One Utility Quoted Google a 12-Year Wait Just to Study a Power Hookup. The Data Center Itself Takes Two.
AI's newest bottleneck isn't a chip or a mineral — it's a 20-year-old bureaucratic waiting line. For decades, the U.S. has studied requests to plug new power onto the grid largely in the order they arrive, one at a time, and each study assumes every project ahead of it gets built. When one drops out — which happens to most of them — everyone behind it can be sent back to square one. Regulators are now rewriting the line itself.
The queue that resets itself
Simplified illustrative model, not real project data: 12 hypothetical interconnection requests (P1–P12) in line, where P4 withdraws partway through. Pick Serial (the pre-2023 one-at-a-time process) or Cluster (FERC Order No. 2023's batch process), then press Run to see how far that one withdrawal reaches back into the line. "Study rounds" are an abstract unit for pacing, not real years.
The plain version
Picture the DMV, except instead of renewing a license, you're asking permission to plug a giant new power source into the electrical grid — and the line has been forming for twenty years.
That's roughly how America's "interconnection queue" has worked since the 1990s. Anyone wanting to add new power to the grid — a wind farm, a gas plant, or lately an AI data center that can draw as much electricity as a small city — gets studied by the local grid operator to make sure adding them won't overload the wires. For decades this happened one project at a time, in the order applications arrived, and each study assumed every project ahead of it would actually get built.
The problem: most don't. Historically, a large majority of projects that enter the queue eventually walk away — costs rise, financing falls through, plans change. But when an earlier project quits, the study behind it, which assumed that project's grid upgrades would exist, is now wrong. So everyone studied after it has to be restudied. It's a single-file line where someone leaving near the front sends everyone behind them back to the waiting room.
Google says this is now its biggest headache: one utility quoted the company 12 years just to study whether a data center could connect — before a single wire gets built. That's longer than the data center itself takes to design and construct.
In 2023, federal regulators ordered grid operators to stop studying requests one by one and start studying them in batches instead, so one withdrawal only delays its own batch, not everyone behind it. PJM, the mid-Atlantic grid operator, says its first batch under the new rules is moving 800-plus projects through together, targeting one to two years instead of the old queue's open-ended wait.
The expert version
For most of its history, U.S. generator interconnection ran as a serial, first-come-first-served process: each request was studied individually, in queue position, against a system-impact model that assumed every higher-queued project ahead of it would be built and would bear its allocated share of any needed network upgrades. That assumption is the process's structural flaw. Interconnection is a long-duration commitment — median time from application to commercial operation is now approaching five years nationally, and PJM data show AI-era generation and load projects reaching commercial operation in 2025 averaged upward of seven years: roughly three-plus years to a signed interconnection service agreement, then another four years to energization. Google's global head of sustainability and climate policy, Marsden Hanna, said in 2026 that some utilities have quoted the company interconnection study timelines alone — before construction — of up to 12 years, longer than a data center's own design-to-energization schedule.
Over that time horizon, withdrawal is the norm, not the exception: Lawrence Berkeley National Laboratory's "Queued Up 2026" report, tracking 38,201 project records across more than 50 transmission providers in the seven major ISO/RTO regions, documents historical completion rates for queued projects running well under 20%. Under the old serial rule, each withdrawal invalidates the shared cost-allocation and system-impact assumptions baked into every study behind it in line, forcing a re-study of the remaining queue — a cascade that compounds as queue depth grows, since national active-queue capacity is now roughly double the size of the country's entire existing generating fleet.
FERC Order No. 2023 (finalized 2023; implementation ongoing through 2026) mandates replacing serial studies with an annual cluster process: a fixed intake window, then a single system-impact study covering the whole batch simultaneously, with cost allocation fixed per cluster rather than recalculated on every individual exit, plus stricter financial-readiness deposits meant to discourage speculative queue entries. PJM's first cluster ("Cycle 1") is studying more than 800 projects totaling roughly 220 GW together and targets issuing interconnection agreements within one to two years; PJM states it has eliminated its backlog of unprocessed new requests. The bottleneck now shifting into view: transformer, switchgear and transmission-line lead times once a project clears study — a physical supply-chain constraint that study reform alone doesn't fix.
Why it matters for tech + supply chain: chip supply is no longer the pacing item for AI buildouts — a paperwork queue designed in the 1990s is, and fixing it matters as much as fixing chip supply did.
Why it matters for tech + supply chain: AI capacity growth is increasingly gated by transmission-study throughput rather than chip supply, and FERC's cluster-study reform is the first structural attempt to unwind decades of cascading-restudy delay — worth tracking alongside whether physical build-out (transformers, lines, substations) becomes the next chokepoint once study reform removes the first one.