GridCARE verified 5 gigawatts of spare grid. It indexes the rest at $350B.

Five gigawatts. In twelve months, across six American utilities, GridCARE says it has qualified that much additional load-serving capacity on grids that were already built, already paid for, and already rated as full.

On October 8, at a summit in San Francisco, the company published the first attempt to put a national price tag on the idea. Its Power Acceleration Index estimates that up to 120 GW of extra load could be served from existing transmission and distribution assets if large customers accept flexible terms. GridCARE puts the net customer benefit at $270–350B, and the household share at $130–170 a year — a 7–9% cut to utility bills.

The distance between those two paragraphs is the entire story. Five gigawatts verified is a portfolio of named substations. 120 gigawatts estimated is a model with published screening assumptions and no site behind it yet.

Utilities size equipment for the worst hour of the worst year, not the average one. That conservatism is why a Stanford study published in December 2025 found Western transmission lines carrying as little as 18% of rated capacity even at peak demand, with most utilities clustered near 30%. If that headroom is real, the case for building new transmission weakens sharply. If it is partly an artefact of how capacity gets counted, the $350B estimate dissolves.

The five gigawatts, named

GridCARE's 5 GW is not one abstract number. It is four disclosed projects and a remainder.

Portland General Electric's Hillsboro site accounts for more than 400 MW, with 80 MW offered to data centres in 2026 and the rest energised by 2029 at the latest. Hillsboro matters because it is where fiber-optic submarine cables come ashore. As we wrote in September, PGE has been treating grid flexibility for data centres as a commercial product since before anyone had a name for the category.

National Grid accounts for 650 MW on its New York network. The utility's chief executive, Zoë Yujnovich, told investors on the May 2026 results call that the company had identified up to 650 MW of flexible connection capacity for large loads — and that doing so reduces connection time and cost. Santee Cooper is working with GridCARE on large-load qualification in South Carolina, where Chris Wagner, senior director of technical services, says traditional assessments would have required lengthy capital-intensive upgrades. Southern California Edison found up to 20 MW on an Inland Empire circuit previously thought fully constrained, the first California utility to qualify capacity this way.

Add another 2 GW across unnamed Pacific Northwest utilities and the announced total clears 3 GW. The remaining ~2 GW of the headline figure sits with two utilities that have not been named.

The twenty-four-fold gap

GridCARE verified roughly 5 GW. Its Index projects 100–120 GW. That is a twenty-four-fold expansion from a demonstrated base, and the Index itself acknowledges it: results at the utilities it has worked with "show these figures are achievable", while actual capacity at any site still depends on network-specific study.

The comparison that matters is not against GridCARE's own ceiling. It is against the one peer-reviewed number available on what flexibility can actually deliver inside today's grid.

A 2026 analysis in Cell Reports Sustainable Energy modelled the theoretical upper bound on integrating data centres into existing US systems without new generation or transmission, using a peak-preserving criterion: new demand must not lift the system's historical net-load peak. In ERCOT, ten hours of annual flexibility accommodates about 2 GW at a 99.95% load factor. Fully flexible year-round operation — an idealised bound requiring perfect foresight and instantaneous response — reaches 59.24 GW. Both figures are theoretical headroom, not deployable capacity, and the authors say so.

Two bank estimates sit in the same range. Bank of America has US demand of 230 GW arriving between 2026 and 2030 against 93 GW of new supply, a shortfall near 100 GW. A Duke University team led by Tyler Norris calculates that accepting 0.25% annual curtailment could let the existing grid absorb up to 76 GW of new load, rising past 100 GW with deeper flexibility — with the curtailment assumption and method disclosed in published work.

Capacity estimates for flexible load on the existing US grid, 2026

Spread the range across four studies and the honest summary is a factor of ten, not a factor of two. The GridCARE figure sits at the optimistic end of a defensible band, and its own floor — the 100 GW case — is close to what Duke reaches under published, conservative assumptions.

Who carries the flex

GridCARE's framing is that flexibility is a qualification problem, not a physics problem. A load that can curtail when the grid is strained is valuable; what is missing is proof, under a utility's own reliability standards, of how much flexibility a site needs, when, and how fast. The company supplies that proof, the utility verifies it in its own planning tools, and GridCARE's operations layer then dispatches within the agreed envelope.

The dependency runs the other way too. Without a defined qualification, the site's curtailment obligations stay open-ended, which makes the revenue stream hard to finance. That is a real problem for the data centre developer.

The risk sits with everyone else. Headroom on a shared network changes hour by hour with weather, demand and equipment outages. When a qualified flexible load is asked to perform, the question is not whether GridCARE's model found capacity. It is who absorbs the consequence when a customer fails to flex, or when two adjacent sites both curtail less than promised on the same hot afternoon. GridCARE's own release concedes it does not increase electricity costs "by a single cent", which is a claim about rate design, not about physics.

The supply-side race

Demand-side qualification is only half the queue. The other half is generation interconnection, and it is being attacked with the same software logic from three directions at once.

In September, AWS introduced Agentic Grid Planning, giving utilities managed AI agents that run interconnection study workflows on the same physics-based simulation software and engineering standards those utilities already use. AWS was careful to attach no gigawatt figure to it. Tapestry, the Google-affiliated moonshot, is running PJM's queue through HyperQ — 811 generation projects totalling 220 GW, reviewed in under an hour. And on June 18, FERC opened docket RM26-4, directing regional transmission organisations to stand up faster study queues for hyperscale compute above a defined threshold with firm offtake attached.

FERC's order reorders who gets studied first. AWS makes each individual study faster. Neither changes how long a transformer takes to manufacture. Stanford's researchers put new high-voltage lines at eight to twelve years to build, with large power transformers already on multi-year lead times — and they found transformer assets under peak stress at 28–68% loading, clustered near 45%, which is materially tighter than the transmission lines.

Software compresses the queue. It does not pour concrete.

Who wrote the study the Index leans on

The Index rests on a December 2025 Stanford study finding that Western utilities used 18–52% of transmission capacity even at peak demand. The senior author is Liang Min, managing director of Stanford's Bits & Watts Initiative. The PhD candidate on the study, Rajanie Prabha, is advised by Ram Rajagopal.

Both are co-founders of GridCARE. So is Arun Majumdar. GridCARE emerged from stealth in May 2025 out of Stanford's Sustainability Accelerator, and its launch materials list all four names together.

This is disclosed, not concealed — the study is publicly attributed and the company relationship is public. It is also the kind of structure that deserves a reader's attention. The research establishing that American grid headroom is large was produced partly by the people now selling software to find it.

Min's own framing is more conservative than the Index. Better use of existing transmission "could go a long way in limiting electricity rate increases for existing customers and in keeping the grid reliable until the new transmission lines can be built". Until. The study positions headroom as a bridge, not a substitute.

Will the existing grid absorb the AI buildout without new transmission?

Development scenarios