The FERC Ruling Watch: One Decision Could Reshape Every AI Nuclear Deal

The single highest-stakes pending regulatory decision in the AI buildout is not at the NRC, not at the EPA, not in any state utility commission. It is at FERC, the Federal Energy Regulatory Commission, in the matter of the Amazon-Talen Susquehanna interconnection service amendment. Whichever way it falls reshapes every direct-feed nuclear deal hyperscalers have signed and will sign for the rest of the decade. This page tracks the watch.

The State Of Play

In March 2024 Amazon Web Services bought Talen Energy's adjacent Cumulus data center campus for $650 million and signed a direct-feed power agreement: an initial 480 MW pull from the Susquehanna nuclear plant, with provisions to scale to 960 MW. The structure is sometimes called co-location or behind-the-meter, depending on whose filing you read. The campus sits inside the substation; the power moves between generator and load without traversing the wider PJM grid.

PJM filed an amended Interconnection Service Agreement (ISA) on Amazon's behalf in mid-2024 to memorialize the scale to 960 MW. In November 2024, FERC rejected the amendment in a split decision. The technical objection: a campus drawing that much power from a generator inside the substation, but not paying full transmission cost-share, was a structural change to how grid costs get allocated. If the structure stands, the rest of the ratepayer base ends up subsidizing infrastructure that does not serve their load.

Talen, AWS, and PJM filed rehearing requests. The proceedings have continued in front of FERC and on parallel state tracks at the Pennsylvania Public Utility Commission. As of May 2026 the matter remains procedurally open. A decision is expected sometime in the 2026 calendar year, though FERC rarely commits to exact dates.

What FERC Has To Decide

The substantive question is straightforward to state and complex to resolve.

When a large industrial load co-locates physically with a generator and pulls power through facilities inside the substation, is it a transmission customer (and therefore owes the full network cost-share) or is it something else (a direct sale, a wholesale arrangement, a behind-the-meter campus)? The Federal Power Act and the existing FERC tariff structure assume the prior decades' pattern: generation feeds the grid, the grid serves load, transmission gets paid for. Co-location at this scale is a structure the existing tariffs were not designed for.

The pure regulatory answer matters less than the precedent. Whatever FERC decides on Susquehanna becomes the template for every subsequent application. Constellation, Vistra, Dominion, Public Service Enterprise Group, and at least four other US nuclear operators have surplus land adjacent to their plants and have publicly or privately explored similar structures. Every one of those waits on this ruling.

The Three Outcomes

Approves. The 480 to 960 MW scale-up goes through with current structural terms. Amazon adds gigawatt of training capacity adjacent to Susquehanna. Constellation, Vistra, and the other nuclear operators with adjacent land start running their own co-location auctions. Land within five miles of an operating reactor effectively gets a new asset class. Microsoft TMI may or may not reopen the structural terms of its own deal to take advantage. The hyperscaler competition for nuclear-adjacent acreage gets aggressive.

Rejects. The bypass structure gets sent back. Amazon Susquehanna renegotiates as a virtual PPA: Talen sells power into the grid, AWS buys the equivalent through the grid, the transmission cost-share gets paid. The economics are still positive for both sides but worse, and the latency benefits of true co-location are lost. Some currently-announced deals (Oracle SMRs, X-energy Amazon, smaller proposed Constellation co-locations) get re-papered. Some get cancelled. The pace of new nuclear-AI deals slows materially through 2027.

Splits or delays. The likeliest actual outcome. FERC writes a conditional approval that requires Amazon to pay some transmission cost-share, but not full. Parties negotiate the cost-share formula for months. The Susquehanna structure goes forward in modified form. Other co-location applications now have a cost-share floor they have to design around. The result is messier than either pure outcome and adds 12 to 24 months of negotiation tail to every subsequent deal.

Who Else Is Watching

Constellation Energy. Owns Three Mile Island, plus Calvert Cliffs, Limerick, Peach Bottom, and others. Microsoft TMI is structured as a virtual PPA, but Constellation has multiple other plants where adjacent co-location would be commercially valuable if the structure clears.

Vistra Energy. Owns Comanche Peak in Texas and has publicly explored co-location proposals. Texas is on the ERCOT grid, not PJM, so the Susquehanna ruling does not directly bind. But ERCOT will take cues.

Dominion Energy. Operates several nuclear plants in Virginia, where Loudoun County data center demand is the densest in the country. Adjacent co-location structures have been floated. Virginia state regulators have additionally weighed in.

Hyperscalers without nuclear deals yet. Meta, Apple, and xAI have all been reported to be exploring nuclear arrangements but have not publicly signed. They wait on the FERC outcome to decide what structure to even pursue. A clean approval brings a wave of announcements in the following six months; a rejection delays them all by a year.

PJM ratepayers and consumer advocates. Filed formal opposition to the original ISA amendment. Their argument is straightforward and not unreasonable: residential and small-business customers should not subsidize billion-dollar hyperscaler infrastructure through avoided transmission costs.

What To Watch

The FERC docket. EL24-149 and related sub-dockets carry the procedural updates. New filings appear on the public docket; rulings post the same day.

The composition of the commission. FERC sits five commissioners but operates often with four during transitions. Voting patterns on infrastructure questions have been unstable in recent commission cycles. A change in commission composition mid-proceeding changes the math.

Parallel state-level proceedings at the Pennsylvania PUC. State actions can constrain the federal answer in ways that show up months before the FERC ruling lands.

Constellation, Vistra, and Dominion earnings calls. Management commentary on nuclear co-location strategy gets specific quarter over quarter; the framing shifts visibly when the underlying regulatory state shifts.

New deal announcements that explicitly do or do not use the bypass structure. Each new deal is a small data point on what the lawyers expect FERC to do.

The Stakes

Hyperscaler capex on US nuclear since the Microsoft TMI announcement is in the tens of billions of dollars across signed and announced deals. The FERC ruling is the load-bearing decision for how those deals close out structurally and whether the next wave behind them happens at all. One regulator, one ruling, one cost-share formula that does not yet exist on paper, deciding whether the AI buildout keeps reopening American nuclear or has to find another way to power the next decade.

We will update this page when the decision lands.

Related: full nuclear-restart analysis at AI Just Reopened American Nuclear. Live tracker of nuclear deals on the AI infrastructure page (Microsoft TMI restart, Amazon Susquehanna campus).