The SMR Graveyard: Projects That Failed and What We Learned From Them
Before the next wave of small modular reactors gets built, it's worth understanding exactly why the last ones didn't.
Nuclear energy has a long and occasionally brutal tradition of optimism outrunning reality. The SMR era is no exception. Over the past two decades, a series of projects that promised to rewrite the economics of nuclear power quietly — and not so quietly — collapsed under the weight of ballooning costs, missing customers, and the particular cruelty of being first-of-a-kind. The graveyard, as one nuclear engineering conference recently called it, is “filled with brilliant designs that couldn’t actually be built — or cost three times the estimate.”
That’s not a reason to give up on SMRs. The technology getting built right now — TerraPower in Wyoming, Ontario Power Generation’s BWRX-300 at Darlington, the microreactor programs multiplying across the U.S. military — is meaningfully different from the projects that collapsed. But the lessons from the failures are worth knowing. Anyone who dismisses the graveyard as ancient history, or waves it away as “that was then,” is probably building the next tombstone.
This is a tour of what went wrong, and what the industry learned by living through it. 🪦
The mPower experiment: too clever, not enough customers
Long before NuScale became the name everyone associates with SMR failure, Babcock & Wilcox was the company everyone was watching. Its mPower reactor — a 195 MW integral pressurized water reactor developed through a joint venture with Bechtel called Generation mPower LLC — was, by 2012, the publicly anointed front-runner. The New York Times called mPower the leader in the SMR race. The U.S. Department of Energy selected it for a five-year cost-share agreement, committing up to $226 million in federal support. The Tennessee Valley Authority signed on to explore permitting at its Clinch River site in Oak Ridge.
And then it started to unravel. Not dramatically. Slowly, the way bad nuclear projects usually die.
In 2013, Babcock & Wilcox tried to sell a majority stake in the mPower joint venture and couldn’t find a buyer
In February 2014, after mPower posted an $87 million operating loss in 2013, the company slashed investment to just $15 million annually
The official reason: “inability to secure significant additional investors or customer engineering, procurement and construction contracts”
By March 2017, Bechtel withdrew from the joint venture entirely, citing the failure to find a utility willing to site the first reactor or an investor willing to put up capital
Babcock & Wilcox paid Bechtel a $30 million settlement to walk away. The DOE had disbursed $111 million of its planned $226 million; that money is gone. The Wikipedia article on B&W mPower captures the end with merciless brevity: “The development project was terminated.”
What killed mPower? Not the engineering. The design was technically sound, NRC-approvable, and conventional enough to be licensable. What killed it was a customer problem. B&W built a product that utilities could not yet convince themselves to buy, in a market that had no experience pricing long-term commitments to first-of-a-kind nuclear technology. The lesson it left behind is one the industry should have absorbed completely by 2023. It didn’t. 💡
The NuScale CFPP: a masterclass in cost creep
If mPower is the forgotten cautionary tale, the NuScale Carbon Free Power Project is the one that made international headlines and caused NuScale’s stock to lose a third of its value in a single day. 📉
The basic facts are by now well documented. Utah Associated Municipal Power Systems (UAMPS) teamed with NuScale to build six 77-MWe power modules at the Department of Energy’s Idaho National Laboratory site, targeting operation by 2029. The DOE backed it with a $1.4 billion cost-share award. NuScale’s design became the first SMR ever certified by the U.S. Nuclear Regulatory Commission. As recently as 2022, this was widely described as the project that would prove commercial SMRs were real.
Then the cost estimates arrived. The trajectory, per NuScale’s Wikipedia entry and documented by multiple analysts, looked like this:
2015: $3.1 billion estimated cost for twelve modules generating 720 MW
2018: $4.2 billion, design resized to 60 MW per module
2020: $6.1 billion, timeline pushed to 2030
2023: $9.3 billion for just 462 MW — construction cost up 75% in 18 months
By January 2023, the target power price had jumped from $58/MWh to $89/MWh — a 53% increase — making NuScale’s electricity more expensive per megawatt than the notoriously costly Vogtle reactors in Georgia. Utilities did the math and flinched. Subscription levels — UAMPS needed 80% commitment from its municipal member utilities to proceed — stayed well short of that threshold. On November 8, 2023, UAMPS and NuScale mutually terminated the project. Utility Dive reported NuScale CEO John Hopkins said: “Once you’re on a dead horse, you dismount quickly. That’s where we are here.”
The clean-energy think tank Clean Air Task Force wrote a useful post-mortem noting that while the cancellation was painful, the project had still yielded something: NuScale’s regulatory journey produced the first-ever reduction in Emergency Planning Zone requirements for an SMR and demonstrated reduced control room staffing requirements — achievements that benefit every subsequent SMR developer, not just NuScale. The costs of the CFPP were, in part, the industry’s tuition. 🔬
That’s a generous reading. The blunter one is that the project underestimated first-of-a-kind risk at almost every turn. Construction costs for novel nuclear projects have a documented tendency to balloon; a 2014 academic study analyzed 180 nuclear projects worldwide and found 175 exceeded their initial budget by an average of 117%. CFPP behaved exactly as the historical data predicted.
