Analysis
I have watched a lot of infrastructure categories acquire a political problem, and the tell is always the same: it stops being a national debate and becomes a county commission meeting. That is where AI data centers are now. Governor Greg Abbott, whose state spent years marketing cheap power and permissive permitting to hyperscalers, told Axios that the industry "dug their own grave" on public opinion in Texas.
The grievances are concrete rather than ideological. Residential electricity bills went up in several markets that hosted large campuses. US data centers tripled their water footprint over ten years, which is a serious argument in a state that has had drought restrictions. And the jobs math is genuinely weak -- a billion-dollar campus employs a few dozen permanent staff, which is a hard trade to sell to a community absorbing the grid load.
What makes this politically strange is the federal counter-signal. President Trump has said rejecting data centers is "a mistake" in national-competitiveness terms, while Senator Bernie Sanders has pushed Democrats toward a pause on AI development that most 2028 contenders have declined to endorse. Neither party has a settled position, which means siting outcomes get decided locally, case by case, by people with no partisan script to follow.
“Residential electricity bills went up in several markets that hosted large campuses.”
The economics that made Texas attractive are also what makes the backlash durable. ERCOT's deregulated market let hyperscalers sign power deals fast, and the same market structure passes wholesale price spikes through to residential customers with very little insulation. When a campus in Abilene or the Permian adds load equal to a mid-size city, the effect shows up on household bills in the same billing cycle, which is the shortest feedback loop between an industrial project and a voter that I can think of.
Room for disagreement: you can read Abbott's comment as political positioning rather than a policy shift, and Texas has approved an enormous amount of capacity that is still under construction regardless of the rhetoric. Capital has not slowed. The counterargument I take seriously is that permitting friction mostly redirects projects to friendlier jurisdictions rather than stopping them, and that the buildout continues at roughly the same pace with a different map.
Still, if you are underwriting an AI infrastructure deal today, the schedule risk that used to sit in transformer lead times now also sits in community opposition, and that is a variable most 2024-vintage models did not include.
