Published: September 13, 2026 | Category: Technology | By Mahesh
Finished Doesn't Mean Connected
Google has publicly reported facing potential transmission grid connection delays of up to 12 years for some new data center projects.[1] Not twelve months. Twelve years, for a facility that may already be fully built, fully staffed with the electricians and technicians Depth Grid's earlier reporting this week found in critically short supply, and sitting idle while it waits for permission to actually draw power from the grid. Depth Grid's pillar article this week on the compute-and-power bottleneck named grid interconnection delay as a core constraint alongside gas turbine backlogs, chip packaging, water, labor, and international competition. This piece is the seventh and final spoke in that series, going specifically into the interconnection queue itself: the formal, often multi-year administrative and engineering process a project must clear after it is financed, permitted, and built, simply to be switched on.
What "the Queue" Actually Is, and Why It's Not About Construction
The interconnection queue is a formal, federally structured process, not an informal waiting list. Electric transmission system operators, including the regional Independent System Operators and Regional Transmission Organizations that manage most of the US grid, require any proposed power plant or, increasingly, any proposed large electricity load like a data center campus, to undergo a series of impact studies before it can actually connect to the transmission system.[2] This process establishes what new transmission equipment or grid upgrades the connection will require, and assigns the cost of building that equipment, and the ordered lists of projects moving through this study process are what the industry and Lawrence Berkeley National Laboratory, the U.S. Department of Energy's own primary research authority on this data, call interconnection queues.[2]
Legal analysis of the specific regulatory mechanics involved makes clear why this process has become such an acute constraint specifically for data centers rather than a general construction delay. AI data centers are large loads that need to connect to the interstate transmission grid, but the Federal Energy Regulatory Commission has historically regulated how power plants connect to the grid far more thoroughly than it has regulated how large loads connect, since load interconnection was never previously a major bottleneck before AI-scale data centers emerged as a genuinely new category of massive, concentrated electricity demand.[3] This regulatory gap is precisely why the Department of Energy, under a directive issued on October 23, 2025, invoked a rarely used federal authority to specifically direct FERC to write the country's first dedicated large-load interconnection rule, an acknowledgment that the existing regulatory framework, built primarily around generators rather than loads, simply was not designed to handle the specific administrative challenge a gigawatt-scale AI campus now presents.[3]
What Berkeley Lab's Own Data Shows
Lawrence Berkeley National Laboratory's own "Queued Up" report, the most authoritative and comprehensive dataset available on this topic, compiled interconnection queue data from more than 50 transmission grid operators, covering seven ISOs and RTOs and roughly 50 non-ISO balancing areas, representing approximately 98 percent of all currently installed U.S. electric generating capacity.[4] The 2026 edition of that report, analyzing data through the end of 2025, documents a queue that has grown to a genuinely staggering scale: total active interconnection queue capacity reached 2,060 gigawatts as of end-2024, roughly double the entire installed capacity of the current U.S. bulk power system.[5] That comparison is worth sitting with directly: the amount of generation and storage capacity currently waiting in line to connect to the American grid is larger than the grid itself.
The specific composition of that backlog, and the wait times attached to it by region, are documented with enough granularity to be genuinely useful for project planning rather than remaining an abstract national statistic. Solar accounts for roughly 1,090 gigawatts of the queued capacity, battery storage roughly 580 gigawatts, wind roughly 220 gigawatts, and natural gas roughly 110 gigawatts.[5] AI data center loads, classified as large loads rather than generators in most regional transmission organizations, sit in separate but parallel queues carrying comparable wait times.[5] Regional variation is significant and directly actionable for anyone comparing potential project locations: PJM zones serving the largest AI campus filings see waits approaching seven years, CAISO waits run five to six years, MISO waits average around five years, SPP waits run four to five years, while ERCOT, which operates a separate large-load interconnection process specifically for Texas, runs three-to-four-year typical waits for AI campus loads above 75 megawatts, generally the fastest major U.S. market for this specific approval process.[5]
| Grid Operator | Typical Wait Time | Notes |
|---|---|---|
| PJM (largest AI campus filings) | Up to ~7 years | Serves the mid-Atlantic and Midwest, includes Virginia's data center corridor |
| CAISO | 5-6 years | California's grid operator |
| MISO | ~5 years | Midwest and parts of the South |
