Google signed a deal with Voltus to finance a 100-megawatt virtual power plant on the largest US grid, the companies said Tuesday. The project, set to operate by 2027, will pay homes and businesses to curb electricity use during peak demand, freeing capacity for Google’s data centers. It marks one of the most concrete efforts yet by a tech giant to use distributed flexibility to meet surging AI energy needs.
Voltus will aggregate electric vehicles, smart thermostats, and other distributed resources across the PJM Interconnection, which serves 65 million people from New Jersey to Illinois. The company will pay participants to dial back consumption or discharge stored energy when the grid is stressed. Google foots the bill for setup, and the extra capacity helps run its data centers in the region.
The program, called “Bring your own capacity,” was unveiled by Voltus in September. Google is the first named customer. “There is no one solution for expanding grid capacity and we’re continuing to explore all options, including the many avenues for load flexibility,” Michael Terrell, Google’s global head of advanced energy, told MIT Technology Review in an emailed statement. Voltus expects to sign up enough devices each year to deliver 100 megawatts of flexible demand.
That is roughly the output of a small gas-fired power plant. The virtual plant should be operational in 2027, according to Voltus. This is not Google’s first flexibility play.
The company already has agreements with utilities across the US to shift or curtail its own data center loads. But as Google noted in a blog post earlier this year, not every facility can ramp down on command. AI inference workloads, which serve real-time user queries, are far less flexible than model training.
A Duke University study last year crystallized the opportunity. It found that if data centers agreed to cut demand for just 40 hours per year, about 100 gigawatts of new load could connect without building new power plants or transmission lines. The math is simple.
Grids are sized for peak demand—the hottest summer evening when air conditioners, TVs, and appliances all run at once. If data centers step aside during those few hours, the system can handle them the rest of the year. The Duke finding ignited a wave of interest in data center flexibility.
But it also exposed a hard question: Who pays for the cuts? Data centers lose revenue when they throttle computing. Regulation offers one path.
A US proposal would let new data centers jump the interconnection queue if they commit to lowering demand during grid emergencies. A new Texas law already requires large users to switch to backup power or curtail in a crisis. Google’s Voltus deal represents a different path.
Instead of cutting its own load, Google pays others to cut theirs. It is a market-based answer to the flexibility puzzle. But will people sign up?
A recent California study on managed electric-vehicle charging offers a cautionary signal. With no payment, only 1% of EV owners enrolled in a program that cedes control of when their car charges. At $40 per month—about 15% of the average power bill—participation rose to just 4.6%.
The study, from Lawrence Berkeley National Laboratory, did not test higher incentives. But the numbers suggest money alone may not overcome consumer reluctance. Voltus and Google have not disclosed how much they will pay participants.
That figure will be decisive. Without a compelling incentive, the virtual plant could struggle to recruit. The political climate adds another layer of friction.
Gallup polling this year found that roughly 70% of Americans oppose AI data centers in their area. Asking those same communities to tweak their thermostats to help a Google server farm may be a tough sell. Here is what they are not telling you.
The Voltus deal is a bet that households will trade a little control for cash. But the California EV data suggests most people value autonomy more than a modest discount. The math does not add up—at least not yet.
Follow the leverage, not the rhetoric. Google needs power. Regulators are watching.
The Voltus deal is a real-world test of whether virtual power plants can bridge the gap between AI ambitions and physical infrastructure. Behind the scenes, the PJM grid is under acute strain. A surge of data center proposals has overwhelmed the interconnection queue.
PJM has warned that retirements of coal and gas plants are outpacing new supply. In that context, 100 megawatts of flexible demand is a drop in the bucket. But if the model works, it could scale fast.
Voltus is not the only company chasing this market. Other aggregators, including AutoGrid and Leap, are pitching similar services. It turns a utility-scale problem into a neighborhood transaction.
Skeptics note that virtual power plants have been hyped for years without delivering transformative scale. The economics are improving. But consumer adoption remains the bottleneck.
The Voltus-Google project will test whether a tech brand can overcome that inertia. The timeline matters. Voltus says the plant will be operational in 2027.
That is an eternity in AI years. By then, Google’s data center fleet will have grown substantially. The company’s carbon footprint is already rising, driven by AI.
Flexible demand could help square that circle. Why It Matters:
A successful 100-megawatt virtual power plant in PJM would set a precedent for how tech giants manage the grid impacts of AI. It could accelerate interconnection for data centers, lower costs for consumers, and reduce the need for new fossil-fuel plants. If participation lags, however, it will reinforce the case for mandatory flexibility rules—a less palatable outcome for the industry.
Key Takeaways:
- Google is paying Voltus to build a 100-megawatt virtual power plant in the PJM grid, operational by 2027, to free capacity for its data centers. - Consumer participation is uncertain: a California study found only 4.6% of EV owners enrolled in managed charging even with a $40 monthly incentive. - The project could scale if successful, but faces headwinds from public opposition to AI data centers and the challenge of recruiting enough households. What Comes Next:
Voltus must now recruit thousands of participants across PJM territory. The company will need to set incentive levels high enough to attract sign-ups while keeping costs manageable for Google. Regulators at PJM and the Federal Energy Regulatory Commission will watch closely.
If the model proves replicable, expect other hyperscalers—Amazon, Microsoft, Meta—to announce similar deals within the next 18 months. If it stumbles, the policy debate will shift toward mandates, not markets.
Key Takeaways
— Google is paying Voltus to build a 100-megawatt virtual power plant in the PJM grid, operational by 2027, to free capacity for its data centers.
— The deal is the first named test of Voltus’s “Bring your own capacity” program, which lets data centers finance distributed flexibility instead of cutting their own load.
— Consumer participation is uncertain: a California study found only 4.6% of EV owners enrolled in managed charging even with a $40 monthly incentive.
— The project could scale if successful, but faces headwinds from public opposition to AI data centers and the challenge of recruiting enough households.
Source: MIT Technology Review
