A former SpaceX engineer is resurrecting the mPower reactor design. The narrative is clean: AI data centers demand endless power, and nuclear offers a zero-carbon baseload solution. The math is seductive. But the trust? That is the variable.
Context: The Energy Liquidity Crisis
We are watching a classic liquidity squeeze — not in dollars, but in electrons. AI model training consumes 10-20 MW per cluster today, and scaling laws push that toward 100 MW by 2028. The grid is not ready. Natural gas is cheap but carbon-heavy. Solar is intermittent. So the market reaches for nuclear. The mPower design, originally shelved by Babcock & Wilcox in 2017, is being pulled from the shelf. The implicit argument: a 180 MW modular reactor can be co-located with a data center, bypassing grid constraints.
But here is the core problem: the article that broke this story offers zero technical data. No reactor type, power class, licensing status, construction timeline, cost per MWh, or customer commitment. It is a narrative without a ledger. As I wrote after the Terra collapse, "The math was sound; the trust was the variable." The same applies here. The math of nuclear power depends on regulatory approval, fuel supply chain, and long-term power purchase agreements — none of which are disclosed.
Core: The Fragility of the Nuclear-Backed Energy Asset
From a macro liquidity perspective, nuclear projects are capital-intensive, long-cycle, and highly sensitive to regulatory certainty. The mPower design was abandoned once because it could not clear the economic and regulatory hurdles. What changed? AI demand. But demand alone does not unlock capital. The real question is: can a startup with a resurrected design, even with SpaceX pedigree, secure the $1-2 billion needed for a first-of-a-kind reactor, obtain NRC certification, and build within 5 years while AI clusters roll out in 18 months?
My experience auditing smart contracts in 2017 taught me that code is easy, but trust is hard. The same applies to nuclear engineering. The article frames the design as a solution, but the real bottleneck is the gap between narrative and execution. "Liquidity is not a floor; it is a horizon." The horizon of nuclear commercialization is 2030+. The horizon of AI energy demand is 2025. That mismatch is where the risk lives.
Contrarian: The Decoupling Thesis
The conventional wisdom is that nuclear will power the AI boom. I see the opposite: nuclear will be too slow, too expensive, and too regulated to serve the immediate needs of data centers. The real decoupling is between the narrative of "nuclear revival" and the physical reality of permitting, construction, and fuel logistics. The market is pricing the story, not the supply chain.
Consider the alternatives: natural gas turbines with carbon offsets can be deployed in 2 years. Long-duration energy storage (iron-air, flow batteries) is scaling. Even grid upgrades, while bureaucratic, are faster than building a reactor. The article ignores these substitutes. That is not an oversight; it is a signal. The author is selling a nuclear narrative, not an energy portfolio analysis.
"Correlation is the smoke; divergence is the fire." The correlation between AI demand and nuclear interest is real, but the divergence will come when data centers sign gas PPAs while nuclear projects still wait for NRC hearings. The fire is the capital allocation risk — investors chasing the nuclear story may miss the faster, cheaper solutions.
Takeaway: Positioning for the Cycle
This is not a call to dismiss nuclear. It is a call to calibrate. The mPower resurrection is a signal worth tracking, but not a trade. The signals to watch are: (1) NRC docket number, (2) a binding PPA with a hyperscaler, (3) EPC contract award. Until then, treat the story as a concept — a concept that may reshape energy markets in a decade, but will not solve the 2025-2027 power crunch.
"History does not repeat; it rhymes in code." The code of energy infrastructure is long, slow, and unforgiving. The AI industry moves in nanoseconds. The two will meet, but not on this timeline.
In the meantime, the best hedge is to monitor the liquidity of energy assets — not the narrative. The math of nuclear is sound, but the trust variable remains uncalibrated.
