Google Just Turned a 100-Megawatt Bet Into a Full Gigawatt of Grid Capacity. It Didn't Build a Single Power Plant to Do It.

If you own a smart thermostat or a home battery somewhere in the 13 states PJM Interconnection runs, from Virginia to Illinois, Google now has a financial stake in how you use it. Not metaphorically. On October 22nd, Google and the demand-response company Voltus announced they've expanded a deal that pays homes and businesses across PJM's territory to cut or shift their electricity use during grid stress, scaling it from 100 megawatts to a full gigawatt. Google is funding it. The companies call the arrangement Bring Your Own Capacity, and the pitch is that instead of building a gas plant or waiting years in an interconnection line for new generation, a hyperscaler can pay to make existing demand more flexible and count that flexibility as capacity of its own.

The mechanics are straightforward once you see them. Voltus already runs a platform that aggregates flexible electricity use, batteries, smart thermostats, industrial processes that can pause for an hour, from homes and businesses across PJM's grid. Under the deal, Google pays for that aggregated capacity the way it might pay a developer for new generation, and the enrolled households and businesses get paid in turn, as bill credits or direct payments, when Voltus calls on them. When the grid is strained, usually a handful of the hottest summer afternoons a year, their devices cut back in concert: a battery discharges instead of drawing, a thermostat drifts a couple of degrees warmer for an hour, a factory floor shifts a process it doesn't need to run right that minute. Google gets to say it's standing behind a gigawatt of capacity. PJM gets a resource it didn't have to permit, finance, or build, and the households and businesses that opted in get paid for something most of them were never using that power for anyway.

This isn't Google's first run at it. The two companies signed their original three-year deal on June 2nd, covering up to 100 megawatts a year, which Voltus and Google both described at the time as the first commercial agreement of its kind between a hyperscaler and a virtual-power-plant operator in a U.S. wholesale market. Four and a half months later, they're multiplying it by ten. That's a fast scale-up for a resource with no published performance data yet. Neither company has disclosed how many megawatts the 100-megawatt phase actually delivered during a real stress event, or how reliably enrolled thermostats and batteries responded when called. Google and Voltus are asking PJM, and everyone paying a PJM-region electric bill, to trust that a gigawatt of distributed flexibility will show up when it's needed, before either company has published proof that even a tenth of that amount did.

The problem this is actually trying to solve

The reason this deal exists traces back to a bottleneck we've written about before. New generation in PJM now waits an average of roughly 40 months to connect to the grid, more than three times the eight-to-eleven-month target federal regulators would like to see. That wait is a big part of why PJM's last four capacity auctions added a combined $29.4 billion in costs tied to data center load, by the market monitor's own count, and why the 2026/27 auction cleared at $329.17 per megawatt-day, the price cap FERC approved after Pennsylvania's governor complained the prior auction had run away. A data center that wants to be online in three years can't wait in that line. That's why hyperscalers have spent 2026 buying nuclear uprates, funding new gas plants, and bankrolling solar farms, anything that adds supply. Google's Voltus deal is a different bet: instead of adding supply, it pays to make existing demand bend, on the theory that a flexible megawatt delivered at 4 p.m. on a July afternoon is worth about as much to the grid as a new one built from scratch, and a lot faster to put in place.

That bet is still a lonely one. Of the hyperscalers racing to secure power for AI data centers this year, Google remains the only one with a Bring Your Own Capacity contract with Voltus, and generally the only one treating demand flexibility as a core part of its capacity strategy rather than a side experiment. Microsoft, Amazon, and Meta have all spent 2026 on the supply side instead: new gas turbines, nuclear restarts, solar power-purchase agreements, the kind of deals that show up as ribbon-cuttings and press releases, and the kind this blog has covered in Georgia, Arkansas, Wisconsin, and Louisiana this year alone. There's a reason beyond preference. A virtual power plant built from thousands of individually small, individually unreliable devices is a harder thing to guarantee than a power plant with a single meter and a single operator answering for it. A gas turbine either runs or it doesn't, and either way one company is accountable for the answer. A thermostat in someone's kitchen might not respond at all, for reasons that have nothing to do with the contract Google signed. Google is making a real bet that the aggregation problem is solvable at scale, with enough devices enrolled that the ones which don't respond get averaged out by the ones that do. It hasn't yet shown the grid proof that it is.

The regulatory ceiling nobody built

The timing lands in an interesting spot. Just two weeks before this announcement, on October 9th, FERC rejected the mandatory version of this same idea when PJM tried to build it into its own tariff. PJM's plan would have let new data centers connect early in exchange for being curtailed first in an emergency, and would have let loads that brought their own new generation skip the arrangement entirely, the grid operator's own version of Bring Your Own Capacity. FERC killed the curtailment framework on jurisdictional grounds, ruling that deciding which customers go dark first belongs to the states PJM operates in, not to PJM's tariff. What survived was a large-load registry and a 50-megawatt definition, nothing that requires any company to do what Google just volunteered to do on its own.

Put those two stories together and the shape of the problem gets clearer. The grid operator tried to build a structural fix and got told that wasn't its call to make. One hyperscaler built a voluntary one instead, at a scale that's still a rounding error against the roughly 30 gigawatts of new data-center demand PJM expects by 2030. Both things can be true at once: Google's deal is a genuine addition to PJM's flexible capacity, funded with real money and aimed at a real bottleneck, and it comes nowhere close to solving the underlying crunch on its own. A gigawatt of promised flexibility is not the same as a gigawatt delivered, and nobody but Google and Voltus has that number yet.

I think the honest way to read this deal is as a bet, not a victory lap, and Google's own announcement doesn't claim otherwise. The real test isn't the contract size. It's next summer, when PJM has its next run of 95-degree afternoons and actually calls on this gigawatt to show up. If it does, at anything close to the scale promised, this becomes a template other hyperscalers have real reason to copy, and a genuine answer to the interconnection queue that doesn't require anyone to wait 40 months or build a gas plant nobody wants next door. If it doesn't, we'll find out the same way GridPermit's own evidence ladder would flag it: as a claim that sounded good in a press release and never got checked against an actual stress test. Right now, it's still just the first kind of fact. Whether it becomes the second is worth watching closer than the next announcement.