Texas regulators are weighing whether to require full emergency curtailment for an expanding AI campus from Crusoe and Ensign Infrastructure after staff at the Public Utility Commission of Texas (PUCT) endorsed operating conditions proposed by the Electric Reliability Council of Texas (ERCOT).

The proceeding is among the first to test Texas’s new rules for colocating large electric loads with existing generation. The approach uses existing interconnections to shorten timelines to power, but raises questions about how reliability obligations should scale when a load sits behind a generator’s point of interconnection (POI).

The application centers on Crusoe and Ensign Infrastructure’s proposal to add a 260 MW “Crusoe Load 2” AI data center behind the 265.5-MW Goodnight 1 wind farm in Armstrong County. Combined with the previously approved 265.5-MW “Crusoe Load 1,” the AI campus would create 525.5 MW of behind-the-meter load sharing Goodnight 1’s existing POI. “Behind-the-meter” in this context means the data centers would be electrically located on the generator’s side of the interconnection with the ERCOT grid.

Related:Texas AI Data Centers: Power, Policy, and Progress

PUCT staff and ERCOT argue that both AI facilities should remain capable of curtailing during grid emergencies so Goodnight 1’s output can be made available to the ERCOT market when needed. Ensign argues the proposed conditions exceed what Senate Bill 6 authorizes for the project.

In a June 26, 2026, initial brief, PUCT staff recommended approving the application subject to the same six operating conditions the Commission imposed earlier this year on Crusoe Load 1. Those conditions include a requirement that Crusoe Load 2 curtail within 30 minutes – or within 10 minutes if it later demonstrates that capability – when ERCOT determines Goodnight 1’s generation is needed to address a transmission emergency or an Energy Emergency Alert.

The dispute follows the Commission’s recent adoption of PUCT Rule 25.205, which implements provisions of Senate Bill 6 and establishes procedures for reviewing net-metering arrangements between existing generators and large loads colocated behind a shared POI.

Commission staff asserts that Crusoe Load 2 should operate under the same framework as Crusoe Load 1 because both facilities share Goodnight 1’s POI. ERCOT makes a similar point, adding that developers could have avoided this review by constructing a separate POI for the expansion. By choosing the existing interconnection, ERCOT argues, the project became subject to SB 6’s review process.

Related:Texas May Have Accidentally Built the Perfect Grid for AI

A Fight Over How Curtailment Should Scale

Ensign does not dispute the Commission’s authority to impose reliability conditions but argues ERCOT’s proposed requirements exceed what PURA §39.169 authorizes here. The company notes that Crusoe Load 1 already must curtail up to 265.5 MW during emergency conditions under the Commission’s earlier order. Because that obligation restores Goodnight 1’s generating capability to the ERCOT market, Ensign argues there is no remaining reliability deficit requiring Crusoe Load 2 to curtail as well. Ensign also urges regulators to evaluate net system effects rather than total behind-the-meter load, citing ERCOT studies that found no thermal or voltage violations and no need for new transmission facilities.

PUCT staff rejects that interpretation, asserting the shared POI – not the sequence of projects behind it – should govern the analysis. On that view, both data centers must remain capable of curtailing so that Goodnight 1’s available capacity is accessible to ERCOT during emergencies. Staff also discounts reliance on the planned Goodnight 2 wind project, arguing it is outside the scope of the current application. ERCOT similarly says the future project does not change the statutory analysis because Crusoe Load 2 would still interconnect behind Goodnight 1.

Related:Interconnection Delays Push Texas Data Center Behind the Meter

Implications Beyond Goodnight

Chris Talley, co-founder of Interconnection.fyi, said the Commission’s decision could influence how developers design future AI campuses in Texas. “The precedent is really about proportionality,” Talley said. “If ERCOT wins, colocating any load with any qualifying standalone generator locks in 100% curtailment risk on the entirety of that load, plus exclusion from all demand response programs, regardless of how small the generator is relative to the load.”

Talley pointed to a hypothetical raised in Ensign’s brief – whether a 1 GW data center colocated with a 50 MW generator could be required to curtail the full 1,000 MW facility in emergencies.

“If ERCOT wins, colocation becomes viable primarily for developers who can build on-site backup generation sized to serve their entire facility load, because ERCOT’s condition requires the full load to fully curtail during emergencies, not just enough to offset what the colocated generator was contributing,” Talley said. He cited Amazon Data Services’ pending “Project Spectrum” application involving the Comanche Peak nuclear plant as a different approach, where the data center would transition to customer-owned backup generation while returning the plant’s output to the grid in emergencies.

Talley cautioned against viewing ERCOT’s position simply as opposition to AI development. “They’re navigating a really hard tradeoff to figure out how to preserve grid reliability as data center load grows without shutting out the developers that Texas wants to attract,” he said. Colocation can reduce the need for new transmission and improve grid resilience if operators can return the generator’s output to ERCOT during emergencies. He also noted that many inference-focused facilities and hyperscale data centers with strict uptime SLAs plan for full backup capacity regardless, which could blunt the policy’s deterrent effect.

A PUCT decision is expected later this summer. Along with several other pending Senate Bill 6 proceedings, the ruling will offer an early indication of how Texas will apply its colocation framework as AI developers race to secure power for new campuses.