FERC Gave the Grid 60 Days to Get Ready for AI. The Deadline Just Quietly Passed.
Before a data center can draw a single watt, its grid operator has to certify the local wires can take it — a process that now runs slower than building the data center itself. In June, federal regulators gave the operators a strict clock to fix it. That clock just ran out, and the fixes are still trickling in.
The widening gap
Drag the year slider (or press Play) to watch two paces diverge: how much new AI data-center demand shows up each year, versus how much new grid supply actually gets connected. Toggle "if connections matched demand" to see what closing the gap would take. Simplified national model using industry-reported order-of-magnitude rates (demand +5-7 GW/yr, supply +2-3 GW/yr) — illustrative, not an official FERC or NERC projection, and it excludes the multi-year backlog that already existed before year zero.
Scrub the years, or race the two paces:
By year 10, data centers have requested about 60 GW of new grid capacity, while only about 25 GW of new supply has actually been connected — a backlog of 35 GW on top of whatever the queue already held.
The plain version
Building an AI data center is the easy part. The hard part is getting permission to draw power from the grid — and that permission process is its own long, opaque bottleneck, separate from and often slower than actually building anything.
Here's the mechanism: before a data center, or any huge new electricity user, can flip the switch, the regional grid operator has to study whether the local wires, substations, and transformers can handle that much extra load without breaking something for everyone else nearby. That study, plus whatever fixes it identifies, plus a signed agreement, can take years — often five, and up to seven in Northern Virginia, the world's biggest data-center market. A related but separate queue governs new power plants trying to connect their supply to that same grid, and it's even more clogged: roughly 8,200 projects, representing more new electricity than the entire U.S. grid currently generates, are sitting in line.
In June, the U.S. energy regulator, FERC, did something unusually blunt: it told the six organizations that run most of the country's power grid — covering the Mid-Atlantic, the Midwest, California, Texas's neighbors on the Plains, New England, and New York — that their rules for connecting giant new customers like data centers looked outdated or unfair, and gave them a strict 30-and-60-day clock to prove otherwise or fix them. That kind of aggressive, all-at-once order is rare for FERC.
Both deadlines have now quietly passed. Some grid operators are still filing their fixes months late — one isn't due until mid-November. It's a reminder that even when Washington moves fast by its own standards, "fast" is still slower than the AI buildout it's trying to keep pace with.
The expert version
On June 18, 2026, the Federal Energy Regulatory Commission issued show-cause orders under Section 206 of the Federal Power Act to all six FERC-jurisdictional RTOs/ISOs — PJM, MISO, SPP, CAISO, ISO-NE, and NYISO — and their transmission owners. Each was directed to do two things: file an informational report within 30 days on how it will secure sufficient generation capacity for existing and new large loads, and within 60 days either justify that its current open-access transmission tariff (OATT) provisions for interconnecting large loads — data centers, manufacturing plants, and similar demand — remain just and reasonable, or propose reforms. Both windows, roughly July 18 and August 17, had elapsed by early September. The action follows an October 2025 directive from Energy Secretary Chris Wright asking FERC to pursue reforms for "timely and orderly" large-load interconnection, and sits alongside a separate December 2025 FERC ruling that found PJM's tariff for co-located generation-and-load arrangements unjust and unreasonable — which PJM addressed through its own compliance filings in January and April 2026, exempting it from the June order's large-load provisions.
Compliance is arriving on a staggered timeline that outruns the original clock: MISO, for instance, expected to file its Large Load Interconnection Reliability Requirements Tariff by the end of August 2026, but committed to a full response to the show-cause order only by November 16, 2026.
This large-load bottleneck compounds a well-documented generation-interconnection queue. Per Lawrence Berkeley National Laboratory's tracking, roughly 8,200 active projects representing about 1,312 GW of generation and 749 GW of storage were queued for grid connection as of end-2025, with average wait times near five years nationally, and up to seven in Northern Virginia. Structurally, data-center demand is being added to the grid at an estimated 5-7 GW per year, while new generation supply is energized at only about 2-3 GW per year — a gap, not a fixed backlog, meaning the shortfall compounds annually regardless of how fast interconnection paperwork itself gets processed. NERC's January 2026 Long-Term Reliability Assessment raised its 10-year summer peak demand growth forecast to 224 GW, 69% above the prior year's estimate and a 24% increase over 2025 peak demand — the steepest compound annual growth rate NERC has recorded since it began tracking in 1995 — attributing most of the increase to new data-center load, and flagging 13 of 23 assessed regions at elevated or high resource-adequacy risk within five years.