Reviving a Reactor Design Is Not a Power Deal: The mPower/AI Data-Center Thesis Under Due Diligence

A freshly relabeled reactor design is circulating through energy headlines with an AI data-center pitch attached. The claim is simple enough to sell: advanced nuclear, revived by a former SpaceX engineer, may become the missing power source for the next generation of high-density compute clusters. That sounds like a thesis. It is not yet a business case. In my due-diligence work, I have learned to separate narrative surface from contract-grade evidence. The first test is brutal. If the article cannot answer who will buy the power, who will approve the plant, who will build it, and who will carry liability for the waste, then it is not an industrial report. It is a signal that the market is trying on a new story. This matters because the market is already pricing in hype faster than engineers can price in construction risk. In 2020, while auditing the interest mechanics around Compound, I found that markets often treat a clever design as if it were a completed economic machine. They do not wait for the operating constraints to appear. The same reflex is happening now around advanced nuclear and AI infrastructure. The difference is that nuclear failure is not a weekend bug. It is a decade-long cost event. The background is not complicated. AI workloads are pushing hyperscalers and data-center operators toward larger, steadier, lower-carbon electricity contracts. The compute racks are denser, the uptime requirements are harsher, and the public carbon disclosures are harder to fake. That creates a plausible opening for nuclear power, especially if a reactor vendor can prove direct supply to an industrial customer instead of competing only in the slow-moving utility market. The mPower story is being framed as an example of that opening. The underlying logic is that a small or modular reactor design may be revived, repositioned, and sold to customers who cannot wait for the old utility procurement cycle. That is the claim the market wants to believe. But the claim is thin. The parsed source material contains almost no primary data. There is no reactor power rating. There is no license status. There is no construction timeline. There is no EPC partner. There is no fuel supplier. There is no insurance structure. There is no power purchase agreement. There is no grid study. There is no waste plan. There is no cost per megawatt-hour. There is no named customer. That absence is not an oversight. It is the finding. A nuclear story without those fields is not an investment case. It is a plot summary. The first layer of analysis is regulatory. Code is law, but capital is king. In nuclear, regulation is closer to law than in almost any other energy business. A revived design does not become commercial simply because an engineer decides it should be useful again. It must survive the licensing path, and in the United States that path runs through the Nuclear Regulatory Commission. The important question is not whether the physics is interesting. The important question is whether the design has entered a recognized review track, whether it has enough safety documentation to survive scrutiny, and whether the agency is willing to certify it at a pace that matches the data-center buildout. If those answers are unknown, the project should be treated as pre-commercial. The former SpaceX label is not a substitute for a design certification record. SpaceX success does not transfer automatically to nuclear safety culture, construction sequencing, or multi-decade operator liability. The second layer is engineering repeatability. In my audit of the 0x protocol, the issue was not that the team had ambition. The issue was that the protocol had hidden edge cases that only appeared under adversarial conditions. Advanced nuclear has the same property. Conceptual elegance does not prove deployability. A reactor design can look clean on paper and still fail once siting, procurement, concrete, piping, instrumentation, quality assurance, and construction management enter the picture. The relevant evidence is not the biography of the founder. The relevant evidence is the existence of a credible demonstration unit, a qualified supplier base, a licensed construction route, and a team that has already delivered regulated infrastructure. None of those are present in the parsed article. That means the right classification is not near-term supply. It is speculative infrastructure. The third layer is economic. This is where the thesis becomes fragile. AI data centers need large, reliable, low-carbon electricity. That does not mean they automatically need nuclear electricity. They need a contract that clears a business screen: availability, price, interconnection feasibility, delivery date, balance-of-plant compatibility, maintenance responsibility, and downside allocation. If nuclear cannot clear that screen, the demand narrative collapses. Nuclear may be zero-carbon at the point of generation, but the commercial question is whether a data-center operator will pay enough to fund a multi-billion-dollar facility, absorb regulatory delay, and accept a long construction window. The parsed material says nothing about power price, capital cost, financing structure, or customer willingness to pay. Without those numbers, the article is describing a want, not a market. The fourth layer is timing. Nuclear supply and AI demand do not run on the same clock. A data-center developer can move quickly once land, water, cooling, labor, and grid capacity are available. A nuclear project moves through approval, procurement, design finalization, site work, construction, fuel loading, testing, and commissioning. The mismatch is severe. The article effectively connects an immediate load problem with a slow-build solution and then stops. That is dangerous because the market can price the connection before the delivery path exists. The right due-diligence question is not whether AI will need more