Multiple Ways to Win

U.S. electricity demand should accelerate structurally through 2050 because electrification, manufacturing reshoring and compute infrastructure reinforce one another, making the power thesis resilient even if AI investment disappoints.

• Recent AI-related selling chained together several uncertain forecasts—regulation, slower capital spending, less data-center construction and weaker electricity demand—as though they were inevitable.
• Predictions about transformative AI have often been unreliable: in 2016 Geoffrey Hinton said deep learning would outperform radiologists within five years, while AGI has supposedly remained 20 years away in every decade since the 1950s.
• U.S. electricity demand grew 4–5% annually for decades but was essentially flat during the first 20 years of the 2000s, leaving the grid underinvested just as demand began to reaccelerate.
• TCW expects approximately 3% annual electricity-load growth through 2050, driven by building, vehicle and industrial electrification, reshoring of manufacturing, and demand for compute infrastructure.
• The EIA forecasts U.S. electricity consumption reaching new highs, with growth from both commercial users such as data centers and industrial users such as manufacturers; the thesis therefore does not depend on AI alone.

Takeaway: Investors should assess the power opportunity across multiple scenarios rather than abandon a decade-long structural thesis because of short-term AI headlines, while recognizing that AI regulation, weak returns or other developments could still reduce one of its three growth drivers.