Definition
The administrative and engineering process by which proposed power generation, storage, or large-load projects request connection to the US transmission grid. Each Regional Transmission Organization (RTO) — PJM, ERCOT, MISO, SPP, ISO-NE, NYISO — maintains its own queue. Projects are studied for grid impact, assigned interconnection costs, and scheduled for energization. The queue is public, updated quarterly or annually, and dispositive: a project with PPAs but no queue position is a press release; a project with queue position is real.
Why It Matters for the Newsletter
The interconnection queue is the document class that decides which AI data center projects actually deliver in 2026-2030. Financial press AI capex coverage rarely cites it; the queue is the first place the binding constraint shows up and the last place chip-layer analysts look. Vocabulary central to the AI Buildout Grid Constraint thesis. Provides the analytical edge for the operator-class reader: read the PUC and FERC filings, track the queue depth, ignore the press release.
Evidence & Examples
- Total queue size (2025 edition). ~2,290 GW active queue vs. ~1,322 GW installed capacity — LBNL framing: “nearly twice” installed capacity. Queued Up 2025 Edition — LBNL - 2025-12-15
- Prior edition (2024). ~2,600 GW queued vs. ~1,279 GW installed; the 12% queue decrease is attributed to FERC Order 2023 forcing withdrawal of speculative projects, not reduced demand. US Interconnection Queue Twice Installed Capacity — Latitude Media - 2024-04-11
- Active project count. 10,303 active projects; ~1,400 GW generation + ~890 GW storage.
- Wait time — US average. Median IR-to-COD for 2024 completions: 55 months (4.6 years); LBNL frames as “over 4 years” for 2018–2024 cohort. Projects completed in 2008 waited under 2 years; 2015: 3 years; 2023: 5 years. Queued Up 2025 Edition — LBNL - 2025-12-15
- Wait time — regional. Northern Virginia now at 7 years; California at 9+ years. Google Intersect Power Acquisition — Introl - 2026-01-20. CAISO median ~75–94 months (~6–8 years); PJM median ~68 months (~5.7 years) for 2024 cohort. Queued Up 2025 Edition — LBNL - 2025-12-15
- Completion rates. 19% of projects (13% of capacity) from 2000–2019 cohort reached commercial operation by end of 2024. Solar 14%; battery 11% (2024 edition figures). Queued Up 2025 Edition — LBNL - 2025-12-15
- Withdrawal rate. ~70% of projects ultimately withdraw; 77% of capacity from 2000–2019 cohort withdrew. ~700+ GW withdrew in 2024; ~500 GW new requests submitted same year. Queued Up 2025 Edition — LBNL - 2025-12-15
- Project size growth (2015 → 2023). Mean solar plants 250% larger; standalone batteries 330% larger; wind 66% larger.
- Cost data. Interconnection costs in 2019-2023 were 44% greater than during the preceding 5 years.
- Regulatory recognition. FERC ordered PJM (Dec 18, 2025) to overhaul rules for co-located and behind-the-meter large loads — first major regulatory acknowledgment that the queue mechanism cannot keep pace with hyperscaler demand. Microsoft Electricity Cost Recovery Commitment — POWER Magazine - 2026-01-22
Tensions & Counterarguments
- Vertical integration as queue bypass. Google’s acquisition of Intersect Power and Microsoft’s Brookfield deal both sidestep traditional interconnection queues by co-locating generation with load (behind-the-meter). Whether this becomes the dominant pattern or remains supplemental to grid-scale connection is open.
- Withdrawal-rate framing. The 70% withdrawal rate is sometimes cited as evidence the queue is “self-cleansing.” LBNL’s analysis suggests instead that withdrawal indicates structural failure: projects reach the front of the queue at price points or timelines that no longer support the economics that justified them.
- The queue isn’t a single entity. “The queue” is shorthand for seven different RTO queues running on different rules; reform of one (FERC’s December 2025 PJM order) doesn’t reform the others. Cross-RTO comparison requires reading each separately.
Related Concepts
- AI Buildout Grid Constraint — master concept; this concept is the document-class layer of that bottleneck
- Chokepoint Control — the queue functions as a chokepoint specifically because RTOs are the sole legitimate path to grid-scale generation deployment
- Hyperscaler Vertical Integration — the dominant workaround pattern (deferred stub)
Key Sources
- Queued Up 2025 Edition — LBNL - 2025-12-15 ← primary; most current authoritative data
- US Interconnection Queue Twice Installed Capacity — Latitude Media - 2024-04-11 ← 2024 edition figures
- Google Intersect Power Acquisition — Introl - 2026-01-20
- Microsoft Electricity Cost Recovery Commitment — POWER Magazine - 2026-01-22
- Big Tech Promised $650B Data Centers Most Not Being Built — Bricks & Bytes - 2026-04-28