| SPP | 4-5 years | Central US, Great Plains |
| ERCOT (large-load process) | 3-4 years | Texas-only, separate process for loads above 75 MW, generally fastest major market |
Source: Lawrence Berkeley National Laboratory, "Queued Up: 2025/2026 Edition," analyzed via SAVRN operator brief, June 2026.[5]
Why 80% of the Queue Never Gets Built at All
A critical detail that separates the raw 2,060 gigawatt headline figure from the realistic amount of capacity actually likely to reach commercial operation is the queue's extraordinarily high attrition rate. Nearly 80 percent of new projects entering the queue ultimately withdraw before completion, primarily due to unpredictable, multi-year delays and prohibitively high assigned grid upgrade costs.[6] This attrition pattern is not a new phenomenon specific to the current AI-driven surge, historical Berkeley Lab data covering projects that entered the queue between 2000 and 2017 found less than a quarter had actually been built as of the end of 2022, confirming this high failure rate has been a persistent structural feature of the U.S. interconnection process for well over two decades, not a recent development caused specifically by AI demand.[7]
The mechanism driving this high withdrawal rate matters directly for how a data center developer should interpret their own position in a queue. Projects generally do not exit the queue at the same rate they enter it, and the underlying reason is a "first-come, first-served" or, more recently, a cluster-based study process in which a single project's required grid upgrade costs can shift substantially as later, larger projects join the same queue ahead of it, sometimes forcing an earlier project to be re-studied entirely and pushed further back.[7] FERC Order 2023, finalized in July 2023 and still being implemented as of this piece's publication, replaced the legacy first-come-first-served model specifically with a cluster-study framework, tighter financial commitments required at each project milestone, and a "reasonable efforts" standard intended to hold regional transmission organizations more directly accountable for actually processing queued projects within a predictable timeframe, though Berkeley Lab's own reporting cautions it remains too early to measure and assess the full impact of these reforms given how recently many of them have taken effect.[4]
The Federal Response Already Underway
The scale of the AI-driven load interconnection problem has produced a genuinely unusual and fast-moving federal regulatory response, one worth tracking closely given how directly it could reshape the wait times documented throughout this piece. On October 23, 2025, the Secretary of Energy invoked a rarely used authority under Section 403 of the Department of Energy Organization Act to direct FERC to open a formal rulemaking specifically on large-load interconnection, attaching an Advance Notice of Proposed Rulemaking containing fourteen guiding principles for how that new rule should be structured.[3] FERC opened the resulting docket, RM26-4, took public comment, and announced on April 16, 2026 that it would take action by June 2026, a genuinely rapid timeline by the standards of federal energy regulatory rulemaking.[3]
FERC followed through directly: on June 18, 2026, the Commission issued tailored orders under Section 206 of the Federal Power Act to each individual U.S. regional grid operator, a historic action fulfilling the DOE's original October 2025 request to drastically accelerate grid interconnection specifically for large-load energy users including AI data centers, amid what the agency's own order describes as unprecedented energy demand.[8] Analysis of FERC's approach notes a deliberate legal strategy behind how the orders were structured: by issuing region-specific orders restricted to FERC's own established jurisdiction, rather than attempting one sweeping national rulemaking, the Commission maximized the legal durability of its actions against likely challenges from state-level stakeholders who might otherwise contest a broader federal rule as regulatory overreach.[8] Individual grid operators have moved in parallel with this federal push: PJM and SPP have already adopted their own large-load tariffs independently, and PJM has separately proposed an Expedited Interconnection Track specifically to accelerate the highest-priority large-load projects through its own internal process.[3]
How Developers Are Routing Around the Queue Entirely
Given the severity and unpredictability of the delays documented throughout this piece, a growing share of AI infrastructure developers are not waiting for federal or regional reform at all, instead pursuing strategies specifically designed to avoid the traditional interconnection queue altogether. Behind-the-meter generation, building dedicated on-site power generation, most commonly the natural gas turbines Depth Grid's earlier reporting this week documented as themselves facing multi-year manufacturing backlogs, that bypasses the public grid entirely, has become one of the two dominant alternative strategies identified in current industry analysis.[6] Crusoe's on-site gas turbine buildout at the Stargate campus in Abilene, Texas, covered in Depth Grid's earlier reporting on the turbine backlog, is a direct, named example of exactly this strategy in active deployment.
