The Nuclear Option: Why Powering AI Is a Risky Bet

With today's discussions about developing new and advanced nuclear energy facilities to support the extraordinary power demands of artificial intelligence and the military, perhaps we should look back at the adverse loss development that accompanied the nuclear buildout of the late 1970s and 1980s.

Three Mile Island in Pennsylvania in 1979 and Chernobyl in Ukraine in 1986 fundamentally changed the risk environment surrounding nuclear power. The consequences extended well beyond property damage and nuclear liability. They reached securities claims, D&O insurance, regulatory proceedings, ratepayer litigation, project financing, municipal bonds and ultimately the insurance industry.

I was very involved in the Long Island Lighting Company (“LILCO”) decommissioning and closure of the Shoreham Nuclear Power Plant and worked with its CEO, William Catacosinos, after the plant had been built but before it entered commercial operation. What began as a nuclear project to supply electricity to Queens, Nassau and Suffolk Counties ultimately resulted in approximately $6 billion in construction expenditures, compared with an original estimate of less than $100 million.

The plant never became operational because of the hysteria surrounding the Three Mile Island meltdown in early 1979. We were deluged with shareholder suits under § 11 of the Securities Act and § 10(b) of the Securities Exchange Act and SEC Rule 10(b)-5 , governmental disputes with the State of New York, regulatory proceedings with the Nuclear Regulatory Commission (“NRC”), and ratepayer litigation. The NRC and State of New York refused to approve an evacuation plan deemed infeasible in such a heavily populated area. The securities claims alleged failure to disclose cost overruns, mismanagement of those costs, and not disclosing challenges of regulatory hurdle approvals.

CEO William Catacosinos, an NYU-educated businessman with a B.A., M.B.A., and Ph.D., pushed hard for multiple policy-limit recoveries, particularly for costs associated with ratepayer and regulatory claims arising from the decommissioning. Shoreham was ultimately transferred for $1 to the Long Island Power Authority (“LIPA”), a public authority created by the New York Legislature in 1986 that eventually acquired LILCO’s electric system and the 57-acre Shoreham site. Mr. Catacosinos received a reported $45 million golden parachute in connection with the LILCO-LIPA transaction pursuant to a negotiated settlement between LILCO and LIPA.

Then There Was WPPSS

At approximately the same time, the Washington Public Power Supply System ("WPPSS") embarked on an extraordinarily ambitious nuclear construction program in the Pacific Northwest. WPPSS was a joint operating agency on behalf of 88 northwest public utilities that failed after committing to a five plant nuclear construction project, resulting in the biggest municipal bond failure in the United States in 1983.

Approximately $2.25 billion in bonds were sold to investors, the proceeds of which were to be used to subsidize construction of the plants backed up by guarantees from the utilities contractually obligated to buy energy from WPPSS and to repay the bondholder even if the project failed. The plants were never completed, WPPSS defaulted on its bond delegation and the bondholder suits followed, as did $1.5 billion in settlements to bond holders from 200 different defendants.

At the center of WPPSS was a fundamental mismatch between the enormous technical and financial complexity of nuclear development and the experience of many of those responsible for overseeing it. WPPSS was an artificially created joint operating agency comprising 88 publicly and privately owned utilities, many with little or no experience in the design or management of nuclear power projects. Its board included administrators from rural utilities who likewise had little experience with nuclear fission or the design, construction and oversight of projects of this magnitude. Investment banks marketed the bonds based on projected demand for additional electricity and the utilities' contractual payment obligations. When the projects collapsed, those assumptions collapsed with them.

The Supreme Court of Washington ruled that the utilities had no authority from their ratepayer customers to obligate repayment and saddle their customers, some in poor rural areas, with insurmountable debt and also risk their access to power from a bankrupt utility.

Have We Learned the Lesson?

The question is whether we have learned from the lessons of the 1980s. Does our current urgency to win the AI race justify the means of getting there? Is the morality of using oil and gas resource extraction justifiable as a means of controlling global prices in the energy sector? Is this race going to lead to regulatory rollbacks by the U.S. Department of Energy, the Nuclear Regulatory Commission, and state energy and nuclear regulators, potentially jeopardizing the safety of millions of people as occurred with the Chernobyl meltdown? Will investment banking greed, regulatory rollbacks, and fear of external and internal cyber threats to our energy infrastructure and grid influence the morality and means of how we win the AI race? What are the alternatives? Perhaps a more balanced, less egotistical approach with ongoing regulatory safeguards is sensible.

The projected electricity requirements of AI and hyperscale data centers are again creating enormous pressure to bring generating capacity online quickly. Natural gas will likely satisfy a substantial portion of the immediate need, while existing nuclear plants are increasingly attractive for their ability to provide reliable, around-the-clock generation.

As one example, New Jersey is confronting these issues today. PSE&G is pursuing uprates at its Salem nuclear units (i.e., modifications that increase the amount of electricity an existing reactor can generate), while the Power NJ Act seeks to encourage new nuclear generation with financial safeguards drawn from past experience, including protections for ratepayers from construction costs before a new plant begins operating and from construction cost overruns.

The lesson of Shoreham and WPPSS is not that nuclear power should not be built. It is that enormous projected demand, abundant capital and technological urgency can create a dangerous confidence that the economics, regulation and execution will eventually work themselves out. And, the reality is that sometimes they don't.

As the insurance industry facilitates the infrastructure needed to power the AI race, it will also be one of the principal risk takers if those assumptions prove wrong. The opportunity is enormous. So is the aggregation of risk.