The Data Center Interconnection Queue: Steps, Waits and Options

If you are planning AI capacity, the data center interconnection queue often sets the date that matters most. Land, buildings and chips can be bought. Power at the size you need, delivered to your fence line, has to be granted by whoever runs the local grid, and they work through requests in order.

This guide covers what the queue is, the steps a large load goes through, why the wait is long, what changed in 2025 and 2026, and the realistic ways to reach power sooner. Numbers come from FERC, DOE, Berkeley Lab and ERCOT, with the date we checked them.

What is an interconnection queue?

An interconnection queue is the list of projects that have formally asked to connect to the grid and are waiting for studies, an agreement and construction. Each request is checked for what it would do to the system and what must be built first.

What is grid interconnection? It is the physical and contractual link between a site and the electric system: lines, substation, protection equipment, and an agreement that sets how much power can flow and who pays for what.

There are two kinds of queue, and they are easy to confuse:

• Generation queues hold power plants and batteries asking to put power onto the grid. These are the queues tracked in Berkeley Lab's annual Queued Up report. At the end of 2025 they held about 1,312 GW of generation and about 749 GW of storage.
• Load queues hold large users asking to take power off the grid: data centers, factories, electrolysers. The local utility runs the process, sometimes with the grid operator. Berkeley Lab's Queued Up dataset does not include load requests at all.

So when you read that U.S. queues hold over 2,000 GW, that is supply, not data centers. A data center grid connection goes through its own process, and it is less standard. Berkeley Lab describes load processes as ranging from ad hoc to well defined.

Still, the two are linked. A large load often needs new generation and transmission to serve it, so your request can be ready while the supply behind it is not.

What steps does a data center grid connection go through?

Most large-load requests follow the same broad path, though each utility names the steps differently. Berkeley Lab summarises it as inquiry, request, studies, agreement, construction and in service.

1. Inquiry. You ask the utility whether a site could take your load, and roughly when. Answers here are informal.
2. Formal request. You submit size, location, ramp schedule and in-service date. Expect to show site control and pay a study deposit.
3. Studies. The utility, and in organised markets the grid operator, models what your load does to the system. The output is a list of upgrades and a cost estimate.
4. Agreement. You sign an interconnection or service agreement setting capacity, dates, upgrades and who pays. This usually needs a larger financial commitment.
5. Construction. The utility builds or upgrades substations, lines and protection. This is where equipment lead times bite.
6. Energisation and ramp. Power is switched on, often in stages, with full load allowed only as later upgrades finish.

Milestones run alongside: site control, deposits, financial commitments and long-lead equipment orders. Miss one and a request can lose its place.

How big is large? DOE's 2025 proposal to FERC defined large loads as greater than 20 MW. ERCOT treats a single site above 75 MW of peak demand as a large user. An AI campus is usually well above both. For how loads add up, see how much power a data center uses.

Why is the data center interconnection queue so long?

The queue is long because requests far outnumber what the grid can absorb quickly, and each needs studies, upgrades and equipment that take years. Five causes explain most of it.

1. Volume. Requests have outrun the system. As of mid-2026, ERCOT was tracking more than 438,000 MW of large-load requests, nearly 90% of them from data centers. Its all-time peak demand record is 85,508 MW, set in August 2023. In PJM, large-load adjustments in the 2026 forecast reach 39 GW by 2030 and 87 GW by 2040, against a peak system of 154 GW.

2. Speculative and duplicate requests. Developers often file for one project at several sites. Each request still has to be studied, and each changes the results for others. The generation side shows how few requests become projects: of capacity that applied from 2000 to 2020, only 13% was operating by the end of 2025 and 75% had withdrawn.

3. Studies and restudies. Studied one at a time, each new project can change the answer for those already in line. ERCOT says this triggered costly restudies that delayed timelines, which is why it moved to batch studies.

4. Network upgrades. If a study finds lines or substations would overload, they must be upgraded first. New transmission takes years to permit and build, and the upgrade can be far from your site.

5. Equipment lead times. A DOE report to Congress said 36-month lead times for large power transformers were commonly quoted, with some reaching 60 months. Before the pandemic, the lead time could be less than a year.

The generation side has numbers. Berkeley Lab found the median time from request to signed agreement was well over 3 years in 2025, and over 5 years from request to operation for projects built in 2025. No national dataset exists yet for load requests.

What changed for large loads in 2025 and 2026?

Regulators now handle large-load connection as a separate issue. The rules are still changing, so check the current state before planning around them.

• DOE's proposal to FERC (October 2025). The Secretary of Energy asked FERC to consider rules for connecting large loads to transmission. It proposed that loads pay 100% of the network upgrades their studies assign, and that curtailable loads get expedited studies.
• FERC show cause orders (June 2026). FERC ordered all six regional grid operators it oversees to justify or reform their large-load tariffs within 60 days, covering study processes, cost shifting, co-location and behind-the-meter generation, and services for flexible loads.
• ERCOT batch studies (2026). ERCOT replaced one-at-a-time studies with a batch process that studies eligible large loads together. The first group, Batch Zero, began in July 2026.
• Generator queue reform. FERC's Order No. 2023 moved generators to a first-ready, first-served cluster study, with stricter financial readiness and site control rules.

One idea comes up again and again: flexibility. A load that agrees to cut use when the grid is stressed may connect before every upgrade is built. Berkeley Lab lists 41 possible solutions for speeding large-load connections, several of them curtailable or non-firm service.

What does time to power mean for an AI tenant?

Time to power is the time from deciding you need capacity to having energised, usable power at your racks, at the size you contracted. For an AI company it often sets every other date.

Three things make it different from a normal construction schedule:

• It is not fully in your control. The building can be on schedule while the substation is not.
• Partial power is common. If your cluster needs full load to be useful, the date that matters is the last stage, not the first.
• Hardware does not wait. Accelerators lose value while they sit in storage. A late power date can cost more than a higher rent.

When comparing sites, ask for the power date in writing and what it depends on: a signed agreement or only a study, firm or curtailable service, and which upgrades are outstanding. A site with no signed agreement is a site without a date. This guide covers only the grid connection; for the full schedule, see how long it takes to build a data center.

What are your options if you cannot wait for the grid?

You have four realistic routes. None is free, and the table below compares them.

Wait in the queue. Right if you have time and need a specific location. Meet every milestone, and ask whether flexible or curtailable service could connect you sooner.

Pick a different market. Some regions have spare capacity or faster processes. The trade-off is location: latency, fiber, staff and tax rules all change.

Bridge power, then grid. Temporary on-site generation runs until the grid connection arrives. It shortens time to power but adds permits, fuel and a second transition later.

Behind the meter. The site runs on its own on-site generation, so the load queue leaves the critical path. Process does not disappear: you still need firm fuel, air permits and generation equipment with its own lead times. See behind-the-meter data centers for the design and gas-powered data centers for the fuel side.