Texas Paused the AI Data Center Boom. Here's Why That's Weird.
Governor Greg Abbott has been pretty clear about where he stands on AI infrastructure. Last year he called Texas the "epicenter of AI development" and spent the state's resources courting data center developers with cheap land, relaxed regulations, and tax breaks. Now he's hit the brakes.
Abbott's office announced on August 3 that all new data center projects must undergo a comprehensive audit before they can connect to the ERCOT power grid. The directive went to both the Public Utility Commission of Texas and the grid operator itself. Developers waiting in the interconnection queue are now on hold.
The timing is notable. Texas has been aggressively pursuing the AI infrastructure boom. The state now trails only Virginia in total data center square footage and has been adding more connections than any other state except for a few early movers according to industry trackers. But the ERCOT interconnection queue tells a different story. As of early August, the queue held more than 1,800 projects requesting over 474 gigawatts of connection capacity. That's roughly five times Texas's all-time peak electricity demand. About 90 percent of those requests come from data centers, according to the grid operator.
ERCOT's own forecasts suggest data center demand could double the state's current peak by 2032.
The audit requirement is broad. Developers will need to provide projections of annual and peak electricity consumption, information on water usage dependencies, and details about any state financial assistance they're receiving. The directive also asks for noise mitigation plans, light controls, setbacks, traffic improvements, and emergency response coordination.
What's missing from the directive is almost as telling as what's in it. Abbott's order doesn't address local air pollution or greenhouse gas emissions from data centers. Several AI companies have already used a Texas permit loophole to install gas-powered turbines and backup diesel generators on their data center campuses without triggering more extensive environmental reviews according to Floodlight. The directive also doesn't mention the water usage associated with power generation itself. Data centers often draw more water from the plants that generate their electricity than from their cooling systems directly.
There's a reason for that gap. The pause doesn't apply to projects building their own on-site power generation. The "behind-the-meter" strategy has become increasingly popular among major AI companies. Meta, Microsoft, Amazon, OpenAI, and Anthropic have all announced plans for on-site generation, according to market intelligence platform CleanView. Most of that capacity currently relies on natural gas, though some developers are turning to mobile generators strapped to semitrucks and aeroderivative turbines originally designed for aircraft.
Texas leads the country in announced behind-the-meter capacity for data centers, sitting at 40 gigawatts. That advantage comes from the Permian Basin gas supply, extensive pipeline infrastructure, and the state's regulatory environment.
The political context matters too. Abbott's approval ratings have been softening over concerns about data centers and transmission lines in rural communities. Polls show Texas voters generally opposing new data centers in their areas. Democratic challenger Gina Hinojosa called the pause "inadequate political theater" and wants a full moratorium instead.
The Texas Tribune reported that the state already provides more than $1 billion in annual tax breaks to data center developers, with estimated sales tax revenue losses of $3.2 billion over two years.
This isn't just a Texas problem. Virginia, which hosts the most data centers in the US, has seen its own legislature debate moratoriums. New York Governor Kathy Hochul announced a yearlong ban on data center construction in July to give the state time to develop standards. The national resistance to data centers appears to be building.
I keep thinking about what happens to all those queued projects. A lot of them are probably still just paper proposals at this point. Getting in line for grid connection is often the first move developers make, and many projects fizzle out before they reach construction. But even a fraction of those 474 gigawatts coming online would fundamentally change how Texas uses electricity.
The pause creates an awkward moment for a state that built its reputation on being friendly to big infrastructure projects. Abbott is now asking for more information from developers who largely ignored his earlier voluntary survey request. Most didn't respond to that one either.
What's interesting here is the tension between the AI boom's appetite for power and the physical limits of the grid. Texas has added more new power generation than any other state, and utility-scale solar capacity grew fourfold between 2021 and 2025. But the interconnection queue suggests that demand is still outpacing supply in a way that concerns even the state's most enthusiastic proponents.
The audit results could reshape Texas's position in the AI infrastructure race. If the findings show that data centers are destabilizing the grid or consuming more water than expected, the state might tighten regulations significantly. If the audit clears most projects, the boom continues.
Either way, the pause signals that the "move fast and connect later" approach to AI infrastructure is running into real physical constraints. The grid doesn't care how much capital you've raised or how big your order book is. It has limits.
Related
More from the blog
Inertia Enterprises' Fusion Fuel Breakthrough: Weeks to Minutes
Inertia Enterprises cut fusion fuel pellet production time from weeks to minutes, solving a critical capex bottleneck for commercial fusion
Orbital Data Centers and the E-Waste Problem in Low Earth Orbit
Orbital data center plans ignore the looming e‑waste crisis in low Earth orbit, where defunct hardware becomes permanent hazardous debris.
Kog Bets the GPU Arms Race Is a Software Problem
French startup Kog claims 30x faster LLM inference through pure software optimization on standard GPUs, challenging the assumption that better hardware is the only path forward.
CoreWeave's $2.58 Billion Quarter Proves AI Is Devouring Crypto's Money
CoreWeave reported $2.58 billion in Q2 revenue with a $104 billion backlog, proving AI infrastructure demand is shifting capital away from crypto toward compute.