GridCARE Index Claims U.S. Grid Can Absorb 120 GW of New Load

A different question about AI's power problem

Most of the argument about powering artificial intelligence has been about building: new gas turbines, new reactors, new transmission lines. GridCARE, a Redwood City startup that grew out of Stanford, is asking how much spare room already exists on the grid. On October 9 the company said it had qualified 5 gigawatts of additional capacity on the existing U.S. grid over the previous 12 months, working with six utilities11. Alongside that figure it released a Power Acceleration Index. The index estimates that today's grid could carry another 100 to 120 GW of demand, and it puts the net value to customers at $270 billion to $350 billion1118.

CEO Amit Narayan described the effort as a reframing. In his account, the industry has spent its energy calculating how much new power America must build, while GridCARE set out to measure how much is already sitting inside the grid the country has paid for11. The company unveiled the index at its Power Acceleration Summit in San Francisco on October 813.

The 120 GW figure deserves attention because of its size. Grid Strategies forecasts about 166 GW of U.S. peak load growth over five years and ties roughly 90 GW of it to data centers29. If GridCARE's estimate holds even in part, flexible use of existing wires could cover a large share of the AI buildout's near-term needs. That is the case to take seriously. The weak points are elsewhere, and they matter.

How the mechanism works

The idea itself is simple. Power lines and substations are sized for the few hours of highest demand each year, so for most of the year they run well below their limits11. A data center or factory that agrees to cut its consumption during those stress hours can connect long before upgrades sized for its worst-case draw would be finished11.

GridCARE says the missing piece has been "qualification." That means showing, under current reliability standards, how much a site must flex, when it must do so, how quickly it must respond, and for how long, across normal conditions and every outage case13. The company argues this matters to developers as much as to utilities. A data center with vague curtailment obligations has unpredictable performance and has trouble raising money13. Its platform runs what it calls quadrillions of grid scenarios covering congestion, weather, outages and demand swings, looking for capacity that standard interconnection studies miss2. GridCARE says the utility then checks the results in its own planning tools13.

The projects behind the claim

GridCARE's first and most-cited project is with Portland General Electric in Hillsboro, Oregon. There it qualified more than 400 megawatts of data center capacity. Eighty MW became available in 2026, and the rest is due to be energized by 2029 at the latest11. Trade coverage reports that Hillsboro data centers had faced a seven-year waitlist before the project16. A company executive said the work produced interconnection agreements for five separate data centers1. GridCARE also says it has qualified more than 2 GW with several other Pacific Northwest utilities13.

The other projects are earlier-stage and smaller. With National Grid in New York, the work identified up to 650 MW of potential capacity for large customers willing to accept flexible terms11. Southern California Edison studied one constrained circuit in the Inland Empire and found up to 20 MW of additional flexible capacity11. In South Carolina, Santee Cooper is still working with GridCARE to find ways to serve large loads on existing infrastructure11.

The wording varies between accounts. GridCARE's announcement says it "qualified" 650 MW with National Grid13. National Grid CEO Zoë Yujnovich, speaking to investors in May, said the companies had identified "up to" 650 MW of "potential" flexible connection capacity13. Most independent trade coverage repeated that more careful framing11. In practice, the 5 GW total combines energized megawatts, signed agreements, and capacity that has been found but not yet contracted.

Where the reporting agrees and where it pulls apart

The coverage agrees on the basic facts: the 5 GW, the six utilities, the 120 GW ceiling, and the up-to-$350 billion value estimate111516. Axios framed the news as part of a wider wave of startups trying to unlock unused grid capacity. It argued that such capacity is the fastest and most efficient route to power for hyperscalers15.

The main difference is how much each outlet qualifies the numbers. Daily Energy Insider repeatedly called the figures projections built on the index's assumptions, not guaranteed savings. It also stressed that any bill reductions depend on regulators approving tariffs11. The company's own release is more confident. It says eligible projects can connect in months "without increasing the electricity costs for everyday Americans by a single cent"13.

Even the size of the idle headroom changes with the source. GridCARE cites a December 2025 Stanford study finding that some Western utilities used as little as 18% of their transmission capacity even at peak13. Its Series A announcement cited utilization of about 30%2. Its website says the grid is less than a third utilized19. Earlier, Narayan put the figure below 40%9. These numbers measure different things in different ways, and the spread is a reminder that "idle capacity" depends on how it is defined.

Independent research points the same way

The 120 GW claim gets real support from outside the company. A Duke University Nicholas Institute analysis found that the largest U.S. balancing areas could reliably take on 76 to 126 GW of new load without building more capacity, if those loads accepted curtailment in just 0.25% to 1% of annual hours2227. Duke work cited by Utility Dive also found that trimming data center peak demand by 1% to 2% could lower electricity rates by 0.5% to 2.8%28. GridCARE's range falls inside Duke's. To us, that suggests the headline number reflects a methodology the research community recognizes, not just a marketing figure.

The operating evidence is growing as well. Boston University researchers modeled AI training and inference workloads with 18% to 55% flexibility relative to average power use while still meeting service requirements28. At Salt River Project, Emerald AI cut an AI workload's power by 25% over a three-hour peak28. Federal analysts note that NERC's summer 2026 peak forecast for ERCOT was lower than the year before, partly because more data centers can now be curtailed in emergencies26.

There is also a reason for caution. An April 2026 modeling paper found that data center flexibility can cut grid investment and operating costs by 3% to 21%. It also found that more flexibility does not always mean less new generation, and in congested networks it sometimes means more24. Using wires more fully is one problem. Having enough power plants to supply the extra energy is another, and the index addresses the first more directly than the second.

Why it matters now

Timing makes this approach attractive to the industry. Traditional interconnection takes four to six years25. Developers have answered by planning roughly 90 GW of data center capacity around their own on-site power, and only about 2% of that generation was operating as of mid-202625. Meanwhile, U.S. utilities requested $23.1 billion in electric and gas rate increases in the first three quarters of 202616. Public resistance to data centers increasingly centers on local power bills28.

GridCARE's pitch speaks to both pressures. Developers get power years sooner, and households might see savings. The company estimates bills could fall 7% to 9%, or $130 to $170 a year, if the value is passed back to customers11. The second half of that promise is the less certain part. It depends on utilities and regulators building mechanisms to return the money, and on local grid conditions supporting the assumed flexibility11.

Demand forecasts are also uncertain. NERC revised several summer 2026 projections downward because large loads were connecting more slowly than expected. One investment bank estimated that only 50% to 60% of near-term data center capacity will arrive on schedule26.

The verdict

Our view is that GridCARE has a real result and an aspirational number. Hillsboro is a working, contracted example of flexible connection at transmission scale. The Duke research makes a national range near 100 GW plausible. The 120 GW and $350 billion figures, though, are screening estimates built on standardized assumptions. The company itself says actual capacity at any site depends on detailed network studies13. GridCARE has a strong argument that existing wires are an underused resource for AI. The industry will believe it once the remaining 4-plus GW of qualified capacity turns into energized load and regulators actually deliver the promised bill credits.