Demand for generative artificial intelligence (Gen AI) has surged recently. For example, ChatGPT’s weekly active user count increased from 100 million to 200 million between November 2023 and August 2024. This in turn has stoked demand for electricity. A standard ChatGPT query consumes ten times more electricity than a Google search and “drinks” one-third to one and a half ounces of water. Data centers already use 1-2% of global electricity and Goldman Sachs estimates this will increase to 3-4% by 2030. To find enough electricity to power these pursuits, tech giants are going nuclear. Recently, Google purchased power from Kairos Powers’ fleet of small modular reactors (SMRs), Amazon invested in Dominion Energy to develop SMRs and Microsoft is partnering with Constellation to reactivate the reactor at Three Mile Island. Chinese startup Deepseek shocked U.S. players (and capital markets) in early 2025 with a more energy efficient large language model, but its energy impacts are debatable in that this may just stimulate even more corporate AI use.
Most P&C insurance policies exclude damages resulting from nuclear accidents (although Swiss Re questioned the robustness of these exclusions), with private risks socially insured by the Price-Anderson Act. Energy companies also purchase liability coverage from a pool of mostly brand-name insurers called American Nuclear Insurers, which paid hundreds of millions after Three Mile Island melted down in 1980 but otherwise has not experienced any significant triggering events.
What this means for actuaries:
A would-be nuclear renaissance may feel similar to the unexpected 32% compound annual growth of global cyber insurance between 2017 and 2022, which illustrated the industry’s (and actuaries’) facility to rise to the challenge of rapidly emerging perils. However, not unlike the fragile nuclear exclusion, academics caution that the industry is ill-equipped to handle “silent cyber” exposure in lines such as property. A non-cyber example of “silent” property exposure is how the industry dodged a bullet during COVID-19 due to a communicable disease exclusion ISO developed in 2009. Informed by history, actuaries should work with experts in their organizations to firm up existing policy language and ensure fairly-priced coverage is available to service a burgeoning market of AI-driven risk (including but not limited to nuclear liability).
Sources:
- https://news.ambest.com/newscontent.aspx?refnum=62140.
- https://portal.ct.gov/-/media/cid/1_stipulation/fin-exam-american-nuclear-insurers-2020-11-30.pdf.
- https://www.casact.org/sites/default/files/database/proceed_proceed68_68255.pdf.
- https://www.swissre.com/risk-knowledge/advancing-societal-benefits-digitalisation/about-cyber-insurance-market.html.
- https://www.goldmansachs.com/insights/articles/AI-poised-to-drive-160-increase-in-power-demand.
- https://www.insurancejournal.com/news/national/2020/11/30/592047.htm.
- https://www.cnbc.com/2024/10/15/big-tech-turns-to-nuclear-energy-to-fuel-power-intensive-ai-ambitions.html.
- https://www.cnbc.com/2024/09/07/how-small-modular-reactors-could-expand-nuclear-power-in-the-us.html.
- https://www.law.virginia.edu/scholarship/publication/kenneth-s-abraham/1195966.
- https://www.washingtonpost.com/archive/business/1980/03/19/reactor-insurance-cost-laid-to-three-mile-island/15de4fc5-bba9-4920-9106-720d28b64e39/.
- https://www.linkedin.com/pulse/ai-vs-laws-electricity-jim-weiss-fcas-xsb7e.
- https://www.technologyreview.com/2025/01/31/1110776/deepseek-might-not-be-such-good-news-for-energy-after-all/.
