Time to power: the new constraint on US industrial development

The number that broke the model

Three to seven years. That is the current US grid interconnection wait time, depending on the ISO. PJM, the most-watched of them, has issued guidance that under its post-FERC Order 2023 cluster-study regime, wait times will sit at one to two years — for new requests, under the new process, assuming nothing slips. The current backlog is longer.

Against that, US data centre demand absorbed about 15,600 MW of new capacity in 2025 alone, more than 130 times the volume of a decade earlier. Wood Mackenzie's Q4 2025 read was that even at this pace, additions were slowing relative to demand. Forecasts converge around 100 GW of US data centre load by 2030. A quarter to a third of that is now expected to deploy behind-the-meter — Cleanview's February 2026 estimate puts the share at roughly 30% of incremental capacity, up from effectively zero one year prior.

Those three numbers — the queue length, the demand growth, and the BTM share — describe the same fact from three angles: the grid path no longer clears the volume, the demand isn't waiting, and the capital has already moved.

What "time to power" actually measures

Time to power is the elapsed clock from "we have a site and a load" to "electrons are flowing into the meter at the contracted rate." It folds in:

• Queue position and study time — the cluster study, the affected-system study, the facilities study, the interconnection agreement.

• Construction time on transmission upgrades — the substation buildout, the new line, the conductor upgrade, the breaker replacement.

• Permitting — federal, state, county, environmental review, sometimes tribal consultation.

• Generation construction — if the deal includes a new generation asset, the procurement-to-COD timeline of that asset.

• PPA negotiation and definitive documentation — the commercial layer that runs in parallel, but which is gated by the others.

Each of these used to be a quarter-by-quarter scheduling problem. They are now the dominant variable in whether a site has commercial value at all. A site at the front of the PJM queue is worth multiples of the same site sixteen positions back, even if the second site has better land economics, cleaner title, and stronger water access.

This is a structural change. The asset is not the land. The asset is the position of the land in a queue that didn't matter five years ago.

What changed

Three things changed within eighteen months.

The first is AI workload demand. The S&P Global read in late 2025 was that US data centres alone would need 22% more grid power by end of 2025 than they pulled the year before. Hyperscalers signed roughly 40 GW of new clean power contracts during 2025 (Microsoft alone). OpenAI's Stargate had committed 7 GW by mid-year and was targeting 10 GW by end of 2025. VoltaGrid plus Energy Transfer plus Oracle alone announced 2.3 GW of off-grid gas for a single hyperscale customer.

The second is the regulatory response, which is mid-flight. FERC Order 2023 (July 2023) restructured the queue to a first-ready, first-served cluster-study model — replacing the old serial first-come queue that had calcified across most ISOs. The order was correct. Implementation is staggered. ISO-NE rolled out its transitional process in 2024-25; PJM's new process is live but the legacy backlog still dominates the calendar through 2027. In February 2025, FERC issued a show-cause order to PJM specifically on co-location tariff clarity. In October 2025, the DOE directed FERC to accelerate interconnection of large loads — a policy directive that will produce a rulemaking, not an immediate effect.

The third is that the buy side adjusted faster than the sell side. Hyperscalers stopped waiting. The 17-year nuclear PPA AWS signed with Talen for the Susquehanna asset is the canonical example: an existing operating plant, behind-the-meter, no interconnection upgrade required, no new construction. Microsoft pulled forward Three Mile Island Unit 1 on similar logic. The pattern is now copied across natural gas BTM (gas reciprocating engines and turbines deployable in 18-24 months), behind-the-meter solar-plus-storage at large scale, and hybrid combinations.

The result: a bifurcated market. Sites with grid-path power are worth one number. Sites with BTM-path power are worth a different number, often higher despite lower nameplate capacity, because they can come online inside the window the offtaker actually has.

The fork developers now face

Every site under control in 2026 sits on one of two paths.

Grid path. Existing or pending interconnection request, transmission-side upgrade plan, utility offtake or wholesale market participation. Time to power: 3 to 7 years from the date of the interconnection request, depending on ISO and study queue position. Cost certainty: moderate — the utility upgrade cost allocation can move materially during the studies. Counterparty profile: the utility, the wholesale market, or a corporate offtaker willing to carry queue risk. Win condition: scale, queue priority, and a long-tenor offtaker with patience.

Behind-the-meter (or co-located) path. On-site generation paired directly with an industrial or hyperscale load, with the grid as backup or curtailment outlet rather than primary delivery channel. Time to power: 18-30 months for natural gas BTM, 24-36 months for solar-plus-storage, 36-60 months for advanced nuclear or geothermal. Cost certainty: higher — the project lives or dies on its own balance sheet, not on a regulated upgrade cost. Counterparty profile: the load itself, structured as a power services agreement, energy as a service, or direct PPA at the meter. Win condition: matched offtaker, financeable technology, regulatory tolerance for the co-location structure (which is jurisdiction-specific and currently under FERC review for PJM).

