Shared Interconnection Buys Speed. Disciplined Developers Already Underwrite the Curtailment.

The Public Utility Commission of Texas staff has endorsed ERCOT's operating conditions for Crusoe Energy and Ensign Infrastructure's AI campus behind the Goodnight 1 wind farm in Armstrong County, one of the first tests of Texas's framework for colocating large loads with existing generation. The fight is over whether Crusoe Load 2, a 260 MW facility, must curtail within 30 minutes during grid emergencies like Crusoe Load 1 (265.5 MW) already does, since both share Goodnight 1's single 265.5 MW point of interconnection.

Ensign argues ERCOT's conditions exceed what Senate Bill 6 and PURA §39.169 authorize, since Load 1's curtailment alone restores the wind farm's full capacity. PUCT staff rejects that: the shared POI, not project sequence, sets the obligation. This ruling will define the tradeoff every behind-the-meter developer now faces: shared interconnection buys speed but binds you to grid-yield obligations.

Brookfield expanded its fuel-cell partnership with Bloom Energy to $25 billion, up from $5 billion, bundling capital, power delivery, and sequencing into one product so hyperscalers can advance campuses while ERCOT's 300+ GW queue clears. Neil Osnato of Persistence Analytics Group framed it as energy certainty becoming financeable. Wärtsilä's 790 MW off-grid Texas project and GE Vernova's gas-turbine approach chase the same certainty. But NERC's 2026 State of Reliability report flags the physics risk: gigawatt-scale campuses shed load instantly during faults, with a February event dropping 1,800 MW and a June event 1,300 MW, plus nine ERCOT crypto-mining load-loss events over 100 MW.

The financing only works if the loads perform. Developers who lock anchor tenants, validate load assumptions, and build ride-through capability before regulation requires it carry the upside. Those banking on speculative demand and instant-shed equipment carry the downside, and now the modeling scrutiny too.

Curtailment authority is live elsewhere. Texas is writing its rules by hand.

PJM deployed its curtailment power June 30: operators without co-located generation get cut first during heat emergencies. Capacity prices there ran from $28.92 per megawatt-day in 2024/25 to $329.17 in 2026/27, with PJM's Market Monitor pinning 63% of one year's increase on data center demand and $9.3 billion in ratepayer costs. Oregon's POWER Act goes further, forcing a 29% rate hike on large PGE loads starting July 8 while residential rates drop 1.6%. Nine other states are weighing similar cost-allocation frameworks. ERCOT and the PUCT don't operate under comparable cost-shift mandates, but the momentum is unmistakable.

Talen's $18 billion, 1,920 MW nuclear deal with AWS and Constellation's Three Mile Island restart for Microsoft both predate PJM's curtailment authority. The developers who secured firm generation early sidestepped the whole math. In Texas, the move is to lock long-term contracts and grid-connected commitments before the itemizing starts.

Campaigning in Bullard (population 5,500), Governor Abbott called for restricting new rural data centers, but his actual framework is three mandates: contribute power generation to ERCOT's grid, run closed-loop water systems, and lose access to certain state incentives. San Marcos became the first Texas city to attempt a data center ban via a 4-3 vote; two counties tried bans and lost legal challenges per the Texas Tribune. In North Richland Hills, Mayor Jack McCarty declared opposition to residential-adjacent data centers just as Provident explores a 50 MW inference facility on Rufe Snow Drive requiring a zoning change. (Fox RGV's account of Abbott's push is thin-sourced; The Real Deal corroborates the specific mandates.)

The framework favors balance-sheet discipline over speculation. Developers arriving with anchored ERCOT-connected power, secured water offsets, and mapped council sentiment navigate it cleanly. The rural greenfield play built on subsidy assumptions is where the execution risk sits.