is cheap. But mPower did not stop because the engineering ran out. BWXT and Babcock and Wilcox shelved it in 2017 under cost overruns and a regulatory environment they could not price ^[4], and by the source's own account the construction cost environment has only got worse since ^[11]. A license moves the drawings. It does not move the cost of pouring concrete.

The schedule is where the deal gets interesting. First commercial operation is targeted for around 2031 ^[20], roughly five years from the signature ^[21], and it would be the first SMR design to reach full commercial operation in the United States ^[9]. The regulatory clock is the awkward input. The design was partway through NRC review nine years ago and does not resume from where it stopped ^[12], so the nine dormant years ^[19] bought preservation of hardware, not of standing.

The physics on offer is deliberately ordinary. At 575 MW thermal and 195 MW electric ^[5], mPower converts at about 34 percent ^[17], which is what a light-water cycle does. And 195 MWe is 19.5 percent of a conventional 1,000 MW plant ^[6][16]: five and a bit units to match one big station ^[16]. That ratio is the whole siting argument. It is also five and a bit sets of licensing, construction and operating overhead, which is the part modular economics has to eat.

The floating plant study with Core Power says more about the founders' read on the bottleneck than any engineering claim does. The stated rationale is permitting, on the theory that international waters and purpose-built offshore zones may move faster than the land-based NRC route ^[10]. That is a jurisdictional argument, not a reactor argument, and it is a fairly direct concession about the domestic path. The only operating precedent is Russian: the Akademik Lomonosov, in service since 2019 ^[13].

The aerospace pitch, per Kellie, is that SpaceX-grade manufacturing discipline applied to a design that already works on paper closes the gap ^[8]. Manufacturing repeatability matters at unit five, not unit one, and the first unit is the one that has to clear a regulator and a construction bid. Meanwhile the field remains crowded with ambition and short of finished plants ^[15], and US civilian nuclear still has no permanent waste answer, which no operator resolves on its own ^[14]. Buying a shelved design is a sensible way to skip the cheapest part of the problem.