In the United States, where data center power demand is surging, new transmission line construction now faces waits of five to seven years. But the root cause of the power shortage isn't necessarily limited to delays in building transmission lines. According to GridCARE CEO Amit Narayan, the US grid, designed conservatively to withstand sudden simultaneous double failures, operates at only about 30% of its physical capacity for most of the year. California-based startup GridCARE has set its sights on this dormant capacity using AI, and has already extracted concrete connection capacity figures of 650MW with National Grid and over 400MW with Portland General Electric. Yet the same Narayan has also mentioned a figure of an entirely different order—300GW—within three to five years. Between the track record and this forecast lies a gap of more than two orders of magnitude.
Why the grid runs at just 30% capacity for most of the year
Grid design is set based on worst-case scenarios. US grid operators adopt a standard called "N-2," planning to always maintain enough reserve capacity so that even if any two transmission lines or transformers fail simultaneously, no blackout occurs. How much electricity flows under normal conditions has almost no bearing on this standard. The idea of building the entire grid with margin to spare for rare events—like lightning strikes coinciding with equipment trouble—has long been the norm.
As a result of this design philosophy, the grid operates at only about 30-40% of physical capacity for most of the year. This is according to GridCARE CEO Amit Narayan himself (as reported by Data Center Dynamics). Peak demand from extreme heat or cold occurs for only a few dozen hours per year; for the rest of the time, transmission lines and transformers operate at levels far below their true capability. This margin doesn't show up in ledger-based rated capacity figures. Because it fluctuates moment to moment with temperature, wind direction, and actual power flow (which routes electricity takes and in what amounts), static capacity tables alone can't capture it—and grid operators have until now simply used conservative figures when reviewing new connection requests.
Amid a data center development rush that has created five-to-seven-year waiting lists for new transmission line construction, a startup has emerged aiming to make this invisible margin visible through real-time simulation. That's GridCARE, founded in 2024 and based in California. CEO Amit Narayan states, "We use AI to deploy AI faster by turning hidden grid capacity into a fast lane, compressing years into months." The company explains that its AI software "Energize" calculates grid scenarios incorporating weather conditions and demand fluctuations, thereby uncovering actual connectable headroom in grids that appear full on paper.
GridCARE's proven track record with National Grid and PGE
Clear evidence that GridCARE's approach isn't merely theoretical comes from its partnership with East Coast utility National Grid. The company announced it used GridCARE's technology on New York state's grid to newly identify 650MW of connection capacity for large, flexible loads. National Grid CEO Zoe Yujnovich states, "GridCARE identified 650 MW of connection capacity on our network in New York for large, flexible loads."
To grasp the scale of 650MW, it helps to compare it against a major project also planned in New York state. Micron is planning a $100 billion semiconductor plant near Syracuse, projected to consume up to 16.17TWh annually by 2043. This translates to roughly 1850MW of continuous power on average—meaning a customer nearly three times the scale of the 650MW GridCARE identified is already waiting in the wings within the same state.
On the West Coast, Portland General Electric (PGE) is pursuing a similar effort in Hillsboro, Oregon. PGE has set a plan to secure over 400MW of connection capacity by 2029, with 80MW of that expected to come online during 2026. While National Grid's 650MW represents already-identified capacity, PGE's plan differs in that it lays out a phased timeline through to activation. What's common to both cases is that hundreds of megawatts of connection headroom were created not by laying a single new transmission line, but simply by changing how the existing grid is operated.
This track record is also reflected in GridCARE's fundraising. On May 14, 2026, the company announced it had raised $64 million (approximately ¥10.4 billion at ¥162.4/USD as of July 9, 2026) in a Series A round led by Sutter Hill Ventures, with participation from Kleiner Perkins, National Grid Partners, and Emerson Collective. This brings total funding raised to $77.5 million. The fact that National Grid itself is listed among the investors, via National Grid Partners, underscores how serious this partnership is. The company also states it has generated over $10 billion (roughly ¥1.624 trillion at the same rate) in economic value to date, though this figure is self-reported and cannot be confirmed as independently verified by National Grid or any other partner or third-party organization.
How well-substantiated is the "300GW" claim?
