On September 22, 2026, Governor Greg Abbott directed the Public Utility Commission of Texas and ERCOT to audit every new data center proposal seeking to connect to the state's electric grid. The reason is straightforward: ERCOT is currently processing 474 gigawatts worth of connection requests. Texas's record peak electricity demand is about 85 gigawatts. Data centers account for roughly 90 percent of those new requests. The math does not work.
This is not a moratorium — not yet. New York already went there, halting new data center construction in parts of the state earlier this year. Texas is taking a softer approach: you can still build, but you have to prove your facility will not destabilize the grid before you get connected. That means disclosing how much power you plan to pull, where your water comes from, what state and local incentives you received, and how you will handle noise complaints from the people who live next door.
Why this matters if you run AI workloads
If you are running AI inference, training models, or operating automation pipelines that depend on cloud infrastructure, this audit directly affects the supply side of your compute costs. There are already at least 335 data centers operating in Texas and at least 248 more planned, making it the second-largest data center market in the country behind Virginia. When the state starts slowing down new connections, the supply of available compute in Texas tightens. Tighter supply means higher prices. Higher prices mean your cloud bill goes up — not because your usage changed, but because the physical infrastructure behind your API calls is hitting a wall.
The 474 GW number is staggering on its own. That is more electricity than the entire United States generates on an average day. Not all of those proposals are serious — many are speculative land grabs by developers who want to secure grid connections and figure out the rest later. But even if only 10 percent of those requests materialize, Texas would need to roughly double its generation capacity. That takes years and billions of dollars. The audit is essentially ERCOT saying: we need to separate real projects from vaporware before we commit resources.
The water problem nobody wants to talk about
Abbott also told state regulators to penalize data centers that do not disclose water usage. This is the less glamorous side of AI infrastructure. Data centers consume enormous amounts of water for cooling, and in a state where drought is a recurring headline, that consumption competes with agriculture, residential use, and everything else. A single large data center campus can use millions of gallons per day. Multiply that by 248 planned facilities and you start to see why the state is paying attention.
The water angle matters for a different reason, too. Cloud providers have been quietly shifting their expansion plans toward regions with cheaper power and more permissive regulation. Texas was at the top of that list. If water and grid constraints start gating new construction, the migration pattern changes. Providers will look harder at the Midwest, the Southeast, and anywhere else with capacity. For customers, that means your data might end up farther from you, which adds latency, which affects real-time applications and API response times.
The bigger picture: AI's infrastructure reckoning
This is not just a Texas story. Data center opposition is becoming bipartisan. Communities that once welcomed the tax revenue are pushing back on noise, water consumption, and grid strain. Virginia — home to "Data Center Alley" — has seen local opposition groups block or delay multiple projects. New York imposed its moratorium. Now Texas is adding friction to the connection process.
The pattern is clear: AI's growth is bumping into physical limits. You cannot run a 500-GW workload on an 85-GW grid no matter how much venture capital you raise. The era of unlimited, frictionless data center expansion is ending, and the states that host this infrastructure are starting to demand accountability.
For anyone building on top of AI services — whether that is WordPress automation, content pipelines, customer support agents, or full-blown agentic workflows — this is the kind of structural shift that shows up in your bill six months from now. The models keep getting cheaper (OpenAI just halved GPT-6 pricing yesterday), but the infrastructure to run them is getting more expensive and harder to build. Those two forces will eventually collide.
What to actually do about it
If you are a small business or developer relying on a single cloud region, this is a good time to think about redundancy. Not because the lights are going out tomorrow, but because the assumption that compute will always be cheap and abundant is eroding. Multi-region deployments, edge computing, and optimizing your inference calls all become more valuable when the underlying infrastructure is under pressure.
For the automation crowd running n8n, Make, or Zapier workflows that call AI APIs: your tool costs are going down, but watch the hosting side. If your workflows depend on low-latency responses from a single region, and that region starts constraining new capacity, you will feel it.
Texas just drew a line. Other states will follow. The question is no longer whether AI infrastructure will face regulation — it is how fast, and how much it will cost the people who depend on it.
Sources: The Verge — Texas says data centers must pass an audit before connecting to the grid, Texas Tribune — Texas regulation of data centers