Breaking

Wood Mackenzie: Over Two-Thirds of U.S. AI Data Center Power Requests Unlikely to Materialize

August 12, 2026 at 07:20 EDT

Infra & Chips
Policy & Regulation

More than two-thirds of the electricity requested for U.S. artificial-intelligence data centers is unlikely to be delivered, according to new Wood Mackenzie projections, as "phantom" applications and speculative pitches flood grid queues and distort demand forecasts.

Aug 12, 2026 · AI Power Demand

Most AI Power Requests Won't Materialize

Wood Mackenzie projects U.S. utilities will commit to only about 28% of 1,066 GW sought for data centers—leaving more than two-thirds as phantom or long-shot load.

Requested load vs likely commitments

1,066 GW

Power requested

~28%

Likely commitments

Column height is proportional: full stack of requests vs the share utilities are expected to commit.

>2/3

Unlikely to materialize

474 GW

Texas ERCOT queue

~40%

Exelon pipeline cut

83%

Of U.S. gen. capacity (requests)

What filters help

Collateral, credit screens, and first-ready queues aim to separate real campuses from shotgun filings.

What still bites

Phantom load still stretches timelines and can push upgrade costs onto ratepayers if plans overshoot.

How phantom demand piles up

Shotgun multi-utility pitches→Queue bloat & slow studies→Only ready projects advance

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.