The figure "300GW" has begun taking on a life of its own. GridCARE CEO Amit Narayan appeared on the energy-focused podcast "Energy Empire" (hosted by Jigar Shah) and reportedly indicated an outlook that roughly 300GW of capacity could be unlocked across the entire US grid within three to five years. It's a number whose sheer magnitude makes for an easy headline.
What GridCARE has actually announced as identified capacity totals just over 1050MW—650MW with National Grid plus over 400MW with PGE combined. The 300GW (300,000MW) the CEO speaks of represents a figure 285 times larger than this proven total. The company explains that it derives this idle capacity by simulating grid operation scenarios on the order of "quadrillions."
Bank of America (BofA) estimates that between 2026 and 2030, US demand of 230GW could be met by only 93GW of supply, potentially resulting in a shortfall of roughly 100GW. Meanwhile, a research team led by Tyler Norris of Duke University calculated that if large customers accept demand curtailment of just 0.25% annually, the existing grid alone could absorb up to 76GW of new load—and with greater flexibility, this could expand to over 100GW. If GridCARE's 300GW were to materialize as stated, it would fill nearly three times the shortfall BofA warns of. Set alongside these external estimates, the anomalous scale of the 300GW figure stands out.
But there's a key difference: Duke University's estimate discloses its curtailment-rate assumptions and calculation methodology in published paper form, whereas GridCARE's 300GW—described as derived from simulations on the order of "quadrillions"—has no publicly disclosed calculation basis or assumptions, leaving nothing verifiable beyond the CEO's statements. Even when discussing the same concept of "hidden capacity," the level of transparency is worlds apart. For grid operators and regulators to judge whether 300GW is achievable, disclosure of this underlying basis is the essential first step.
GridCARE and Google-affiliated Tapestry are driving a broader trend in grid optimization
GridCARE's efforts are part of a larger trend. As surging data center demand has pushed connection queues for new generation facilities and large loads out to five to seven years, an approach has rapidly emerged in 2025-2026 that optimizes operations through software to speed up connections, rather than physically reinforcing existing infrastructure. The shared goal is to compress the process from application to connection—from a matter of years down to months.
A leading example is the effort by PJM Interconnection, which operates one of the largest wholesale electricity markets in the US. According to Data Center Dynamics, PJM introduced software called "HyperQ," developed by Google-affiliated Tapestry, and used it to review connection applications for 811 generation projects totaling 220GW in under an hour. This achievement of 811 projects and 220GW belongs to Tapestry, and both the developer and the target differ from GridCARE. What GridCARE handles is the discovery of connection capacity for large loads such as data centers—that is, optimization on the demand side. Tapestry's "HyperQ," meanwhile, is a mechanism for the supply side, accelerating connection reviews for generation facilities.
Even so, both are heading in the same direction. Rather than building new grid infrastructure from scratch, both aim to compress the wait time from application to connection by reading the existing grid as it truly is. PJM's processing of 811 projects and 220GW demonstrates supply-side review speed with concrete numbers, while GridCARE's 650MW with National Grid and 400MW-plus with PGE demonstrate demand-side connections with concrete numbers—both sharing this common thread.
The next focal point, one that also resonates with Japan's data center power challenges
Several questions remain that even GridCARE itself has not clarified. When the reserve margin, long maintained under the conservative N-2 assumption, is whittled down through software, it's unclear from published materials how much resilience against rare events—such as multiple simultaneous failures of transmission lines or transformers—would still be preserved. It also remains unclear how rate structures and regulators such as FERC (the Federal Energy Regulatory Commission) will come to view such methods.
The direction this movement points to also overlaps with Japan's power situation. Domestically, major utilities such as TEPCO and Kansai Electric Power are also being forced to respond to a rush of new data center construction for generative AI. What the US case demonstrates is that the options for grid expansion are not limited to laying new transmission lines—they also extend into the operational realm of how precisely existing equipment can be utilized. How concretely Japan's power companies can present figures for their own grids, similar to what National Grid and PGE have publicly disclosed, will determine whether this approach can also take hold domestically.
If other major utilities besides National Grid follow PJM's precedent of shrinking review times to under an hour using Tapestry's "HyperQ," the connection queues currently stretching five to seven years could potentially shrink to a matter of months. The next hurdle for GridCARE will be whether it can expand adoption of Energize to grid operators beyond National Grid and PGE, and disclose the basis for its 300GW figure in a form that can actually be verified.