For the past two years, the hardest constraint on artificial intelligence has not been chips, talent, or capital. It has been a spot on the power grid.
A company could raise billions, buy a mountain of Nvidia silicon, and sign a lease on a data center, then wait years for permission to actually plug the building into the electrical system. In the largest US grid, the line to connect grew so long that it descended into open dysfunction, with major utilities threatening to walk away. The queue had become the bottleneck behind every AI ambition in the country.
On Thursday, June 18, federal regulators decided to blow up the queue.
The Federal Energy Regulatory Commission approved a series of orders directing six regional grid operators to fast-track interconnection for data centers and other large electricity users. The target is blunt: handle a request for power within 90 days, a process that today can stretch for years. The commissioners approved it unanimously.
"This is the biggest priority our country is facing at the moment," said FERC Chair Laura Swett. "We are taking historic action to push our country's electric markets and economy into the future."
The Order Rewrites the Rules of Getting Power
The directives do more than wave data centers to the front of the line. They reset the obligations on both sides of the connection.
Under the orders, the six regional grid operators must demonstrate that large loads can connect "in a timely and orderly manner." Data centers, in exchange for the fast lane, take on the bill. They are responsible for the cost of their own interconnection and for the grid upgrades needed to deliver power at the scale they demand. The fast lane is real, and it is not free.
FERC also set tight clocks running. Grid operators have 30 days to report how much spare generating capacity they actually have, if any. They have 60 days to defend or revise the electricity rates in their regions, with particular scrutiny on tariffs that fail to account for these enormous new customers. Utilities will have to report data-center-related costs to the federal agency directly, a transparency measure aimed at the soaring bills landing on ordinary ratepayers.
Two smaller provisions matter for anyone watching the infrastructure layer. FERC told operators to consider "alternative transmission technologies," an opening for grid-tech startups working on hardware like solid-state transformers and superconducting transmission lines. It also pushed operators to be more accommodating of behind-the-meter power, the on-site generation that AI developers have been building out of desperation when the grid would not connect them.
| What FERC ordered | The deadline or rule | Who pays |
|---|---|---|
| Connect large loads to the grid | Within 90 days | Data center funds its interconnection |
| Report spare generating capacity | 30 days | Grid operator |
| Defend or revise electricity tariffs | 60 days | Grid operator |
| Cover needed grid upgrades | Ongoing | Data center / hyperscaler |
| Manage demand in peak stress | Ongoing | Data center may curtail or bring own power |
The Grid Was Drowning Before AI Showed Up
To understand why a power-market regulator just became a central player in the AI story, look at what the demand curve did to a system built for the opposite.
Grid operators spent two decades getting comfortable with near-zero demand growth. Then AI arrived. Electricity demand from data centers is expected to nearly triple through 2035. A single large data center can add the power draw of a small city within a few years, and the grid was never designed to absorb new loads that big, that fast.
The strain is already visible in prices. Wholesale electricity rates have climbed as much as 267% compared with five years ago, according to Bloomberg. The pain has concentrated in places like PJM Interconnection, the largest US grid operator, which covers 13 states and 67 million customers and has drawn heavy criticism for slow connections and surging bills.
The connection backlog has a brutal logic to it. New power plants need to connect to the grid too, and by the end of 2023, grid-connection requests for power plants exceeded the entire capacity of the existing fleet. The line to get onto the grid was longer than the grid itself could serve. AI data centers were trying to join the back of a queue that was already mathematically impossible.
That is the wall FERC just tried to knock down. The agency was prodded into action by Energy Secretary Chris Wright, who warned last October that connection delays threatened US competitiveness in AI. For practitioners watching compute capacity and cost, this is the policy that sits underneath the hyperscaler spending that is now outrunning cash flow and the data-center buildouts companies like Oracle are betting their balance sheets on.
The Order Skips the Part That Actually Matters
Here is the catch that determines whether any of this works: FERC sped up the line without adding anything to supply.
The orders give data centers a fast lane to connect. They do nothing to fix the shortage of generating capacity that made the line so long in the first place. You can move to the front of the queue, but if there is no spare power where you are standing, a faster connection just gets you to the bottleneck sooner.
That is why the fast lane comes with strings. Under the plans, hyperscalers could be required to bring their own power or curtail demand during times of high stress on the system. The implicit message to AI companies is that the grid will connect you quickly, but it may also ask you to switch off when the system is straining, or to generate your own electricity when it cannot spare any.
For teams running large training and inference fleets, that is a meaningful operational shift. Guaranteed, always-on grid power at a fixed price is quietly becoming a thing of the past for the largest loads. Flexibility, on-site generation, and the willingness to throttle during peak demand are turning into the price of admission.
The Politics Are About Your Power Bill, Not Your GPUs
The fast lane did not arrive in a friendly climate. Public sentiment toward AI and data centers has soured considerably, and the data-center buildout has become a hot-button issue heading into the November midterms, precisely because of those rising consumer bills.
FERC tried to thread that needle, speeding connections while attempting to shield households from the cost. It stopped short of a nationwide rule, leaving states to ensure that costs are allocated fairly. "This order may be remembered less for its technical reforms and more for recognizing that large-load interconnection is now a core political, planning, and economic issue," said Robert Montejo, a partner at the law firm Duane Morris. "We're out of the wild west era of data center development."
The energy backdrop makes the supply gap sharper. The same week as the FERC orders, the Trump administration moved to cancel offshore wind capacity that would have helped feed exactly this demand.
On June 17, the administration agreed to pay $765 million to developer Invenergy to scrap offshore wind leases near California, Maine, and New York. One canceled project alone would have generated up to 2.4 gigawatts at peak, enough to power roughly 1.8 million homes. The administration has now spent about 2.6 billion dollars in total to unwind offshore wind, while simultaneously ordering the grid to make room for AI loads that need every megawatt it can find.
Not everyone sees a problem. New York Independent System Operator spokesman Kevin Lanahan said FERC's approach gives operators the flexibility to handle region-specific conditions rather than forcing a one-size-fits-all rule onto very different grids. The order is a framework, and a framework can be filled in sensibly or badly depending on who is doing the filling.
The Bottom Line
FERC just did the one thing it had the clear authority to do. It cleared a path. Data centers that were staring down a multi-year wait can now, in theory, connect to the grid in a single quarter, and the AI buildout gets a faster on-ramp than it has had at any point in this cycle.
What FERC could not do is conjure electricity. The order accelerates access to a supply that is still short, still expensive, and still arriving more slowly than the demand chasing it. Speeding up the line does not add a single megawatt to the other end of it.
So the real test starts now. Over the next 60 days, six grid operators have to open their books and show how much spare power they actually have. If the answer is "plenty," the AI industry just got a green light. If the answer is "almost none," then the United States has built the fastest possible on-ramp to a road that still dead-ends at a power plant nobody has finished building. The grid will tell us which it is, and it will not be persuaded by a press release.
Sources
- AI data centers just got a government-mandated fast lane to the grid — TechCrunch (Jun 18, 2026)
- US Acts to Speed Up Power Grid Hook-Ups for AI Data Centers — Insurance Journal / Bloomberg (Jun 22, 2026)
- FERC launches aggressive, targeted action to speed large load integration — FERC (Jun 18, 2026)
- US Regulator Makes 'Interventionist' Pivot to Speed Data Centers — Claims Journal (Jun 22, 2026)
- Trump administration to pay $765 mln to scrap four more offshore wind leases — Reuters (Jun 17, 2026)
- The biggest U.S. power grid is under strain from AI and no one is happy — TechCrunch (May 8, 2026)