A site that can credibly do either is worth more than a site that can only do one. A site that can only do grid path is worth what its queue position says it's worth, no more. A site that can only do BTM path is worth what its load-counterparty matching says it's worth, no more.

The work of being a developer in 2026 is, increasingly, the work of structuring a site so that it has both options live until close, and then collapsing to whichever one the offtaker can actually transact on.

Why this is a matching problem, not a permitting problem

The instinct from inside development is to treat time to power as a regulatory or engineering variable: speed up the studies, win the queue position, optimise the substation design. All true. None of it solves the deeper problem.

The deeper problem is that the offtaker's clock and the queue's clock don't align. A hyperscaler, a tier-one industrial, or an emerging-market sovereign load has a deployment window measured in twelve to twenty-four months from intent to operating. The queue's clock — even under FERC Order 2023 — is measured in three to seven years for transmission-connected load growth. The structural answer to that misalignment is not faster studies. The structural answer is matching the offtaker to a path (and a site) that fits the offtaker's clock, not the queue's clock.

That matching is not happening at scale because the existing distribution channels — broker networks, PPA marketplaces, internal procurement teams — were built for a slower market. They optimise for price discovery, which was the binding constraint in 2018. Price discovery is not the binding constraint now. Time discovery is — finding the path-and-counterparty combination that closes inside the window. The market hasn't built infrastructure for that yet; it's being built deal by deal, mostly bilaterally, mostly at hyperscaler scale.

For a site developer below hyperscaler scale, the implication is concrete: the value of your site is not the engineering quality of your interconnection plan. The value of your site is the count of credible offtakers whose clock matches your path, multiplied by your probability of closing one of them inside their window. If that count is one, the site is exposed. If it's zero, the site doesn't have a deal — it has a story.

What this means for developers in the US

A few specific things change in how a US developer should think about a portfolio in 2026.

Pre-flight every site for both paths. Don't write off the BTM path because the site was originally underwritten for grid-tied. Walk the parcel and ask: what does this look like as a co-located 200-500 MW load with on-site generation? What gas, water, and permit access constraints would change? In a non-trivial fraction of cases, the BTM-path value is higher than the grid-path value, and the site was underwritten against the wrong number.

Qualify offtakers on clock, not just credit. A AAA offtaker with an 18-month window is more valuable than an A-rated offtaker with a four-year window for a site that can deliver in 24 months. Existing developer-side intake processes don't really test for clock alignment; they test for credit and price. Add a clock test.

Build the option discipline. Site options are the developer's working capital. The current market is consuming site options faster than developers can replace them — partly because deal mortality at term-sheet stage is 60-70%, partly because option windows weren't sized for current matching delays. Re-baseline the option terms on your live sites against current matching realities, not the 2022 baseline.

Geographic optionality matters. US is the headline market, but a portfolio that includes Australasian, West African, East African, South-East Asian, or Latin American sites benefits from different time-to-power constraints in those markets. Some are slower; some are dramatically faster, especially where DFI capital is willing to underwrite an early offtake counterparty against sovereign credit support. A pure-US portfolio is more exposed to PJM's queue dynamics than a globally diversified one.

Watch the FERC and DOE rulemaking calendar — but don't bet the option on it. The DOE October 2025 directive will produce something. FERC's PJM show-cause order will produce something. Both will probably help. Neither will help on a timeline that matters for a 2026 site option. Underwrite to the current rule, not the projected rule.

What changes for developer compensation

One specific note for developers reading this with a model open: SitePower's structure on the developer side does not include a developer-side success fee. The matching layer's economics sit with the offtaker side and the capital side, where the willingness to pay for time-to-close compression is concentrated. Developers running US (or international) sites through SitePower do not give up margin on the development asset to access the matching capability.

That is a structural choice, not a promotion. A developer-side fee structure would distort the matching incentive — pushing the developer to accept counterparties at the margin to clear the fee — which would erode the deal-mortality benefit that the matching layer exists to produce. Keeping fees off the developer side keeps the qualification gate honest.

What to do this quarter

If you run a US site portfolio: pull the live options, classify each one as grid-path, BTM-path, or both-path, and note the time-to-power range under each scenario. The exercise takes a half-day. The output reorders the portfolio.

If a site that you previously underwrote as grid-tied has a credible BTM path — and most do, more than the original underwriting would have flagged — the next step is matching it to an offtaker whose clock fits. That is the work of the matching layer, and it's the work that's currently sitting underdone across most non-hyperscale deals in the country.

If the site doesn't have a credible BTM path and the queue position doesn't clear inside the option window, the honest move is to fail it fast. Hold capital for the next site. The cost of carrying a site through a third option extension while the offtaker decides is higher than the cost of admitting the matching doesn't work and recycling the capital.

The constraint has changed. The model has to follow.