Related Publications
Related Services
Originally published in Public Utilities Fortnightly
The rapid expansion of digitalization and artificial intelligence (AI) is redrawing the map of the electric grid. Hyperscale data centers – deployed by companies like Amazon, Google, Microsoft, and Meta – are multiplying across the country, demanding unprecedented volumes of reliable electricity. By 2030, they are projected to consume nearly twelve percent of total U.S. electricity. These aren’t just corporate facilities. They are the computational engines behind AI training, national security infrastructure, and economic competitiveness.
Against this backdrop, global pressure is mounting. China has declared its ambition to become the world’s AI superpower by 2030, backing that claim with aggressive infrastructure investment. In response, the Trump 47 administration has made AI dominance a strategic priority. Power – both literal and geopolitical – is at the center of this contest.
But here lies the problem: data centers are fast. The grid is not. Hyperscalers can construct new facilities in eighteen to twenty-four months. Transmission infrastructure takes three to ten years – sometimes more. That timeline mismatch has become a defining challenge of energy deployment today.
And so, a new model has emerged: co-location. Rather than wait in the transmission queue or depend on grid upgrades, large loads are siting directly next to generation in order to bypass the delays associated with grid expansion.
They’re taking power at the source – avoiding delay, reducing congestion, and asserting greater control over their supply. It’s flexible, fast, and infrastructure-light. But it’s also raising hard questions about how the grid is planned, who pays for shared services, and where regulatory boundaries lie.
Some customers say they’ve checked out of the grid – declaring themselves self-supplied. But the system shows they haven’t left. Others are barreling forward at hyper speed, demanding everything all the time from a grid built for stability, not sprinting. The Federal Energy Regulatory Commission is now tasked with sorting out what these changes mean for law, policy, and fairness.
This article explores some of those questions, beginning with a metaphor – and a warning.
“You can check out any time you like,” the Eagles sang, “but you can never leave.” That line has become a touchstone in the debate over co-located load. Customers who claim to be off the grid often remain wired in – quietly drawing backup power, relying on grid reliability, or selling their surplus. They say they’ve departed. But the grid still sees them there.
The FERC has undertaken a leading role, as it should. Its recent Commissioner-led technical conference and a subsequent show cause order involving the regional grid operator in the Mid-Atlantic region, the PJM Interconnection, have the debate raging.
The Commission is not just evaluating commercial arrangements. It is probing the physical, operational, and legal meaning of off-grid and, importantly to both industry players and electricity consumers, it is evaluating the optimal approaches to cost responsibilities and the impact of co-location on electricity supply and grid reliability.
In its February 2025 order, FERC consolidated the technical conference record and a pending complaint to launch a formal Section 206 investigation into whether PJM’s tariff is unjust, unreasonable, or unduly discriminatory for failing to address large co-located loads.
The Commission identified significant deficiencies in PJM’s tariffs, including a lack of clear rules on transmission service, ancillary charges, interconnection requirements, and capacity market obligations for co-located configurations. It raised concerns about cost shifting, inconsistent treatment of similarly situated projects, and reliability risks stemming from PJM’s current practices.
The order underscores FERC’s jurisdictional oversight of wholesale services and high-voltage grid operations and access, while inviting broader comment on how isolated load arrangements interact with Commission-regulated systems.
With this proceeding, FERC has signaled that the regulatory treatment of co-location must be both coherent and durable – anchored in cost causation, operational visibility, and just and reasonable outcomes.
At its core, co-location refers to siting a large load – often a data center – adjacent to a power plant. This arrangement can take several forms:
Fully Integrated Load: Takes network transmission service and pays standard charges.
Fully Isolated Load: Uses only on-site generation, with no physical or contractual tie to the grid.
Partially Grid-Connected Load: Self-supplied most of the time but retains backup access or shares interconnection facilities.
It is the third category that raises the most questions. These are the checked out but still connected arrangements. And they lie at the center of FERC’s current docket.
When a facility holds itself out as off-grid but maintains a latent connection – however narrow – it imposes planning, reliability, and fairness consequences. The Commission’s task is to determine when such facilities cross the legal and functional line into grid participation, even if their contracts say otherwise.
