A new analysis warns that delays in data center construction and interconnection queue backlogs are threatening the pace of AI infrastructure growth needed to meet frontier-model training timelines.
Datacenter projects across the U.S. and internationally face mounting obstacles, with major buildouts stalled amid local opposition and regulatory scrutiny. Despite backing from U.S. President Donald Trump, NVIDIA CEO Jensen Huang, and other industry leaders, AI infrastructure expansion is being impeded by two converging forces: critical energy shortages and rising political resistance.
The trend is now unmistakable. This week, Texas Governor Greg Abbott announced a halt to major datacenter projects pending an audit by the Electric Reliability Council of Texas (ERCOT) into energy and water consumption, local impact mitigation plans, and project financing. The audit, introduced in August 2026 with a warning that noncompliance could result in grid exclusion, has effectively stalled all pending developments. "Simply put, Texans must come first," Governor Abbott stated. "Data centers must pay their own way, protect our grid and water, and complete the audits. Until they do, TCEQ [the Texas Commission on Environmental Quality] will issue no permits sought by data center projects."
The Texas pause could affect nearly 50 gigawatts of planned capacity, representing more than $8 billion in anticipated fees through the first quarter of 2027.
Texas is not isolated. In July, New York State imposed a one-year moratorium on datacenter development for facilities of 50 megawatts and above, following substantial community opposition. Multiple other states have proposed similar moratoriums. According to Data Center Watch, regulatory scrutiny of datacenters expanded across 49 states in the second quarter of 2026, with 45 projects valued at $68 billion disrupted by local opposition during the same period.
The delays are affecting major projects directly. Oracle Corp. invoked force majeure—circumstances beyond its control—in communications to Stack Infrastructure, a developer owned by Blue Owl Capital, regarding potential payment deferrals on the Jupiter Project in Doña Ana County, New Mexico. The facility represents a key component of the $500 billion Stargate constellation of international datacenters, announced in January 2025 by Trump, OpenAI, SoftBank, Oracle, and MGX.
Projected to deliver 2.45 gigawatts of compute capacity at a cost of $50 billion over five years, Jupiter has encountered a critical setback: the New Mexico State Land Office has repeatedly denied right-of-way permits for the gas pipeline required to fuel the facility. Oracle, which is financing the project as part of its $300 billion agreement with OpenAI, is now seeking protection from payment obligations should Jupiter miss its planned 2028 online date.
Other Stargate projects have also undergone scope revision. In early March 2026, OpenAI and Oracle announced they would not expand their Abilene, Texas facility from 1.2 gigawatts to approximately 2 gigawatts as originally planned; Microsoft subsequently committed to deploying 700 megawatts of capacity at the site instead.
The obstacles are not confined to the United States. Geopolitical tensions have threatened Stargate development in the UAE. In the United Kingdom, OpenAI halted an 8,000-GPU Nscale datacenter in April 2026, citing power costs and regulatory hurdles. OpenAI is reportedly considering Canada as an alternative, though Trump administration tariffs pose risks to buildouts there as well.
These infrastructure delays coincide with public statements from Anthropic, OpenAI, and SpaceX suggesting deliberate deceleration of AI development to reduce risks of the technology advancing beyond human control. Whether this represents genuine concern or an effort to reframe infrastructure delays as responsible governance remains disputed. Given both companies' IPO ambitions, some skeptics view the narrative as an attempt to mask product stagnation as principled societal protection.
The irony is acute: the very datacenter delays threatening to slow AI progress may force advanced-model developers to abandon planned computational expansions. These firms are the primary customers for the massive capacity buildouts planned across the United States. OpenAI alone has committed to approximately $750 billion in datacenter spending by 2030.
Trump has characterized opposition to these buildouts as a "sick conspiracy" and maintains that his leadership is sufficient to guard against any risks posed by AI. The upcoming midterm elections will test this assertion, pitting competing voices across the political spectrum against one another and revealing credibility gaps on fundamental questions about AI governance. The outcome may deliver precisely the slowdown in AI development that the major labs have not anticipated and cannot control.