Skip to content
Rung 10 Datacenters & Digital InfrastructureSwitch rung

Grid Connection Queue

Active Frontier
grid-interconnect-queuedatacenter-power-demandbehind-the-meter

A new data centre cannot serve a single token until the grid connects it, and there is a line for that. There are really two lines, and the rung's sources measure different ones:

  • The generation queue — new power plants waiting to connect so they can sell power. Carbon Direct measures this one in PJM and ERCOT.
  • The load queue — new users of power, data centres above all, waiting for a connection big enough to run. The IEA measures this one.

A data centre can be stuck in the second line even when the first is full of plants, because the wires between them take longest of all to build.

Key Claims

  • The generation queue holds enough on paper; the wait is the problem. PJM and ERCOT together hold over 300 GW across about 1,500 projects, enough "on paper" for projected data-centre demand through 2030, but waits in data-centre load zones run three to four years. Gas clears faster than renewables and storage, which make up 77% and 87% of the two queues. Evidence: moderate (analysis) (Carbon Direct)
  • The load queue is global. At least 150 GW of data-centre projects are in grid queues at advanced stages, and one fifth of the world's data-centre build-out is at risk of delay from grid congestion. Worldwide, more than 2,500 GW of projects of all kinds wait in queues. Evidence: moderate (IEA forecast) (IEA Electricity 2026)
  • Most of the load queue is not real. From US utility data, only around 20% of data-centre connection requests turn into load in the short to medium term. Australia: 8 GW likely out of 44 GW requested (a study paid for by AWS). Brazil: 6 GW under review or late-stage out of more than 26 GW requested. The IEA puts the gap down to duplicate and "phantom" requests for projects not expected to be built. Evidence: moderate (IEA, US data; weak for Australia) (IEA Electricity 2026)
  • The wires are the slow part. New grid lines take 5 to 15 years; a data centre takes 1 to 3. Grid component prices have nearly doubled in five years. Evidence: moderate (IEA) (IEA Electricity 2026)
  • Utilities now make builders prove they are serious, and pay for it. Dominion Energy Virginia proposed a 14-year commitment to pay 60% of generation and 85% of network costs whether the power is used or not; ComEd charges a $1 million refundable deposit per request of 50 MW or more; PG&E cuts the wait from 18–22 months to about 2–5 for builders who pay for transmission upfront; Ireland requires new data centres to bring their own dispatchable generation or storage. Evidence: moderate (IEA, citing named utilities and regulators) (IEA Electricity 2026)
  • Behind-the-meter power is the way around the line. Generating on site can "simplify or avoid" the generator queue, at the cost of siting and balancing; the IEA calls it an emerging trend where connections are the bottleneck, while noting most such sites stay grid-connected anyway. Evidence: moderate (Power 2026; IEA Electricity 2026)
  • Blocked projects are already counted in billions. 36 or more projects worth $162B were blocked or delayed going into 2026. Evidence: weak (trade press) (Hyperscalers in 2026)

Why it matters on the token-to-task path

A queued megawatt is capex spent, or committed, that serves nothing. Every month of wait adds idle capital to the cost of each token the site eventually serves. The 20% conversion rate cuts the other way: forecasts that count requested gigawatts as future supply overstate it about five times, so the case that serving capacity will soon outrun demand, and push token prices down on its own, is weaker than the pipeline numbers suggest.

Benchmarks & Data

  • PJM + ERCOT generation queue: >300 GW, ~1,500 projects; 3–4-year waits in data-centre load zones (May 2026)
  • Global data-centre load queue, advanced stage: ≥150 GW; ~1/5 of global build at risk of grid delay (IEA, early 2026)
  • US data-centre connection requests that materialise: ~20%
  • Australia 8 of 44 GW; Brazil 6 of 26+ GW (Nov 2025)
  • Grid build 5–15 years vs data centre 1–3 years
  • Dominion proposal: 14 years, 60% generation / 85% network costs; ComEd: $1M deposit per ≥50 MW request

Open Questions

  • How fast do the load queues actually clear? Neither source gives a wait time for a data-centre connection, only for generation (3–4 years in PJM/ERCOT).
  • Does the 20% conversion hold outside the US, and is it falling as deposits and commitments like Dominion's filter out speculative requests?
  • Who pays for the long commitments? A 14-year take-or-pay grid charge is a fixed cost that a 2–4-year GPU cannot carry on its own; no source shows how it is priced into leases.

Related Concepts

Changelog

  • 2026-09-24 — Created. Graduated from the frontier page's "still to compile" list (it had been a sub-section of a planned power concept that was never written). Built from the IEA's Electricity 2026 (new), Carbon Direct and Data Center Knowledge (compiled earlier), and the Power 2026 primer (ingested 2026-08-06, compiled here for the first time).

Related concepts