Speed is now the dominant force in power deployment. Hyperscale facilities can no longer afford to wait for multi-year transmission projects. AI workloads require vast computing clusters, and downtime or delay risks forfeiting market share, technical advantages, and geopolitical position.
That urgency has shifted the load-generation relationship. Historically, power plants were sited far from population centers. High-voltage transmission brought electricity to load, with long-range planning guiding the buildout. Today, large loads are reversing that flow – placing themselves next to generation instead of waiting for generation to come to them.
This shift is a paradigm change. And it collides directly with FERC’s Order No. 1920, which requires proactive, long-term regional transmission planning based on anticipated load and policy shifts. But co-location injects uncertainty into that process.
If planners assume traditional geography – remote generation serving distant urban load – they may overbuild transmission that isn’t needed. Lines could be placed where no one plugs in. Billions of dollars could be spent on a bridge to nowhere.
Co-location doesn’t eliminate the need for transmission. But it demands that planners stay attuned to the new tempo and topology of load growth. Failing to do so risks locking in infrastructure for a past that’s already been left behind. Interestingly, two of the most well-known hyperscalers, Amazon and Google, have staked out contrasting approaches to co-located load.
Amazon emphasizes physical separation. If a load does not rely on the grid, it should not pay grid-related charges. Its focus is on preserving speed, independence, and cost control – arguing that fully isolated co-located load falls outside FERC’s jurisdiction and should be treated accordingly.
Google, by contrast, sees co-location as a means to accelerate clean energy deployment within the grid. It supports front-of-the-meter configurations and insists that co-located load participate fully in market structures, contribute to planning, and bear appropriate costs.
In short, Amazon advocates regulatory restraint; Google calls for proactive oversight. One seeks freedom from the grid. Amazon focuses on accountability within it. Isolating data centers from the grid, though has major implications both for the grid and for electricity consumers.
But can large data centers really check out of Hotel California and is it sound policy for FERC to suspend disbelief and assume that they can? If time is of the essence, the straightest path is consistent with Amazon’s position.
Reliability is a system-wide obligation. If the grid must be ready to serve co-located load in an emergency – or protect against unintended back-feed – then the load has affected the system. Transmission owners have warned that such arrangements can disrupt voltage stability, planning models, and reserve requirements. The PJM Market Monitor has called it a “real and growing” issue.
On the cost side, the argument is simple: if a customer benefits from the grid, even occasionally, they should help pay for it. Opponents respond: “We’ve built our own infrastructure. We’re taking the risk. If we never schedule transmission, we shouldn’t be charged for it.”
But FERC’s approach has been consistent. Functional reality trumps formal assertion. The grid doesn’t care what the contract says. It cares whether the system must be built and maintained to accommodate the load.
Some argue that these matters belong to the states – that backup supply and localized service arrangements fall outside FERC’s reach. But when interconnection facilities tie co-located load to wholesale markets, or when grid reliability is implicated, FERC’s jurisdiction is triggered.
In Docket No. EL25-49 and other proceedings, the Commission has begun drawing those lines. Its deficiency letter in ER24-2888 makes clear that access to the grid – even for reliability-only events – may be sufficient to require participation in the transmission cost structure.
This is not about expanding federal reach. It’s about ensuring that shared infrastructure isn’t subsidized by some while used quietly by others.
For transmission-owning public utilities, the obligations remain constant:
Apply filed tariffs consistently and without undue discrimination;
Study all interconnection requests, even those claiming isolation;
Coordinate with state commissions but act where federal reliability or cost causation arises.
Engage constructively in FERC processes that shape the regulatory framework for these new arrangements; and
Public utilities are not simply gatekeepers. They are system stewards. Their actions in the coming months will help determine whether co-location becomes a tool of innovation or a source of inequity.
We are in a moment of divergence. Load is moving faster than infrastructure. Technology is moving faster than law. And declarations of independence are outpacing the physical realities of shared systems.
FERC Commissioner Willie Phillips has framed the issue as one of national urgency. Chair Mark Christie has emphasized fairness and open access. Commissioner Rosner has urged procedural clarity and openness to new models.
The Commission doesn’t need to choose between innovation and reliability. But it must ensure that one doesn’t come at the expense of the other. Co-location may be the fastest lane in the system. But even fast lanes need rules, guardrails, and cost responsibility.