The loud debate over reviving U.S. nuclear plants misses the quiet part: restarts shift risk from the visible to the systemic. The Nuclear Regulatory Commission has cleared a key environmental review for Michigan’s Palisades plant, and the operator targets late 2025 for power production. That headline reads like stability. But reliability is not a press release. It is a function of time, incentives, and entropy. You can move heat away from a reactor core. You cannot repeal the bathtub curve of failure, the long tail of human error, or the compounding effects of aging hardware and aging institutions.
Regulatory green lights are not guarantees. They are threshold tests, not stress tests. An environmental review can say a restart poses no significant impact, while the real dangers live in the operational tail: how equipment behaves after years of thermal cycling, how teams respond to low probability events, how supply chains deliver when a key component fails. Palisades shut in 2022 and changed hands in 2022. The operator intends to pass a gauntlet of inspections and deliver electrons next year. Critics, including former plant engineers, point to degraded steam generator tubes and deferred work that does not care about political timelines. The core claim is simple: dormancy does not preserve reliability; it resets it. In reliability engineering, many systems show a burst of early-life failures on restart, then a stable middle period, then end-of-life fatigue. Nuclear boasts high capacity factors once online. The restart phase is not once online.
Nuclear safety is not only metallurgy and welds. It is institutional memory. It is the tacit knowledge of crews who have seen edge cases and caught weak signals in the noise. The regulator itself has seen turnover, with a notable wave of departures in recent years. That is not an accusation, but it is a variable. Talent attrition creates information asymmetry between regulator and operator at exactly the time oversight should be most granular. The plant workforce faces similar issues. Specialized skills atrophied during the pause, vendors exited, inventories thinned. Obsolescence creeps in. Replacing a pump is easy. Replacing a vendor ecosystem and a culture of skepticism is not. The lesson from aviation and chemical plants is consistent: life extension projects work when you overinvest in inspection and training and assume your first estimates are wrong. Underinvest, and the system becomes brittle. On paper you have a capacity asset. In practice, you own a reliability option with unknown strike price.
The federal government has put up a large loan to bring Palisades back. The climate case for firm, carbon free power is strong, and the political will is clear. Here is the problem. When Washington becomes financier and cheerleader, the regulator faces a classic game theory trap. The arbiter in a repeated game with high public stakes begins to internalize the sponsor’s goal. Strip away the rhetoric and you have a principal agent problem: taxpayers bear a slice of tail risk, operators capture upside, and the referee is under pressure to keep the match going. The Price Anderson framework caps private liability for catastrophic nuclear accidents. That is not scandalous; it existed for decades. It does, however, change behavior at the margin. Private actors optimize to the rules they face. Subsidized debt and limited liability nudge the system toward speed and scale. In the short run, that looks like momentum. In the long run, incentives compounded by time become culture. Culture determines whether weak signals are escalated or rationalized.
Pro nuclear arguments often cite high capacity factors and long lifespans. True. System planners, however, must price reliability at the portfolio level. An 800 megawatt unit that trips offline in a stressed grid can be more destabilizing than 800 megawatts of distributed assets that fail independently. One is a single point failure with long repair tails; the other is a set of small, mostly uncorrelated failures. Factories know this as common mode risk. Power markets dress it up as contingency reserves. The question is not nuclear versus wind and solar in a vacuum. It is system design under uncertainty. A grid with firm, dispatchable power is valuable. But fragility hides in concentration and in the operational realities of old machines. The correct comparison is not nameplate megawatts. It is expected unserved energy and the cost of contingencies. Can the regional market carry a sudden 800 megawatt hole in February and in August, both under transmission constraints, without cascading effects and price spikes that bleed into the broader economy
Financial markets like clean narratives. Nuclear restarts provide one: revive a carbon free workhorse with a big government loan, collect steady cash flows, and claim a climate win. The headline variable is capacity factor. The tail variables are restart risk, supply chain brittleness, and governance quality. Those tails do not fit neat spreadsheets. They require fat tail thinking, not Gaussian comfort. If you discount a 1 in 10,000 event as negligible, you have not done the expected loss math honestly because the loss distribution is not bounded in a way your model assumes. Insurance regimes that cap worst case liabilities make sense for social reasons. For investors, they can blur true cost of risk. When the downside is socialized and the upside is private, project finance looks good even if the social expected value is ambiguous. That is the kind of mispricing that does not show up until correlated stress hits the system. We saw versions of this in mortgage finance. Tail risk lived in the tranches no one examined with skepticism.
The argument that the loan should have gone to wind or solar misses and hits the point. It misses because firm power is scarce and useful. It hits because every public dollar is a choice under uncertainty. What is the option value of building more flexible capacity, storage, and demand response versus placing a large bet on a single plant with restart and life extension risk What is the system value of upgrading transmission that reduces congestion and spreads weather risk How many dollars should go to measures that degrade gracefully under stress rather than those that fail in big chunks This is not an anti nuclear point. New nuclear designs with modularity could reduce single unit risk. Restarts are a different asset class. They require a margin of safety larger than the model suggests. The case for nuclear is strongest when it explicitly pays for redundancy. That means either higher private capital at risk or contractual clawbacks linked to uptime and safety performance.
Inversion helps. Instead of asking how fast a plant can be back on the grid, ask what would make it better after shocks. That list is not a press release. It is over provisioned inspection budgets, third party audits with real veto power, staged power ramps with long dwell times, transparent probabilistic risk assessments that expose uncomfortable tails, and operator compensation tied to long term safety metrics, not first power dates. It is deep bench staffing, apprenticeships to rebuild tacit knowledge, and a parts inventory sized for low probability failures. It is digital upgrades hardened against cybersecurity threats, which are not a footnote for legacy control systems. It is grid level hedges too: more flexible reserves, demand response that can shed load without economic panic, and contracts that price outages realistically. Antifragility means the system absorbs shocks and learns faster than entropy degrades it. That is expensive. It is cheaper than pretending.
If Palisades restarts smoothly, success will breed copycats. If it stumbles, the backlash will be blunt and political, and it will set back firm, carbon free power for a decade. Either way, the decision writes a playbook. The only durable path is to surface the hidden fragilities now and pay to remove them. Markets will not do this on their own when the downside is diffuse and the upside is narrow. Regulators are not built to optimize across decades when turnover, budgets, and politics change every two years. That is the paradox of nuclear restarts in 2025. They promise stability while introducing new layers of complexity. The solution is not to walk away; it is to design like an engineer and invest like a probabilist. Build systems that fail small, learn fast, and never bet the grid on an assumption that the hard part is behind you just because a permit cleared.