power. It is whether this specific reactor can arrive with that power before the customer’s site plan changes, the utility expands capacity, or the customer finds a cheaper procurement path. The fifth layer is infrastructure. A reactor does not power a campus by existing. It needs a delivery architecture. That architecture may be grid-connected, direct-fired to a large industrial user, embedded in a private microgrid, or structured as a dedicated generation asset behind a negotiated transmission arrangement. Each model changes the risk profile. Direct supply can reduce some grid uncertainty, but it does not remove the need for capacity planning, contingency supply, maintenance windows, and dispatch compatibility. The parsed article says only that the reactor would power AI data centers. It does not explain the medium between the reactor and the rack. In regulated energy systems, that medium is where projects actually die. The sixth layer is commercial credibility. During the FTX collapse, I spent time tracing commingled assets and balance-sheet failures. The lesson was that labels hide obligations. A company can be described as innovative, backed by credible talent, or aligned with a strong macro trend, and still be underfunded, over-licensed, or unable to deliver. The same discipline applies here. A former aerospace engineer can be exceptional and still lack the nuclear operating record, safety organization, regulator relationships, insurance capacity, and long-term liability framework required for this business. The investor job is to test the entity behind the story, not to accept the story as evidence of the entity. The seventh layer is ESG validity. The zero-carbon argument is real, but incomplete. Nuclear generation is low-carbon in operation, but the full chain still includes uranium supply, fuel fabrication, construction embodied emissions, long-term operations, decommissioning, and spent fuel management. If the goal is to help a hyperscaler improve Scope 2 disclosure, the electricity source must be verifiable, attributable, and accepted under credible accounting frameworks. The article does not address whether the power can be used in an auditable carbon claim, let alone whether the reactor has a defensible lifecycle emissions profile. That leaves the green claim in the same category as many other unproven market narratives: directionally plausible, contractually weak. There is a contrarian angle that should not be ignored. The bulls are not wrong about the structural problem. Data centers do need durable, low-carbon baseload power. Gas is politically exposed. Renewables require storage or firming. Grid queues are slow. Public disclosure is tightening. In that environment, nuclear deserves to be back on the procurement menu. The real issue is not whether nuclear belongs in the conversation. It is whether this specific project is ready to survive the conversation. If the revived mPower design can prove a licensing route, an engineering delivery path, a cost stack that data-center customers can absorb, and a direct supply structure that avoids the worst grid delays, it could become a meaningful exception in the advanced nuclear market. That outcome is possible. It is not the base case based on the available evidence. Hype is leverage in reverse. The more a market overloads a project with macro meaning, the less careful it becomes about the underlying mechanics. AI demand is being used as a universal justification for a reactor design that has not yet shown commercial maturity. The narrative does the work of missing data. That is useful for attention. It is not useful for capital allocation. A mature investor should treat the story as a request for proof, not a signal to assume success. The project should be followed, but it should not be credited as a solved power solution. What should be watched next is narrow and technical. The first real signal is regulatory progress: entry into a formal review track, design certification milestones, or public confirmation that the safety case is being developed for an approved pathway. The second is engineering progress: site selection, EPC partner, construction plan, and a demonstration schedule that is not aspirational. The third is commercial progress: a power purchase agreement, a memorandum that can escalate into a binding contract, or a named data-center customer willing to absorb long-lead uncertainty. The fourth is financial progress: debt, equity, loan guarantees, or insurance capacity that matches the project’s regulated risk. Without those signals, the right posture is observation, not conviction. The takeaway is unsentimental. Reviving an old reactor design is not a strategy. It is a starting point. The only way to know whether the mPower/AI data-center thesis is real is to force the project through the same gates that any critical infrastructure asset must clear: regulatory approval, engineering repeatability, economic viability, customer commitment, and liability closure. If the project can clear those gates, it may become a credible part of the next energy stack. If it cannot, the market will have confused a powerful demand trend with a non-existent supply solution. The next question is not whether AI will keep consuming more power. The next question is whether this project can produce it, sell it, and survive the decades required to deliver it. The final judgment is simple. This story is a signal, not a conclusion. Advanced nuclear may deserve a renewed look because AI infrastructure has changed the shape of electricity demand. But a revived design with no disclosed license path, no construction proof, no financing stack, and no customer contract is not evidence that the market has changed. It is evidence that the market is searching for a new place to put its expectations. The burden is on the project to move from narrative to obligation. Until then, the correct due-diligence verdict is not enthusiasm. It is patience, verification, and pressure on the missing fields.