A design once shelved by Babcock & Wilcox in 2017 has resurfaced, resurrected by a former SpaceX engineer with a new thesis: power the AI datacenter boom. The market narrative is seductive — infinite compute needs infinite electrons, and nuclear is the only zero-carbon source dense enough to saturate the load.
The consensus is wrong because it ignores the cost of attention. The market is treating a design resurrection as if it were a commercial deployment. A thumbnail sketch, a press release, and the narrative machinery runs. History doesn't repeat, but it rhymes. The 2017 ICO era taught me that a whitepaper is not a product, and a reactor diagram is not a power plant.
This is not a commentary on the technology. It is an audit of the narrative. And the audit reveals a gaping hole between the demand-side story and the supply-side reality.
The Context: A Market Starving for Baselines
AI datacenters are not simply energy consumers; they are energy monopolists. Their load profile is not a sine wave but a horizontal line — high, continuous, and unforgiving. This is not the domain of solar-plus-storage or wind with a battery. Intermittency is a structural liability. The market is desperately seeking baseload, and gas-fired generation, while pragmatic, carries a carbon liability that the world's most scrutinized companies cannot tolerate.
This is the opening that advanced nuclear claims to fill. A small modular reactor, sited on a dedicated campus, delivering 24/7 zero-carbon power via a direct wire. It's a compelling pitch. The problem is that the pitch is not the product. The road to commercial reality is paved with NRC dockets, not press releases.
The Core: The Four-Gate Audit of Nuclear Revival
From my experience auditing 200+ ICO whitepapers, I learned to separate the narrative from the architecture. The same filter applies to physical infrastructure projects. To assess the validity of this nuclear resurrection, one must apply a four-gate test. The first gate is Regulatory: What is the NRC status? Has the design been submitted for pre-application review? Has a Design Certification Application been docketed? The resurrection of a design is not the same as the resurrection of its license. The original mPower design was shelved for business reasons, but the regulatory path is not a lock-step. An NRC review is a multi-year, multi-million-dollar exercise. Without a clear regulatory path, the project is a concept, not a project.
The second gate is Engineering: Is there a demonstration plan? A single-unit prototype? A fuel qualification test? The market is fatigued with novel designs that fail at the critical stage of 'demonstration'. The hardest part of the nuclear industry is not the physics but the materials. It's the welding. It's the procurement. It's the ability to build a concrete silo to spec and schedule. The term 'former SpaceX engineer' is a brand; it is not a nuclear supply chain. It does not confer the ability to source, fabricate, and test a reactor pressure vessel.
The third gate is Economic: What is the Levelized Cost of Electricity (LCOE) and the construction cost? The narrative of the AI datacenter is one of price-insensitivity. It is a story about urgency. But that urgency translates into a willingness to pay an arbitrage. If nuclear costs $150/MWh and gas costs $70/MWh, the only reason for a cloud provider to sign the PPA is for carbon attributes. And even then, the process for verifying a "zero-carbon" source is not trivial. The carbon accounting rules for nuclear are complex. The fuel cycle is not emission-free.
The fourth gate is Time: What is the schedule? A datacenter is built in 18 months. A nuclear plant is built in 7-10 years. The mismatch is not a gap; it is a chasm. The AI infrastructure cycle is in hyperdrive. This year's needs are urgent. Next year's are obsolete. If this project's earliest power date is 2032, it is not serving the AI boom; it is serving the AI aftermath.
The consensus is that AI needs power. It does. But this doesn't mean it needs nuclear. It needs power, now. It needs the power grid to be more flexible. It needs gas turbines to be more efficient. It needs long-duration storage. The nuclear pitch is a long-duration play in a market that demands the short-term. Volatility is the fee for admission to the future; but so is time.
The Contrarian: The Missing Bridge of Power
Here is the blind spot: the industry assumes that a reactor can be connected directly to a datacenter. But a reactor does not come with a power cord. The bridge between the two is the grid or a dedicated substation. The article omits the most critical piece of infrastructure: the interconnection. Who builds the high-voltage line? Who bears the cost of the transmission? Who provides the reactive power for the voltage regulation?
This is not a trivial question. A large datacenter is a massive load that introduces harmonics and imbalances. The utility must approve the interconnection. The utility must also guarantee a certain power quality. This process is often more lengthy than the reactor licensing itself. The reality is that a dedicated unit is an island. It needs a system. And the system is the grid. The grid is the sovereign. Code is law, but capital decides who writes it. In this case, the capital is the transmission tariff.
And then there is the fuel cycle. The narrative stops at the reactor vessel. It does not address the fuel fabrication, the enrichment, the spent fuel storage, or the decommissioning. The project is a liability chain. The operator is not just a power producer; it is a custodian of radioactive material for decades. This is a responsibility that does not fit the "fast-moving" culture of the AI industry. The culture of AI is one of fast iteration. The culture of nuclear is one of slow, deliberate, safety-first protocol. These two cultures are not naturally aligned.
The narrative is a single story: 'AI needs power; nuclear has it.' But the implementation is a cross-disciplinary problem: 'AI needs power, now, at a specific location, with a specific rate, with a specific price, and with a specific interconnection path.' The first narrative is a rallying cry. The second is the reality of project finance.
The Takeaway: The Signal vs. The Noise
This is a signal to watch, not a signal to act on. The announcement is a protocol-level event, not a market event. The real signals are: 1) The announcement of a design certification application with the NRC; 2) The signing of a binding PPA with a hyperscaler; 3) The announcement of a site selection and a construction schedule. Until one of those occurs, the project is a concept.
The market will continue to trade on the demand side. The datacenter power is a real and growing secular trend. But the supply side, the nuclear supply, is a function of regulatory, engineering, and financial reality. The gap between the two is the opportunity. But it is also the risk. I will continue to watch the grid interconnection dockets. I will watch the PPA price. The code is not yet written. The capital has not yet been deployed. I will wait for the first gas turbine to be cancelled. I will wait for the first reactor to be poured. The first contract is the first proof. Everything else is a narrative.
The demand is real. The supply is not. The next few years will reveal whether this is a narrative or a trajectory. The next decade will reveal if the physics is right. The capital will decide who writes the code. The rest is just noise.
