The interconnection queue is now the longest lead item in a data center build

Chips arrive in months. Buildings go up in a year or two. But in most major U.S. markets, the grid connection — the thing everything else waits on — now takes longer than either. Here is what the data says, and how behind-the-meter generation changes the critical path.

How long the queue actually is

Lawrence Berkeley National Laboratory's Queued Up series is the standard reference for U.S. interconnection timelines. Its findings are stark: the median time from interconnection request to commercial operation has roughly doubled since the 2000s, and for projects built in 2025 the median exceeded five years. Large loads that trigger network upgrades — and a gigawatt campus almost always does — commonly face four-to-seven-year timelines in major markets, with transmission-dependent projects at the long end.

The queue is not a formality that runs alongside your build — it gates it. A campus that breaks ground today on a standard grid-served model is, in effect, scheduling its first megawatt for the early 2030s.

Why AI demand broke the queue model

Interconnection queues were designed for a world where load grew a percent or two a year and generators waited patiently for study cycles. AI training campuses invert every assumption: the load is enormous, it arrives in one step rather than gradually, and the business case is measured against model-generation cycles, not utility planning horizons.

The market has responded with long-term nuclear offtakes — hyperscalers have announced roughly 10 GW of them — but almost none of that capacity delivers this decade. Contracted capacity is not delivered capacity. The binding constraint for anyone who needs gigawatt-scale compute before 2030 is not willingness to pay; it is a delivery date someone will sign.

What behind the meter actually changes

Behind-the-meter generation places the power plant on the same site as the load, selling electricity directly to the campus under a power purchase agreement. That changes the structure of the problem, not just its speed:

• First power tracks plant construction, not queue position. Gas-plus-storage generation is a known quantity with a known build time; your energization date is an engineering commitment, not a study-cycle outcome.
• The grid becomes redundancy, not the critical path. An interconnection can be developed in parallel for backup and export optionality — valuable, but nothing waits on it.
• One counterparty owns the date. When generation and load share a fence line, delivery obligations can carry liquidated damages instead of best-efforts study schedules.

What this looks like in practice

Thorium One Power develops behind-the-meter campuses in the Permian Basin around exactly this logic: a 1.1 GW bridge gas fleet with storage targeting first power in 2028, molten-salt nuclear phasing in from 2033 to drive the blended price and carbon intensity down on a published schedule, and 10.4 million gallons per day of drought-proof cooling water from produced-water desalination — power and water under mirrored 20-year agreements with one counterparty, with liquidated damages on commercial operation dates. Grid interconnection is developed in parallel as redundancy and export optionality.

The water half of that package has its own economics and its own moat — covered in the companion piece on cooling a gigawatt campus without freshwater.

Questions this page answers

Sources & notes: interconnection timelines — Lawrence Berkeley National Laboratory, Queued Up series (2025 Edition: median time from interconnection request to commercial operation exceeded five years for projects built in 2025; durations have roughly doubled since the 2000s). Hyperscaler nuclear offtake volumes are aggregate public announcements, most with delivery dates in the 2030s. Thorium One Power campus capacity, water and date figures are targeted and indicative, subject to diligence and final engineering.