Yes. For new AI data centers in the US, power is now the slowest input to secure. A typical large power transformer takes 2.5 to 3 years to procure (NLR, July 2026), and the median power plant built in 2025 waited 61 months for its grid connection (LBNL, June 2026).

Where is the constraint showing up: equipment, interconnection or generation?

It shows up in all three, and each one has a public number behind it.

Constraint Latest figure Source
Large power transformers 2.5 to 3 years to procure a typical unit; up to 5 years for extra-high-voltage units NLR, July 2026
Transformer supply 30% supply deficit for power transformers in 2025; imports about 80% of US supply Wood Mackenzie, August 2025
Gas turbines 116 GW of gas equipment backlog and slot reservations at end of June GE Vernova, July 2026
Generator interconnection 2,061 GW waiting in queues; median 61 months from request to operation LBNL, June 2026
Large-load requests in Texas About 474 GW requested, roughly 90% from data centers Utility Dive, August 2026

Equipment first. Wood Mackenzie reported in August 2025 that US demand for power transformers rose 116% since 2019, with a 30% supply deficit in 2025. The National Laboratory of the Rockies puts extra-high-voltage units at up to 5 years (July 2026). On generation, GE Vernova's July 2026 results show gas backlog and slot reservations up from 100 GW to 116 GW in one quarter, against a factory rate of 20 GW a year from the third quarter of 2026.

Interconnection is the longest wait. LBNL's Queued Up 2026 edition counts 2,061 GW of generation and storage in queues, larger than the 1,374 GW US installed fleet. The median project built in 2025 took 61 months from request to operation, against 36 months in 2015. Only about 19% of projects that applied from 2000 to 2020 were operating by the end of 2025 (LBNL, June 2026).

How much new load are grid planners expecting?

They expect data centers to take a much larger share of US electricity by 2030. The LBNL 2025 update (June 2026) puts data center use at 192 TWh in 2024, or 4.7% of US electricity, and 649 TWh in 2030 in its reference case, or 11.8%. The scenario range for 2030 is 9.5% to 15.3%.

Grid operators show the same direction. PJM's January 2026 forecast has summer peak growing 3.6% a year to 222,106 MW in 2036, an increase of 65,733 MW, and names data center load as a driver in several zones. In Texas, large-load requests to ERCOT total about 474 GW, roughly 90% of them from data centers, and on August 3, 2026 the governor directed regulators and ERCOT to audit data center requests (Utility Dive, August 2026).

Which buyers now have budget and urgency?

Hyperscalers, the utilities that serve them, and the generators that sell to both. Hyperscalers sign long contracts for power that arrives years out:

These deals fix supply for the 2030s and leave a gap between today and the day that power arrives. Storage, flexible load, grid software, power electronics and on-site power sell into that gap, because they can add capacity before new plants and lines are built. Equipment makers see the same pull: GE Vernova reported data center orders of over $5 billion year-to-date at the end of June 2026, more than double its 2025 total (GE Vernova, July 2026).

What does this mean for a vendor's sales plan?

Sell months saved to energization, and sell them to the person who owns that date. At a hyperscaler or data center developer, that is the energy procurement or power development lead. At a utility, it is the large-load interconnection team and transmission planning. At an independent power producer, it is origination. Each one has a different approval path, covered in who signs the purchase at a utility, a data center and an IPP.

Expect two clocks. Data center buyers move on the construction schedule. Utilities move on regulatory and budget cycles, which is why the utility sales cycle needs its own forecast line. Fast-track study processes also create short windows: MISO and SPP run Expedited Resource Addition Studies and PJM runs a Reliability Resource Initiative, which promise timelines as short as three months from request to agreement (LBNL, June 2026). Projects in those lanes need equipment, software and financing fast.

Which segments are we counting, and how big are they?

About 2,500 private companies sell into the power bottleneck (our data, Oct 2026). In October 2026 we counted companies with 21 to 500 staff in a B2B company database by industry tag and removed utilities, co-ops, installers and associations.

Segment (21 to 500 staff) Estimated companies
Grid software, VPP, demand response and DER about 700
Energy storage technology and developers about 1,300
Data-center power and infrastructure about 180
Power electronics about 180
Energy trading and energy fintech about 180
Total, overlap removed about 2,500

These are estimates from tag counts with overlap removed (our data, Oct 2026). Storage is the largest group by far, so storage sellers compete for the same developer and utility contacts. The data-center power group is small, and each company in it sells to a short list of buyers.

What should a vendor do this quarter?

  1. Write your value in months of time to power for one named buyer type. Use a figure you can defend in a customer reference call.
  2. Pick one constraint from the table above and build the account list around it: transformer-limited sites, queue-limited generators or load-limited utilities.
  3. Name the owner of the energization date at each target account before the first email.
  4. Track public triggers: a new hyperscaler power agreement, a large-load filing, an expedited study cluster or a utility load forecast revision.
  5. Build two pipelines with separate stage lengths, one for data center buyers and one for utilities.
  6. Re-check every external figure in your deck each quarter. GE Vernova, LBNL and the grid operators publish new numbers on fixed schedules.

To see where your pipeline misses the buyers with budget, start with The Diagnostic.