The second major alternative strategy is the flexible interconnection agreement, in which a developer accepts specific operational constraints, such as agreeing to curtail electricity draw during periods of peak regional grid stress, in exchange for a meaningfully faster connection than the standard queue process would otherwise provide.[6] New grid-enhancing technologies are also gaining genuine commercial traction as a complementary, faster, and cheaper alternative to waiting for entirely new transmission construction, with companies providing AI-based software platforms specifically to accelerate the underlying interconnection studies themselves, a case of AI infrastructure development using AI tooling to help solve its own core physical bottleneck.[6] A concrete illustration of how fast an interconnection can move when these accelerated pathways are actually used successfully: a 200 megawatt data campus authorized in early 2026 through one such accelerated process reached first-token operation inside the first quarter of 2027, a timeline dramatically faster than the four-to-seven-year waits documented as typical across the major regional grid operators earlier in this piece.[5]
What This Means for Anyone Evaluating a Project's Real Timeline
Request a project's specific queue position and study status directly, not just its announced target completion date. Given that a project's realistic timeline depends heavily on precisely where it sits in a specific regional transmission organization's cluster-study process, and given the documented near-80-percent withdrawal rate across the queue as a whole, an investor or partner evaluating a specific data center project should request the project's actual queue position, current study phase, and any history of re-studies or cost reallocation, rather than relying solely on a developer's announced target completion date.
Weight regional interconnection timelines directly against the site-selection factors covered elsewhere in this series. With ERCOT's three-to-four-year large-load process running meaningfully faster than PJM's seven-year waits for major campus filings, a developer with genuine geographic flexibility should weigh interconnection timeline explicitly alongside the water availability, labor market depth, and gas turbine access factors Depth Grid's earlier reporting this week identified as the other major regional variables shaping where a project can realistically be built on schedule.
Treat behind-the-meter generation and flexible interconnection agreements as genuine schedule de-risking tools, not simply cost trade-offs. Given how directly a queue-bypass strategy can compress a project's timeline, as illustrated by the 200 megawatt campus reaching operation within roughly a year through an accelerated process versus the four-to-seven-year standard queue wait, any project with a genuinely time-sensitive commercial rationale should evaluate on-site generation or a flexible interconnection agreement as a primary path rather than a fallback option to consider only after the standard queue process has already proven too slow.
Common Questions
This analysis is editorial commentary based on publicly available sources cited above. It is not financial, investment, engineering, or regulatory advice. Interconnection queue data, wait times, and regulatory timelines cited reflect data available as of publication and change frequently as FERC rules and regional grid operator processes evolve; verify current figures directly with Lawrence Berkeley National Laboratory, FERC, and the relevant regional grid operator before making decisions based on this information.
Sources
- ComputeLaw, "How AI Data Centers Connect to the Grid (FERC and the Power Queue)," citing Google's own reported delay figures, updated August 3, 2026. Link
- Lawrence Berkeley National Laboratory, "Queued Up: 2026 Edition, Characteristics of Power Plants Seeking Transmission Interconnection as of the End of 2025," Energy Markets & Policy Program. Link
- ComputeLaw, "How AI Data Centers Connect to the Grid (FERC and the Power Queue)," citing DOE Section 403 Directive and FERC Docket RM26-4, updated August 3, 2026. Link
- Lawrence Berkeley National Laboratory, "Queued Up: Characteristics of Power Plants Seeking Transmission Interconnection," Energy Markets & Policy Program, data updated May 2026. Link
- SAVRN, "AI Data Center Grid Interconnection: 2026 Operator Brief," citing Lawrence Berkeley National Laboratory Queued Up 2025 Edition, June 17, 2026. Link
- EnkiAI, "Grid Interconnection Delays 2026: A Threat to US Energy," April 28, 2026. Link
- Utility Dive, "US Grid Interconnection Backlog Jumps 40%, With Wait Times Expected to Grow as IRA Spurs More Renewables," citing Berkeley Lab historical completion-rate data. Link
- American Action Forum, "FERC Data Center Orders Accelerate Grid Connection," citing FERC's June 18, 2026 Section 206 orders, June 18, 2026. Link
Read More on Depth Grid
- The Compute-and-Power Bottleneck: Why AI's Next Decade Will Be Decided by Electrons, Not Algorithms
- Order a Gas Turbine Today, Get It in 2031: The Backlog Quietly Gating Every AI Data Center
- Microsoft's Own President Says Electricians, Not Chips, Are the Biggest Constraint on AI Expansion
- China Builds a Data Center in Months. The US Takes Years. Speed Is the New AI Race.
Article by Mahesh | Depth Grid