Over the past month, a single narrative has circulated through tech and energy circles: the mPower nuclear reactor design has been revived by a former SpaceX engineer to power AI data centers. The claim is clean. The story is seductive. The data is absent.
Let me be precise about what the source material actually contains. Not one verifiable figure. No reactor type. No power rating. No licensing status. No construction timeline. No cost data. No customer agreement. No regulatory pathway. The entire report is a narrative bridge between two known trends: AI energy demand is rising, and nuclear designs have historically struggled to reach commercial deployment. That bridge is built on assumption, not evidence.
In my line of work, I audit smart contracts. When a protocol lands on my desk with a whitepaper full of promises and zero code, I close the ticket. The math does not get reviewed if the inputs do not exist.
Context: The Revival That Was Never a Startup
The mPower reactor was not a new invention when it was shelved. It was a small modular reactor design originally developed by Babcock & Wilcox, intended for the kind of steady-state baseload power that industrial customers need. It was abandoned. Not because the design was technically impossible. It was abandoned because the regulatory, commercial, and construction path did not close.
Now it is being resurrected by a team led by a former SpaceX engineer, repositioned as the answer to AI data center electricity demand.
This is where the narrative takes a familiar shape. Replace "revolutionary L2 scaling solution" with "advanced nuclear reactor." Replace "TPS" with "megawatts." The pattern is identical. A compelling story, a pressing need, and a complete absence of engineering verification.
Here is the fundamental question the article does not answer: why did this design get shelved in the first place? The original mPower program was terminated after years of effort and significant capital expenditure. The design was not revived because of a new technical breakthrough. It was revived because AI data centers need power and the market is looking for new sources.
The math does not care about the narrative.
The Problem is Not Supply. It is Verification.
Let me break down what actually matters for a nuclear project serving a data center. This is not a claim about the reactor's thermal efficiency. It is about the full path from design to delivery.
First, regulatory licensing. The Nuclear Regulatory Commission does not approve reactor designs based on pitch decks. The review process takes years. And this is after the design has undergone testing, safety analysis, and a licensing path. The article does not mention whether the mPower design is even in the NRC review queue. It does not mention a single filing, a single review milestone, or a single agency engagement.
The absence of this information is not a minor omission. It is the entire story. Regulatory approval is not a step in the process. It is the process.
Second, engineering verification. The article does not mention whether this team has built a prototype, conducted a demonstration run, or secured an engineering, procurement, and construction (EPC) partner. In the nuclear world, the design is a fraction of the effort. The challenge is in materials, quality control, construction, and long-term operation. A former SpaceX engineer brings expertise in fast-moving, iterative hardware. The nuclear industry is the opposite of that.
Third, economic reality. The article provides no levelized cost of energy. No construction cost estimate. No operating cost. No fuel cost. No capital structure. Nothing. This is the one data point that determines whether the whole project is viable. The article cannot mention it because there is no data to mention.
Fourth, the commercial close. Who is buying the power? Is there a signed power purchase agreement? A memorandum of understanding? A letter of intent? The article mentions none of these. And this is the critical part: AI data centers have power needs, but the centers do not automatically agree to pay a premium for nuclear power. They have existing options. Grid power, natural gas, and utility-scale batteries are all on the table. Nuclear must compete on price, reliability, and timeline.
The Contrarian View: The Timeline Mismatch Is the Real Vulnerability
The data center demand is real. The nuclear supply is not. That is the core structural tension that the article completely ignores.
Here is the timeline problem. Nuclear projects from design to operation take a decade or more. That is the realistic experience. AI data centers are being built on a scale of months. The demand exists today. The supply will exist, at best, in the 2030s.
The article treats "AI data center demand" and "nuclear reactor supply" as if they exist in the same time frame. They do not. The demand curve is steep and immediate. The supply curve is shallow and slow. This mismatch is the defining feature of the entire narrative.
The second blind spot is the "revived design" framing. In my line of work, when I see a design that was abandoned and then resurrected, I ask a question: what changed? The answer is usually not the technical design. It is the market narrative. The design did not become safer or more efficient. The market simply became more desperate.
The third blind spot is the physical infrastructure that sits between the reactor and the data center. The article treats "power supply" as a single point. It is not. There is the grid connection, the transformation, the transmission, the redundant feed, the backup systems, and the dispatch rights. Who is responsible for each? The article does not address any of it.
An Unverified Narrative Is a Risk, Not an Opportunity
Let me be direct about what this article is. It is a narrative. It is not an analysis. It does not provide a single piece of verifiable data that would allow an investor, a data center operator, or an energy planner to make a decision.
The information density is so low that the entire piece could be condensed into one sentence: "A reactor design was shelved, and some engineers are trying to bring it back to power AI data centers." That is not a thesis. It is a rumor with a timestamp.
What would change my mind? Specific, verifiable milestones.
First, if the mPower team publishes a regulatory submission. A filing with the NRC, a design certification application, or even a pre-application meeting notice. That would be proof that the project is moving toward real world deployment.
Second, if they announce a customer commitment. A power purchase agreement, a memorandum of understanding, or a public statement from a data center operator that they intend to buy nuclear power. That would be proof that the commercial model works.
Third, if they release engineering data. A construction schedule, a cost estimate, a project timeline. That would be proof that the project has moved beyond the pitch deck.
None of these things exist in the article. And the absence of them is not a detail. It is the definition of the project's stage.
Security is not a feature; it is the foundation.
This is the core principle of my work, and it applies here as well. The security of an energy system is not an add-on to the project. It is the project. The reactor must be safe, the grid must be stable, the regulatory path must be clear, and the economics must work. Every one of those is a precondition for the project to exist at all.
The mPower story has none of those preconditions verified. That does not mean the project will fail. It means the project is not yet a real thing.
The lesson is not that nuclear power is impossible. The lesson is that a demand narrative does not replace an engineering reality. The AI industry needs power. That is a fact. Whether nuclear power will be the source of that power is a separate question that cannot be answered by a story about a resurrected design.
The Takeaway: Watch the Milestones, Not the Headlines
Here is my forward-looking judgment. This is a story to watch, not a story to act on. The signal is the narrative, not the outcome. The project will become meaningful only when it crosses the four gates: regulatory approval, engineering contract, customer commitment, and cost data. Until then, the "nuclear power for AI" is an idea, not a system.
The math does not check out because the math has not been provided. That is not skepticism. That is process.
If the mPower team publishes regulatory progress, signs a customer agreement, and provides cost data, the equation changes. The narrative becomes a project. The project becomes a potential opportunity. But until that point, the information is a story. It is not a fact.
Trust the code, verify the trust. There is no code here. There is only a narrative. And the narrative is not a substitute for verification.