The Biden administration recently announced plans to work with tech firms on power usage. Technological revolutions have always come with novel challenges: building infrastructure, navigating dual-purpose applications and preparing our workforce. For the growing industry surrounding AI, this challenge is energy. Progress in AI will soon face the limits of the power available on the grid. As the administration now recognizes, maximizing AI’s potential will require an energy breakthrough. While the nation needs diverse approaches, the best solution is the promise of fusion.

Currently, AI models are run from data centers. These models require power to build, or “train,” and to run, or “inference.” Both are energy-intensive. Combined, the vast expansion of AI means that the electricity consumption of building and running AI models may soon match the power generation of small countries — and then grow beyond that. Estimates project that a single utility in Virginia will see annual data center power demand grow by nearly 79,000 GWh by 2038 — the equivalent of adding nearly 6.1 million homes to the grid.