X-energy and the SPAC misfire
The NuScale debacle sent ripples through the entire SMR sector, and the X-energy/Ares Acquisition Corporation SPAC merger is the clearest example. 💊
X-energy, a Maryland-based developer of the Xe-100 high-temperature gas-cooled reactor using TRISO fuel, had announced in December 2022 a deal to go public by merging with Ares Acquisition Corporation (NYSE: AAC), a special-purpose acquisition company. The deal initially valued X-energy at roughly $2 billion. By June 2023, under amended terms, that valuation had been revised down to $1.8 billion. By October 31, 2023, the deal was dead.
The stated reasons were:
“Persistently volatile public market conditions”
“Peer-company trading performance” — meaning NuScale’s collapse scared investors away from the whole sector
A mutual judgment that the risks of becoming a public company outweighed the benefits given the climate at the time
Neither party paid a termination fee. X-energy went back to raising private capital, and Wikipedia’s X-energy entry notes directly that the SPAC collapse was triggered partly by NuScale’s CFPP cancellation and its “effect on the market.” One domino hit another.
The postscript is instructive. X-energy survived, raised $235 million from existing investors in December 2023, secured a massive Amazon commitment for up to 5 gigawatts of power by 2039, and went public in May 2026 at $23 per share — 15 times oversubscribed — in what was described as the largest nuclear IPO on record. The lesson from the X-energy SPAC failure is, I think, simpler than the lesson from NuScale: market timing matters, and 2023 was a terrible year to go public in nuclear. The technology was not the problem. The macroeconomic moment was. 🚀
The BWRX-300: alive but expensive — and the warning it carries
The GE Vernova Hitachi BWRX-300 is not a graveyard story. Construction began at Ontario Power Generation’s Darlington site in May 2025; the first unit is now on track for late 2029. But the BWRX-300’s cost trajectory is close enough to past failures that it deserves to be in any honest discussion of what can go wrong.
When GE Hitachi first pitched the BWRX-300 around 2018, the promise was $700 million per 300-MW unit — roughly $2,250/kW — low enough to compete with natural gas. That estimate is now, generously speaking, a historical curiosity.
Ontario’s final investment decision in May 2025, approved by the provincial government, set the budget for the first unit at $7.7 billion Canadian (about $5.6 billion USD), per World Nuclear News’s detailed cost breakdown. The total projected cost for all four Darlington units is $20.9 billion Canadian in 2024 dollars. The first unit alone costs roughly US$15,000/kW — a number that puts it in the same range as Vogtle, widely cited as the most expensive power plant ever constructed. 🏗️
The IEEFA, in its May 2024 analysis titled SMRs: Still Too Expensive, Too Slow and Too Risky, tracked the cost-per-kilowatt for the BWRX-300 design from $2,883/kW in 2020 to a range of $7,408 to $12,347/kW by 2023 — before a single unit had been poured.
What distinguishes the BWRX-300 from the mPower and NuScale stories is that Ontario Power Generation went ahead anyway. OPG is publicly owned and takes a long view. It expects the second, third, and fourth units to cost dramatically less than the first as construction processes normalize. That logic is plausible — every mature reactor fleet started with an expensive first-of-a-kind unit — but it requires faith in future cost reductions that the industry has promised, and failed to deliver, before.
Do you think the “fleet economics” argument — that FOAK costs will fall sharply by the third or fourth unit — is credible given the history? The answer probably tells you more about your priors on nuclear than any individual data point.
What the graveyard actually teaches us
Look across mPower, the NuScale CFPP, the X-energy SPAC, and the BWRX-300 cost blowout, and a few patterns emerge clearly. ⚡
First: the customer problem comes before the engineering problem. Every failed or struggling project has had adequate technology. What it has lacked is committed buyers at the price the technology actually costs. The subscription model UAMPS used for CFPP — where municipal utilities had to commit before the first concrete was poured — is probably not viable for first-of-a-kind nuclear. You’re asking risk-averse public utilities to underwrite an experiment. The commercial structure has to evolve before the economics can follow.
Second: first-of-a-kind costs are not first-estimates costs. The 2014 study finding that 175 of 180 nuclear projects worldwide exceeded budget by an average of 117% isn’t ancient history. It’s the base rate. Every estimate for a novel nuclear design should be read as a lower bound, not a target. Projects that treat early cost numbers as gospel — and stake their subscription models on them — will be surprised.
Third: negative contagion is real. The X-energy SPAC didn’t fail because of X-energy’s technology. It failed because NuScale’s collapse scared the public markets. The industry’s projects are more interconnected than developers usually admit. One high-profile failure poisons the financing environment for everyone.
Fourth: what looks like failure sometimes isn’t. NuScale’s CFPP produced NRC design certification — still the only one any SMR has — and regulatory precedents that reduce Emergency Planning Zone requirements for the entire next generation of designs. The mPower project trained a workforce and produced engineering data that informed other designs. The industry paid a steep tuition, but it did learn.
The projects now under construction — Darlington, TerraPower’s Natrium in Wyoming, the military microreactor programs — carry the marks of these lessons. They have better-defined customer commitments. They have more government backing to cushion first-of-a-kind risk. Whether that’s enough to avoid the next tombstone in the graveyard is the question the 2030s will answer.
What failure do you think the SMR industry is still repeating today — and which previous mistake do you most clearly see in the projects now moving forward?