The Federal Energy Regulatory Commission is forcing regional grid operators to justify how they handle massive new electricity loads from AI data centers, while companies like Microsoft and Amazon Web Services have already moved to lock down dedicated nuclear power plants. These parallel actions reflect a collision between the speed at which tech companies are scaling AI infrastructure and the pace at which the U.S. power grid can absorb that demand. The result is a regulatory scramble that could reshape how electricity is generated, sold, and delivered across the country.
FERC’s show-cause orders signal a grid under pressure
The strain is no longer theoretical. FERC has issued Section 206 show-cause orders to regional transmission organizations and independent system operators, demanding they justify or reform the tariff rules governing how large loads connect to the grid. The agency’s own language identifies “data centers and other large energy users” as a key source of that pressure. These orders carry legal weight: each RTO must demonstrate that its existing interconnection process is just and reasonable, or propose changes.
The practical effect is that grid operators in different regions now face different levels of exposure. PJM Interconnection, which manages the grid across 13 states and the District of Columbia, sits at the center of the conflict because several of the largest data center co-location proposals involve plants within its territory. FERC separately launched a formal review of PJM’s tariff rules for co-location of data centers running AI at generating facilities, asking whether current rules adequately protect reliability and fairly allocate costs. That review targets a specific arrangement: data centers that physically sit next to, or directly connect to, a power plant and draw electricity without routing it through the broader transmission system.
This matters for anyone who pays an electric bill. When a large data center bypasses the grid to buy power directly from a generator, the remaining customers may bear a larger share of the fixed costs that keep the transmission network running. FERC’s inquiry is essentially asking whether these behind-the-meter deals shift costs onto households and businesses that had no say in the arrangement. If the agency ultimately finds that current tariffs allow unreasonable cost shifts, it could order PJM and other operators to redesign how they treat these private connections or require new charges to cover shared infrastructure.
FERC’s actions also highlight a timing problem. Large data centers can move from concept to operation in a few years, while major transmission upgrades often take a decade or more. The show-cause orders are an attempt to compress that gap by forcing RTOs to revisit how quickly they process new large-load requests and how they ensure that reliability standards keep pace with concentrated, always-on demand from AI and other digital infrastructure.
Microsoft, AWS, and the race for dedicated nuclear capacity
Tech companies are not waiting for regulators to sort out the rules. The U.S. Energy Information Administration has documented two high-profile nuclear deals that illustrate how far the industry has moved. Microsoft signed a power purchase agreement with Constellation Energy to restart Three Mile Island Unit 1, the reactor adjacent to the site of the 1979 partial meltdown but a separate and previously operational unit. Amazon Web Services secured an arrangement with Talen Energy at the Susquehanna facility in Pennsylvania. Both deals reflect a calculation that only large, emissions-free baseload generators can reliably feed the constant, heavy electricity demand that AI training and inference require.
These are not small pilot projects. Nuclear plants produce hundreds of megawatts around the clock, and dedicating that output to a single corporate buyer represents a fundamentally different model from the traditional utility structure where generators sell into a shared wholesale market. The EIA’s analysis frames these transactions as evidence of data center operators actively seeking alternatives to conventional grid supply, driven by both the scale of their power needs and corporate clean-energy commitments.
The tension between these private deals and public grid planning is already producing legal disputes. Constellation Energy Generation filed a formal complaint against PJM Interconnection over how the grid operator treats co-located loads, recorded under FERC accession number 20241122-5285. The complaint signals that generators and data center developers believe current PJM rules are blocking or penalizing arrangements they consider commercially viable and technically sound. At the same time, PJM and other operators are responsible for maintaining system-wide reliability, and they face pressure not to allow bespoke deals that could undermine broader planning assumptions.
For companies like Microsoft and AWS, dedicated nuclear capacity offers two advantages: predictable, long-term pricing and a clear emissions profile that can be counted toward internal climate targets. For regulators, the same deals raise questions about whether critical, zero-carbon resources are being walled off from the wider grid just as many states are trying to decarbonize electricity for all consumers, not only for a handful of hyperscale buyers.
Unresolved questions about grid equity and regional divergence
Several critical questions remain open. The full text and supporting exhibits from Constellation’s complaint against PJM have not been publicly released beyond the docket cover sheet. Without those details, the specific load forecasts and cost-allocation arguments driving the dispute are unknown. FERC’s show-cause orders also do not provide a public breakdown of how much of the new large-load demand is specifically tied to AI workloads versus other industrial users such as manufacturing or cryptocurrency mining.
The underlying contracts between Microsoft and Constellation, and between AWS and Talen, have not been disclosed in full. The EIA’s reporting confirms the existence of these power purchase agreements but does not include granular data on contracted capacity, pricing, or delivery timelines. That gap makes it difficult to assess how much nuclear generation is being effectively removed from the shared grid and what the downstream price effects will be for other ratepayers. Without visibility into the volume and duration of these deals, regulators and consumer advocates are left to infer their impact from limited public filings and aggregate planning documents.
Regional grid operators have not yet filed their formal responses to FERC’s co-location docket. Those filings will reveal whether RTOs plan to accommodate behind-the-meter nuclear arrangements through new tariff structures or resist them to protect existing cost-sharing frameworks. Some regions may seek to impose special charges on large, directly connected loads to ensure they contribute to transmission and reliability costs. Others may encourage co-location as a way to relieve congestion on stressed parts of the network, particularly if dedicated plants sit near major data center clusters.
The lack of uniform rules raises the prospect of regional divergence. If one RTO adopts flexible tariffs that welcome co-located nuclear-backed data centers while another takes a restrictive stance, developers could gravitate toward the more permissive region. That migration would not only shape where AI infrastructure is built but also influence where new generation and transmission investments occur, potentially leaving slower-moving regions with older, more expensive grids and fewer large customers to share the costs.
Equity concerns run through each of these decisions. Households and small businesses cannot negotiate bespoke nuclear contracts or site their operations next to power plants. They rely on the shared grid and on regulators to ensure that everyone who benefits from that system pays a fair share. FERC’s current proceedings are, in effect, an early test of whether the regulatory framework can adapt to AI-era demand without allowing the most powerful buyers to opt out of common obligations.
As the dockets progress, the core tension will remain: AI companies want rapid, firm, low-carbon power commitments, while grid planners and regulators are tasked with maintaining reliability and fairness for millions of customers who will never step foot inside a data center. How FERC resolves its show-cause orders and co-location review will help determine whether dedicated nuclear deals become a narrow exception or a defining feature of the next phase of the U.S. power system.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
