FERC Ordered Every US Grid Operator to Reform AI Interconnection This Week — and the Government Pulled an AI Model Offline the Same Week

FERC's Show Cause Orders Are the Fastest Major Grid Regulatory Action in a Generation

The show cause orders FERC issued on June 18 under Section 206 of the Federal Power Act represent an unusual exercise of regulatory authority. Section 206 allows FERC to initiate proceedings to determine whether existing rates, terms, or conditions of service are unjust or unreasonable, and to require utilities to show cause why they should not be changed. Applying this authority to interconnection queue frameworks — rather than to specific rate disputes — and directing it simultaneously at all six major regional grid operators is a move without close precedent in FERC's modern history. The practical effect is to compress what would normally be a multi-year rulemaking process into a proceeding that can require responses in weeks and produce interim orders within months. FERC's stated justification is that CoreWeave alone is targeting 1.7 gigawatts of data center capacity by end of 2026, and that the current interconnection frameworks produce average wait times of four years or more for large-load customers, meaning facilities under construction today will not receive utility interconnection until 2029 or 2030 under current queue management.

The geographic concentration of AI data center development makes the show cause orders more consequential than a national average would suggest. Virginia's data center market — the largest in the world, accounting for roughly 70 percent of US data center capacity — operates primarily within PJM Interconnection's territory, one of the six grid operators receiving a FERC show cause order. The interconnection reforms PJM proposes or is ordered to implement will determine whether the physical infrastructure of the AI era gets built in the places where AI companies want to build it, or whether data center siting decisions shift toward regions with faster interconnection queues at the cost of proximity to population centers, fiber infrastructure, and the skilled workforce that data center operations require. The industry's response to the FERC orders — which will require each regional operator to either defend the status quo or propose specific reforms by a FERC-set deadline — will be the most consequential grid policy proceeding for AI infrastructure in the US for at least the next decade.

The Export Control Collision Is the Regulatory Paradox No One Has a Framework For

The federal government's June 12 export control directive that forced an AI lab to pull its two most capable frontier models offline globally — on national security grounds — arrived six days before FERC declared AI grid integration a national priority. The juxtaposition captures a genuine regulatory contradiction that the US government does not yet have a framework for resolving. The argument for accelerating AI infrastructure is that US AI capability is a strategic national asset that requires massive physical investment in power, cooling, and connectivity infrastructure to maintain. The argument for export-controlling advanced AI models is that those same models represent a national security threat when available to foreign adversaries. Both arguments are being made simultaneously by different agencies of the same government, and they point toward opposite conclusions about how aggressively to build and deploy AI systems that are both economically strategic and potentially dangerous. The administration's response to the contradiction — accelerate infrastructure while restricting access — is a coherent short-term posture, but it raises a question that neither FERC nor any export control authority has the institutional mandate to answer: at what capability level does AI become something that the US builds strategic infrastructure to house but does not actually deploy broadly?

For the technology industry, the practical planning implication of these two developments arriving in the same week is that AI infrastructure investment and AI product deployment are now governed by separate and potentially conflicting federal policy tracks. A company investing in US data center capacity in response to FERC's grid integration priority is making an infrastructure bet that is consistent with one axis of federal policy. A company deploying frontier AI models is operating on a different axis that can produce export control restrictions, access revocations, and liability exposure that the infrastructure investment was not designed to accommodate. The companies best positioned in this environment are those building infrastructure designed for a range of model capability levels — not optimized for the most powerful systems currently deployable — so that their physical assets remain useful regardless of how the export control and AI governance debates resolve over the next 18 to 24 months.