← Back to feed
6

Opinion: AI Infrastructure Buildout Is Largely Fictional, Grid Data Shows

Infra1 source·Mar 25

Summary

  • • Article argues only ~33% of 241GW disclosed US data center capacity is actively under development
  • • 58% of committed data center capacity lacks a confirmed power generation source
  • • PJM has committed power to data centers at 3x the rate new generation capacity comes online
  • • Author contends the AI buildout is driven by speculative announcements, not executable plans
Adjust signal

Details

1.Insight

Only ~33% of 241GW disclosed US data center capacity is under active development

Drawing on Wood Mackenzie data, the article contends the remaining two-thirds of the pipeline consists of speculative land deals, uncleared permits, and capacity assumptions tied to power generation that does not yet exist. Economist Paul Kedrosky is cited in support of this framing.

2.Industry Update

US data center capacity pipeline additions halved from Q3 to Q4 2025

The article attributes this to persistent 'load-queue challenges' — the backlog of projects waiting for grid interconnection approval. The author reads this deceleration as early evidence that the buildout's momentum is already softening.

3.Insight

58% of committed data center capacity depends on wires-only utility arrangements with no confirmed generation source

In a wires-only arrangement, a utility agrees to deliver power to a site's fence line but takes no responsibility for generating that power. The article argues this means the majority of planned AI campuses have no confirmed path to the electricity they would need to operate.

4.Insight

PJM, one of the largest utility providers in America, has committed power to data centers at 3x the rate new generation capacity is coming online

The article frames this gap not as a minor timing lag but as a structural mismatch that cannot be resolved by optimism alone. It is presented as a concrete illustration of the broader argument that physical grid constraints will eventually override speculative planning.

5.Financials

Total US data center capex ~$948 billion; capex growth decelerated for first time since 2023

The author treats this deceleration as a leading indicator that the AI infrastructure boom is beginning to lose momentum, even before the harder collisions with grid and power constraints materialize.

6.Market Impact

Texas leads US data center pipeline; New Mexico, Indiana, Wyoming saw relative Q4 2025 growth

Geographic distribution data from the article provides context on where speculative and active development is concentrated, though the author's broader argument applies across all regions.

7.Insight

Author argues AI buildout sustained by collective belief in inevitability rather than executable infrastructure plans

The article's central thesis is that the industry operates on the assumption that obvious problems will resolve themselves — and that this faith cannot override the physics of power generation and grid capacity.

Insight = attributed opinion or argument from the author, Industry Update = reported industry development, Financials = capital expenditure data, Market Impact = geographic or competitive market data

What This Means

If the article's argument holds, the AI industry's infrastructure ambitions are significantly overstated, and a painful correction may be coming as grid realities catch up to speculative announcements — with consequences for AI development timelines and the companies betting on continued capacity growth.

Sources

Similar Events