Grid Connection Queues: Why Power Is the Data Center Bottleneck

Grid connection queues, not construction schedules, now set the delivery date for most large data center projects. In PJM, active large-load projects wait 36-48 months; in Dominion Energy’s Northern Virginia territory, the average is close to seven years. A campus that takes 18-24 months to build can sit finished and empty for years waiting on a utility interconnection agreement. Buyers who treat power as a line item rather than the critical path are the ones getting the multi-year surprise.

Key takeaways

• PJM large loads: 36-48 months for active projects in data center growth zones, with some cases running past five years, per Ascend Analytics.
• Dominion (Northern Virginia): roughly seven years average for a 100 MW-plus connection; the utility’s contracted pipeline grew from ~40 GW in early 2025 to 47.1 GW by October 2025.
• ERCOT (Texas): faster at 4-6 years for a 500 MW transmission-level interconnection, but the large-load queue grew from roughly 60 GW to 238.6 GW in about a year, and Senate Bill 6 now requires $50,000/MW in financial security above 75 MW.
• The queue is now the slower half of the timeline. A data center takes 1-3 years to build; new transmission and generation capacity takes 5-15 years to plan and permit — the mismatch is structural, not temporary.
• Behind-the-meter power is a partial workaround, not a shortcut. Gas turbine backlogs (GE Vernova 116 GW, Siemens Energy ~69 GW) now run three years or more themselves; only smaller aeroderivative units deliver in 12-18 months.
• FERC is rewriting the rules. June 2026 show-cause orders direct all six US RTOs/ISOs to justify or reform how they interconnect large and co-located loads; a compliance timeline runs through late 2026 and into 2027.
• Southeast Asia uses quota allocation instead of an open queue. Singapore’s DC-CFA rounds and Malaysia’s TNB Green Lane Pathway cap available capacity but avoid the US-style multi-year backlog for approved projects.

For live capacity and pricing context by market, see the colocation price index, the data center catalog, and the power sector hub.

How bad the US queues actually are

Three numbers frame the problem. First, the national interconnection backlog: US generation and storage projects awaiting grid access totaled roughly 2,600 GW as of 2026, more than double the country’s entire installed generating capacity. Second, the AI-load-specific queue: PJM and ERCOT alone saw more than 10 GW of newly announced data center projects since October 2025, with only about 30% targeting online dates by 2027. Third, the withdrawal rate: nearly 80% of queued interconnection requests are eventually withdrawn, mostly because multi-year delays and grid upgrade cost allocations make the project uneconomical before it reaches the front of the line.

The result is a queue that grows faster than it clears. ERCOT’s large-load queue alone expanded from roughly 60 GW to 238.6 GW in about a year, with data centers accounting for 77.5% of that total, per Utility Dive and Latitude Media reporting. That growth rate — not any single project — is what pushes wait times out further every quarter.

Market-by-market wait times

Market / operator	Typical wait for a large load	Queue model	Notes
Dominion Energy (Northern Virginia)	~7 years for 100 MW+	First-come, first-served	70 GW in queue; only ~25 GW have assigned connection dates through 2031
PJM (broader footprint)	36-48 months, up to 5+ years in constrained zones	First-come, first-served, under FERC-ordered reform	New GS-5 rate class shifts infrastructure cost to developers
ERCOT (Texas)	4-6 years for 500 MW transmission-tier	Connect-and-manage	Queue at 238.6 GW; SB6 now requires $50,000/MW security above 75 MW
Singapore	Allocation-based, not a queue	DC-CFA capacity call	DC-CFA2 opened 200 MW in 2026; requires 50%+ green power
Malaysia (TNB, Peninsular)	As little as 12 months (Green Lane), down from 36	Fast-track for qualifying projects	49 ESAs signed representing 7.1 GW as of September 2025

Figures compiled from Ascend Analytics, Bloomberg, Utility Dive, Latitude Media, The Edge Malaysia and KWM/Morgan Lewis 2026 reporting; utility queue positions and connection dates change frequently as projects are added, studied or withdrawn.

The contrast is structural, not incidental. US markets run open, first-come-first-served queues where anyone can file an interconnection request — which is exactly why the queue fills with speculative and duplicate requests that later withdraw. Singapore and Malaysia instead cap the pool upfront through a call-for-applications or fast-track framework, trading open access for a shorter, more predictable timeline on the capacity that is actually awarded. Neither model eliminates scarcity — Singapore’s 2026 round only opened 200 MW nationally — but the quota approach avoids the multi-year backlog dynamic that now defines PJM and ERCOT.

