Will nuclear power AI data centers?

Data centers throughout the U.S. consumed 183 terawatt-hours of energy last year, or more than the state of Ohio. As AI adoption accelerates, by 2030 they could draw 12% of total U.S. grid power¹. The continued growth of AI is reliant on access to even more electricity.

Natural gas supplies 40% of electricity for data centers and it is expected to be the largest source of additional supply in the near term. But building new power plants takes time, gas turbines are in short supply, and impatient operators have been searching for alternatives. 

Against this backdrop, plans to restart mothballed nuclear plants and build new ones have captured the public imagination.

This enthusiasm has translated into market performance. Uranium prices have doubled since 2021, and an ETF of nuclear-linked equities has posted gains of 68% year-to-date.

Amid this exuberance, investors should consider why nuclear has fallen out of favor, with generation declining since the 1990s² compared to gas and renewables.

1. Long timelines: Nuclear projects average 9 years from groundbreaking to commercial operation, often with massive cost overruns. The last nuclear plant built in the U.S. was completed in 2024 for $30bn, double its budget of $14bn, and 7 years later than planned³. For data centers needing immediate power, this timeline mismatch is a challenge. 
2. High costs: While modern nuclear plants are relatively cheap to operate once built, construction represents 80% of total costs. Improved technologies for extraction of fossil fuels have made gas cheaper per unit than nuclear, and solar costs are declining rapidly. 
3. Import reliance: The uranium used by American reactors is mostly imported⁴, making it vulnerable to tariffs and geopolitical volatility. By contrast, there are plentiful domestic gas supplies and renewable resources.

Longer-term bets

Small Modular Reactors (SMRs) promise faster construction times, thanks to simpler designs. There are currently only two SMRs operational today, in remote parts of China and Russia, and construction is underway on a unit in Canada. It remains to be seen whether the unit economics at scale will be competitive against conventional nuclear or other alternatives.

Nuclear fusion has been hailed as the ultimate long-term energy breakthrough. Theoretically, it could use cheap natural fuels and emit less radioactive waste than conventional fission. However, fusion remains unproven outside of laboratory settings.

Grid-scale deployment of these technologies could still be decades away. Fusion experiments still consume more total energy than they generate, and the small-cap companies working on nuclear development typically spend more cash than they earn.

No uranium bullet

Nuclear power will likely be one component of a diversified energy solution for AI. 

Newer technologies may form part of a diversified venture portfolio. Long-term private market investments can be an appropriate way to match investment horizons with the timelines for data center power.

To gain exposure to energy providers offering solutions today rather than decades into the future, investors might consider private infrastructure funds, which can directly invest in energy generation assets and have wide latitude to partner with data centers.

¹ Source: "What we know about energy use at U.S. data centers amid the AI boom." Pew Research Center

² Source: "Nuclear Explained." U.S. Energy Information Administration

³ Source: "Plant Vogtle Unit 4 begins commercial operation." U.S. Energy Information Administration

⁴ Source: "U.S. nuclear generators import nearly all the uranium concentrate they use." U.S. Energy Information Administration

cea08089-c6e9-11f0-ad2f-f315a9e1eaa0