AI Boom Strains US Power Grid and Infrastructure Planning

The rapid expansion of artificial intelligence is placing significant strain on the United States power grid, with Goldman Sachs Group Inc. estimating a need for 50 gigawatts of new capacity. This surge has led to a volatile environment where tech companies request massive projects across multiple regions to secure power, a practice known as data center shopping. However, the risk of stranded assets is rising; for example, Microsoft Corporation abandoned proposed projects in the U.S. and Europe totaling 2 gigawatts in March.

David Rosner, Chairman of the Federal Energy Regulatory Commission, warned that inaccurate load forecasts jeopardize billions of dollars in investments. While some tech leaders aim for aggressive growth, Constellation Energy CEO Joe Dominguez cautioned that projected electricity loads may be overstated. The financial stakes are high, as utilities may pass infrastructure costs, averaging $102 million per gigawatt, onto ratepayers if proposed centers do not materialize.

To bypass long interconnection delays and grid constraints, AI firms are pursuing behind-the-meter power generation. These strategies include investing in small modular reactors, mobile gas generation, and restarting shuttered nuclear plants. Nvidia CEO Jensen Huang has advocated for data centers to self-generate power to move faster than grid integration. Meanwhile, the speed of capacity expansion remains uncertain due to regulatory hurdles and political opposition from President Donald Trump toward wind and solar energy.