Why the queue is longer than the build

A data center’s own construction — site work, shell, MEP fit-out, commissioning — runs 12-24 months for a standard build and up to 36 months for a large hyperscale campus, per our data center construction cost guide. Grid infrastructure runs on a different clock entirely:

• Interconnection studies stack sequentially. A feasibility study, system impact study and facilities study each take months, and a single large load can trigger new studies for every other project sharing the same substation or transmission corridor.
• Transmission builds take 5-15 years. New high-voltage lines require their own permitting, environmental review and, often, siting fights with landowners and municipalities — independent of how fast the utility wants to move.
• Generation adequacy is a separate constraint. Connecting a load doesn’t guarantee there’s enough generation to serve it at peak; FERC ordered all six RTOs/ISOs to file generation adequacy reports by July 20, 2026, specifically because load growth has outpaced generation build in several regions.
• Cost allocation disputes add delay. Who pays for the substation upgrade or transmission reinforcement a new data center triggers is frequently contested, and unresolved cost-allocation questions are a recurring cause of queue slowdowns, not just paperwork volume.

This is the same underlying scarcity discussed in our guide on rack power density: as average deployed density climbs, each new campus asks the grid for a bigger single bite, which makes queue and cost-allocation friction worse, not better.

Regulatory response: FERC’s 2026 large-load rulemaking

FERC has spent 2026 trying to force RTOs and ISOs to change how they handle large and co-located loads. Key milestones:

• December 2025: FERC found PJM’s tariff for loads co-located with generation “unjust and unreasonable” and ordered new rules; PJM’s compliance filing was partially accepted and partially rejected by FERC in April 2026.
• June 2026: FERC issued Section 206 show-cause orders to all six US RTOs/ISOs (PJM, MISO, SPP, ISO-NE, NYISO, CAISO — plus ERCOT, which sits outside FERC jurisdiction but faces parallel state-level pressure), directing each to justify its existing large-load interconnection tariff or propose reforms.
• July 20, 2026 deadline: each RTO/ISO owed FERC a generation adequacy report addressing whether enough generation exists to serve current and new large loads.
• Docket RM26-4: the broader rulemaking on interconnection of large loads to the interstate transmission system remains open, with further orders expected into 2027.

None of this shortens today’s queue. It’s aimed at the next cycle of projects and at standardizing rules — like flexible/curtailable large-load tariffs and co-location terms — that currently vary RTO by RTO and create uncertainty for anyone planning a multi-year buildout.

Workarounds operators are actually using

Behind-the-meter and co-located generation. Building or contracting on-site power — usually gas turbines, occasionally an adjacent nuclear plant — lets a facility start operating without waiting for a full grid interconnection, then optionally grid-tie later. The catch: this only moves the bottleneck. GE Vernova’s gas turbine order backlog reached 116 GW with deliveries booked into 2031, and Siemens Energy’s backlog hit roughly 69 GW with three-year-plus lead times on large frame turbines. Smaller aeroderivative packaged units can ship in 12-18 months — genuinely faster than most grid queues — but at a cost and reliability profile that suits a bridge, not a permanent primary supply.

Flexible and curtailable load contracts. Some utilities and RTOs now offer faster interconnection in exchange for accepting curtailment during grid stress events — the load gets a connection sooner but agrees to reduce draw on demand. This trades unconditional supply for queue position, which works for training workloads with schedule flexibility and works poorly for latency-sensitive inference or colocation tenants who can’t predict their own load profile.

Early, speculative interconnection filings. Because queue position is typically set by filing date, some developers now file for a location years before a tenant or even a confirmed use case exists, then sell or lease the queue position itself. FERC and RTOs have responded with tighter site-control and financial-security requirements (Texas SB6’s $50,000/MW deposit above 75 MW is one example) specifically to filter out these speculative filings.

Market selection as a power strategy. Increasingly, the site-selection question is inverted: instead of picking a market and then discovering the power timeline, buyers filter markets by published queue data first. This is why secondary US markets and non-US markets with faster or quota-based access — see our guides on the Singapore market and Johor, Malaysia — have absorbed demand that Northern Virginia and PJM can no longer serve on a competitive timeline.

What this means for a colocation or campus buyer

A power-first due diligence process now matters more than a real-estate-first one. Before signing a term sheet or LOI, verify — not take on faith — where in the interconnection process the site actually sits:

1. Ask for the interconnection agreement status, not just “power is available.” A signed and executed interconnection agreement with an assigned in-service date is a different asset than a study in progress or a queue application.
2. Check the utility’s public queue data where it exists. PJM, ERCOT and several state commissions publish queue position and study status; treat a landlord’s verbal timeline as a starting point for verification, not the answer.
3. Model the realistic date, not the marketed one. If the market average is 4-7 years and the site isn’t already through system impact and facilities studies, assume the long end of that range.
4. Weigh behind-the-meter power honestly. It can beat the grid queue by years, but only if the turbine or generation order is already placed — a 2026 order for large-frame gas turbines won’t arrive before 2029-2031 given current backlogs.
5. Compare quota-based markets on their own terms. Singapore and Malaysia trade open access for a capped but faster-moving allocation; that can be the better bet for a 2027-2028 go-live even before comparing headline power pricing, covered in our colocation pricing guide.

Run the numbers through our colocation price index and the sourcing checklist in our data center due diligence guide before treating any quoted power date as firm — in 2026, the grid connection, not the shell and core, is what determines whether a data center actually opens on schedule.