The Government's Big Move to Solve AI's Power Crunch Speeds Up the Paperwork. It Doesn't Add a Single Watt.
FERC unanimously ordered all six US grid operators to fast-track how they connect large AI data centers — a mandate reaching two-thirds of US electricity demand and 200 million people. It sounds like the cavalry. The fine print: it fast-tracks the connection *process*, not the power *supply*. FERC can compress a multi-year interconnection queue and reassign who pays for the wires, but it cannot add one watt of generation to a grid that's short. That's why the hyperscalers already routed around it — building their own gas now and betting on nuclear by 2030. Washington sped up the line to a well that's running low.
The single most forceful thing the US government has done about AI's electricity problem is a FERC order, issued unanimously in June, telling all six regional grid operators to fast-track how they connect large data centers — a mandate covering two-thirds of the country's electricity demand and about 200 million people. It sounds like the cavalry arriving. Read the fine print and it's stranger than that: FERC fast-tracked the connection process, not the power supply. It can compress a multi-year interconnection queue and it can reassign who pays for the upgrades, but it cannot add a single watt of generation to a grid that doesn't have enough. The flagship federal move to solve AI's power crunch fixes the paperwork, not the power.
That's the whole story, and it's a revealing one. This is what it looks like when the binding constraint on AI stops being chips and becomes electricity: the fastest lever available to Washington can only speed the line, not lengthen the supply. FERC did the doable thing — order the queue to move faster — precisely because the hard thing, building generation, is outside its authority and years behind AI's timeline. The speed of the response is itself a tell about which problem is actually solvable.
The mechanics are real, to be clear. FERC issued Section 206 show-cause orders to all six RTOs and ISOs — PJM, MISO, SPP, CAISO, ISO-NE, NYISO — giving them 60 days to justify or reform how they handle large-load (over 50-megawatt) interconnection, explicitly to speed access while "protecting other customers from costs caused by large loads." The bottleneck it targets is genuine: data centers are heading toward roughly 12% of US electricity by 2028 and already drive about half of the country's electricity-demand growth, and the multi-year queue to plug in is the wall everyone hits. Shortening that queue matters.
But look at what FERC can and cannot touch. It can shorten the wait to connect. It can shift who pays for the substations and transmission. It cannot make more electricity exist. The queue is a symptom; the disease is that the US isn't building generation fast enough, and the two things that could — new gas and new nuclear — either arrive privately and off-grid (dedicated gas plants like the one Chevron is building for Microsoft, first power 2028) or arrive around 2030 (the roughly 9.8 gigawatts of nuclear the hyperscalers have committed to, stuck behind a six-year permitting wall). Fast-tracking interconnection into a grid that's still short on supply is widening the on-ramp to a highway that's already full.
Which is exactly why the companies with the most at stake didn't wait for FERC. The hyperscalers signed a White House pledge to keep their power costs off ordinary ratepayers, and — more tellingly — they started building their own generation, because the federal government can order the line to move but can't conjure the electricity to fill it. When the buyers most desperate for power respond to a federal fix by routing around the grid entirely, that tells you what the fix is worth. The "who pays" question FERC is trying to referee — speed access, but protect other customers from the cost — is the same fight the ratepayer pledge is trying to settle, and neither one adds a watt.
So watch two things. The near one: the 60-day clock, and whether the grid operators actually reform their tariffs or simply file paperwork justifying the status quo — show-cause orders can be answered with a shrug. The real one: not the length of the queue but the size of the generation gap, whether the US builds power fast enough that a shorter queue means anything at all. My read is that FERC did the only thing it could do quickly, and "quickly" is the giveaway — the slow, hard thing, generation, is the actual crisis, and nobody has a fast answer to it. It's the mirror image of China's $295-billion state-directed grid build: there, the government is building the power; here, the government is speeding the paperwork and leaving the power to Chevron and to nuclear's 2030 timeline. What would change my mind is Washington pairing the interconnection fast-track with something that actually adds baseload on AI's schedule. Until then, the government sped up the line to a well that's running low, and called it action.