The AI Data Center Just Became National Infrastructure
On June 18, the Federal Energy Regulatory Commission issued orders to all six regional grid operators in the United States. The orders didn't make much news outside of energy industry circles, which is a shame, because they represent something worth paying attention to. FERC directed PJM, MISO, SPP, CAISO, ISO New England, and NYISO to either justify their existing interconnection rules or propose reforms specifically designed to accelerate power connections for large-load customers. The commission bypassed the normal rulemaking process, which typically takes years. FERC Chairman Laura Swett explained the urgency simply: "This is the biggest priority our country is facing at the moment."
She was talking about AI data centers. And she was not being rhetorical.
What "national infrastructure" means in practice is that a technology stops being treated as an industry and starts being treated as a road. You don't permit roads through a normal market process; you designate them, prioritize them, route around obstacles to build them. When FERC tells six grid operators to reorganize around one category of customer in 60 days flat, that's the regulatory equivalent of eminent domain. The AI data center is being made a road.
The orders cover five areas: faster interconnection study timelines, rules preventing data centers from shifting transmission costs onto residential and commercial customers, frameworks for co-location (data centers running their own on-site generation), new flexible-load services, and study processes for generation facilities built specifically to serve AI computation. That last point is worth sitting with. FERC is now creating regulatory frameworks for power plants whose entire purpose is to generate electricity for AI.
Every era of transformative technology eventually bends physical infrastructure policy. The railroads forced federal land grants and eminent domain authority in the 1860s. Telecommunications forced the Communications Act of 1934. The interstate highway system required fifty years of incremental federal authority before Congress funded it all at once in 1956. Each time, the pattern is similar: a technology achieves economic and strategic weight such that existing infrastructure can no longer contain it, and the government steps in to reshape physical reality.
AI took about three years to get from "interesting research" to "national grid emergency." That is unusually fast.
Part of what makes this moment strange is that the decision was made with almost no public visibility. The DOE Secretary asked FERC for action in 2025. FERC agreed. Six grid operators now have 60 days to respond. Those regions serve approximately 200 million Americans across more than 30 states, and their power infrastructure is being reorganized around AI data center priorities on a timeline that did not include a public comment period of any real consequence.
This is not a straightforward criticism. There are genuine arguments that AI infrastructure is economically and strategically important enough to justify priority treatment. The competitive dynamics reshaping the AI market right now are playing out against an infrastructure competition most people do not see. Data center build-out in the United States is a strategic bet on whether American AI labs will have the compute advantage to stay ahead. If you accept that premise, prioritizing their grid connections follows.
But "the government is making a major infrastructure commitment on behalf of AI companies, without a public vote" is a sentence that probably deserved more deliberation than it got.
The cost-shifting provisions are where the practical stakes become clearest. FERC's order explicitly tells grid operators to build safeguards ensuring that faster interconnection for data centers does not translate to higher bills for everyone else. This is a recognition that queue priority is a zero-sum resource: if a one-gigawatt data center moves to the front of the interconnection queue, something else moves back. FERC wants that "something else" not to be a residential customer's electricity rate.
Whether those safeguards will work is an open question. Transmission costs are notoriously difficult to allocate fairly, and data centers at the scale being built today are genuinely without regulatory precedent. The rules being written now will be tested against circumstances that did not exist when they were drafted.
Meanwhile, the physical requirements keep growing. Training and inference at frontier scale is an electricity problem before it is anything else. The regulatory debate around AI has mostly focused on models, outputs, and capabilities. The geopolitical dimensions get some attention. But the binding constraint on AI is not the algorithm. It's the watts. FERC's June 18 orders represent the federal government catching up to a reality that data center developers have understood for years.
What happens next matters more than what happened last week. The 60-day deadline for grid operators runs into mid-August. After that, FERC reviews responses and potentially sets national standards for how AI infrastructure connects to the grid for the next decade. Those standards will shape where data centers get built, which regions become AI infrastructure hubs, and which communities get the tax revenue and the electricity bills that come with hosting the computation.
The models running on those data centers generate the headlines. The grid changes making them possible are generating policy. One of those things is moving much faster than most people realize, and with considerably less scrutiny.
The About.chat newsletter covers AI's real-world impact, not just the model releases. Subscribe here.
