Texas froze new data center grid connections until every project is audited
Governor Abbott ordered ERCOT and the PUC to audit every pending data center project's power, water, and ownership before connecting anything new. The queue stands at 474 gigawatts, five times state peak demand.
ERCOT's queue hit 474 gigawatts, five times peak demand. Texas just pulled the handbrake.
Texas Governor Greg Abbott ordered state energy regulators on August 3, 2026 to stop approving new data center connections to the ERCOT grid until every pending project passes a compliance audit. The number behind the order: ERCOT's interconnection queue has reached 474 gigawatts of requests, roughly five times the state's peak demand, and data centers account for 90 percent of the incoming load.
What the order actually does
The directive tells ERCOT and the Public Utility Commission of Texas to collect and verify, project by project: expected power consumption, water use, on-site generation plans, tax incentives, ownership structures, and impacts on surrounding communities. No timeline has been announced for finishing the audits or resuming approvals; ERCOT says it will consult the commission before the PUCT's August 20 open meeting. The Data Center Coalition, the trade group for the largest operators, said it hopes the audit separates responsible developers from the rest, which tells you the queue includes projects even the industry does not expect to be real.
A queue five times the grid
474 gigawatts is not a forecast of demand, it is a stack of applications, and that is the problem the audit is trying to solve. Speculative interconnection requests, filed to reserve grid capacity for projects that may never break ground, make it impossible for a grid operator to plan real generation against real load. Texas built its reputation on being the fastest place in America to plug in a data center, which is exactly why its queue filled with paper projects first.
Why a build studio cares
Texas is the largest single buildout region for the compute our AI work runs on, and this is the second time in a week the physical layer pushed back, after the memory shortage chipmakers warned will last through 2028. Every AI roadmap quietly assumes compute keeps getting cheaper and more available. Interconnection freezes, audits, and water politics are the mechanisms by which that assumption can fail, so we track them the way we track a dependency with a known issue.
Next step: read the Texas Tribune's report and TechCrunch's coverage. If your product's unit economics depend on inference staying cheap, write to us at hello@gattyworks.com.