Mar 5, 2025 - Energy & Climate

New analysis tackles physical barriers to AI "dominance"

Ben Geman

an illustration of a toll entrance barrier with binary code as the caution stripes — Illustration: Tiffany Herring/Axios

A think tank's new analysis provides eye-popping estimates on data center power and chip needs — and offers Trump 2.0 officials ideas for meeting them.

Why it matters: "[W]hile the AI revolution is digital in nature, its binding constraint is physical infrastructure," states the Center for Strategic & International Studies report about keeping the U.S. lead on AI.

Stunning stat: It sees up to 90 gigawatts of new U.S. data center energy needs by 2030, though the range of estimates is large.

That's up from 32 GW today. Their tally includes data centers overall, not just AI needs, but that's where growth is focused.

State of play: The report floats ideas for using President Trump's energy "emergency" declaration to speed permitting.

It urges a "particular focus on the Northern Virginia computing cluster where data center expansion faces severe constraints."
It calls for fast-tracking gas pipelines, transmission, and generation of varying stripes — including Virginia offshore wind, despite Trump's opposition.

What we're watching: The many recommendations also suggest a muscular DOE role in speeding nuclear projects.

One concept: offtake contracts for new projects to help support 10 GW by 2030.
DOE would sell offtake capacity to hyperscalers or "transfer [it] to utilities to enable ratepayer access to nuclear power."

Add Axios on Google

New analysis tackles physical barriers to AI "dominance"

What to read next