🟢
Confirmed
Aug 22, 2026
Texas Halted 1,800 Data Centers
Governor Greg Abbott says his directive has stopped up to 1,800 data center projects — and the conditions he set are the part that matters. Facilities must not take water communities need, must not take power the grid needs, must not disrupt neighborhoods, and must lower the cost of electricity. That third and fourth requirement together are close to unmeetable. The market has spent a year debating whether AI's power constraint is a grid interconnection problem, on the assumption that developers could build their own generation and route around the queue. Texas just demonstrated that the binding constraint is permission to build at all, and permission is not something a gas turbine solves.
THE SIGNAL
Governor Greg Abbott, posting directly: "My directive has halted up to 1,800 data center projects. I established guidelines and guardrails to protect Texas communities: data centers must not take water needed by local communities, they must not take power needed by the Texas power grid, they must lower the cost of electricity, they must not disrupt neighborhoods or rural communities. These guidelines are now the universally applied standard in Texas."
This follows a statewide moratorium pending project audit announced earlier in August. Texas had been on track to pass Virginia as the largest US data center market.
Two other states have moved in the same direction. New York imposed a one-year moratorium on facilities above 50MW in July. Pennsylvania imposed strict developer rules on August 18, and the announcement weighed on independent power producers and on HPC operators simultaneously.
Pennsylvania sits inside PJM. PJM's 2028/2029 capacity auction cleared at the $325 per MW-day price cap, came up roughly 6.8GW short of its 20% reserve margin target, and attracted only about 525MW of new resources. PJM describes these as the first auctions in its history in which the entire RTO fell short of the reliability requirement.
View source ↗
2026-08-21
THESIS CONNECTION
The consensus workaround for the power constraint has been on-site generation. If the interconnection queue is the bottleneck, build your own gas turbines and bypass the grid entirely. SpaceX is doing exactly that for its Texas chip facility. PJM's own proposed framework offers developers the choice: bring your own new capacity and get firm service, or accept interruptibility. That is a policy explicitly designed around self-generation as the answer.
Texas is a different kind of constraint and the distinction is the whole finding. An interconnection queue is a technical problem with a technical solution. A permitting standard is a political decision, and no amount of private generation satisfies a governor who has decided the projects themselves are the problem.
Look at the four conditions. Not taking community water is meetable with closed-loop cooling at higher cost. Not taking grid power is meetable with on-site generation. Not disrupting neighborhoods is a siting question. But requiring that a data center lower the cost of electricity for everyone else inverts the entire economics of the load. PJM's capacity prices went from under $35 per MW-day to $325 in three auction cycles, and the market monitor attributes a substantial share of that increase to data center demand. A facility that consumes gigawatts cannot make power cheaper for its neighbors. It can only fail to make it more expensive, and Texas has written the stronger version into the standard.
The market appears to have understood this before the file did. When Pennsylvania moved, the selloff hit independent power producers and data center operators together. If this were purely an interconnection story, the on-site generation complex should have caught a bid on the same news — more restrictions on grid access means more demand for private turbines. It did not. The tape priced it as project risk, not routing risk.
That matters for how the capex numbers should be read. Alphabet, Amazon, Meta and Microsoft carry roughly $1.52 trillion of purchase commitments and $904 billion of leases not yet started. "Leases not yet started" is a category that depends entirely on projects reaching construction. A permitting regime that halts 1,800 projects in the fastest-growing market in the country is a direct claim on that line.
WHAT TO WATCH
- The number itself needs sourcing beyond the governor's own post. 1,800 is large enough that the definition matters — whether it counts distinct facilities, permit applications, or projects at any stage of inquiry. Pull the underlying directive and the audit scope before citing it.
- The discriminating test: whether the turbine and on-premise generation names diverge from the IPPs and data center operators from here. Divergence means the market prices the workaround as viable. Continued correlation means it does not, and the entire bring-your-own-power thesis weakens with it.
- Watch whether a fourth state moves, and whether any of the three softens under pressure. Three is a pattern. A reversal in Texas would falsify this quickly, and the state has strong incentives to attract the investment back.
- Watch PJM's backstop procurement auction running September 30 to October 21, which is attempting to fill the 6.8GW shortfall. FERC's chairman has warned the commission will impose reforms if PJM does not act by September.
- The counterweight worth tracking honestly: none of this reduces demand for compute. It relocates it. Projects blocked in Texas, New York and Pennsylvania may reappear in states competing for the investment, which would make this a siting story rather than a capacity story. Watch permit filings in the Southeast and Mountain West.
⚡ The power constraint moves from a queue problem with an engineering workaround to a permission problem with a political one, which makes the timing of the buildout dependent on state politics rather than on turbine lead times.