Texas Keeps New Data Center Connections On Hold
Yesterday's clearest message was that data-center development is being tested less by announced demand than by whether projects can clear the practical gates of power access, local approval and finance. Texas's continuing pause on new ERCOT connections remains the most consequential example because it directly affects when prospective campuses can become real load.
Pennsylvania added a different form of friction: proposed statewide standards on developer obligations coincided with a major local siting rejection. Nebraska is beginning its own planning exercise, while the financing behind OpenAI's proposed Ohio campus remains under negotiation. Taken together, these are jurisdiction-specific constraints, not evidence of a single national policy shift.
Texas remains the largest immediate power-access constraint. Governor Greg Abbott's ordered review has kept new data-center connections on hold while ERCOT examines a queue reported at roughly 474GW, compared with recent Texas peak demand of about 91GW. That gap is not a forecast of built load; many requests may never become executable projects. But it explains why queue position alone is no longer enough. For developers, the audit leaves energization schedules, power contracts and financing dependent on a review whose duration and project-level criteria have not been made clear.
Pennsylvania made the local and statewide dimensions of development risk visible at once. Governor Josh Shapiro is pursuing proposed Responsible Infrastructure Standards, known as GRID, that would press developers on power costs, disclosure, water protection and community benefits after broader legislative restrictions did not pass. Separately, Upper Merion Township rejected MLP Ventures' proposed five-building, 4.6-million-square-foot King of Prussia campus. Data Center Dynamics reported that supervisors cited insufficient detail on facility plans and community impacts; MLP intends to appeal. The two actions are not formally linked, but together they show how incomplete impact planning can become a substantive obstacle rather than a communications problem.
The proposed 10GW OpenAI campus in Ohio remains alive in financing discussions, but its support structure appears unsettled. Nvidia is reportedly considering up to $3 billion in SB Energy, while talks include about $100 billion in credit support. At the same time, Nvidia's expected initial guarantee has reportedly fallen below $120 billion from an earlier $250 billion figure. None of those terms is a disclosed completed transaction. The practical point is that a project of this scale must secure credible risk-sharing arrangements as well as land, equipment and electricity.
Nebraska's new 17-member Data Center Task Force is an earlier-stage development, but it is worth noting. The group will examine water, grid and local-infrastructure effects and help counties develop zoning and policy frameworks, following the state's July decision to end taxpayer subsidies for large-scale facilities. It has not yet met or proposed rules, so it is not a new operating restriction. Still, it creates a formal venue for resource and local-government concerns before individual projects reach approval hearings.
Key Points
- Grid queues are increasingly being treated as a planning and reliability problem, not simply a pipeline of future customers. Texas is using its review to distinguish requested capacity from projects that can demonstrate readiness, financing and an acceptable infrastructure plan. That is a consequential change in what developers must prove before access to power can be assumed.
- Disclosure and local legitimacy are becoming part of project execution. The King of Prussia decision turned a lack of detail on plans and community effects into a tangible siting setback, while Pennsylvania's proposed standards and Nebraska's task force point toward more structured scrutiny of power, water and community obligations. Recent briefings have shown similar pressures elsewhere; yesterday made the connection between information quality and approval risk more concrete.
- For the largest AI campuses, financeability is now inseparable from physical delivery. The revised Ohio discussions suggest that credit exposure, lease commitments and guarantees can alter a project's pace even before construction, transmission or power-delivery milestones are settled.
Implications
Project underwriting is widening. A credible data-center proposal increasingly needs a defensible path through at least four linked tests: interconnection, infrastructure-cost allocation, local impact review and capital support. Weakness in any one can delay the others, particularly where power delivery depends on new grid investment.
The emerging constraints remain fragmented. Texas is focused on grid reliability and large-load review; Pennsylvania is combining state-level expectations with municipal discretion; Nebraska is still gathering information. Developers should not read these actions as a uniform U.S. regime, but neither can they rely on a national growth narrative to overcome state and local conditions.
Watchpoints
Watch
Whether Texas specifies the duration and criteria of the ERCOT audit, including how pending queue requests will be treated.
Watch
Whether Pennsylvania's proposed GRID standards become enforceable requirements and whether MLP's appeal changes the King of Prussia outcome.
Watch
What Nebraska's task force recommends after its expected September meeting, particularly on county zoning and infrastructure planning.
Watch
Whether the Ohio campus produces binding lease, investment, guarantee and power-development commitments rather than revised reported discussions.
Fallout
Data-center expansion is becoming more conditional on verifiable power, local-impact and financing plans, although the response remains uneven across jurisdictions.
Final Thought
The important change is not that data-center demand has disappeared. It is that demand is increasingly being asked to prove it can be connected, financed and accepted before it is allowed to shape the grid.
