The Problem Nobody Wants to Say Out Loud

AI is breaking the American power grid.

PJM, the grid operator covering 13 states from Illinois to New Jersey, has been running record-high capacity auctions for years. Reserve margins are shrinking. Load growth from data centers is aggressive and accelerating. And the pipeline of new power plants cannot keep up.

Google knows this. So instead of waiting for someone else to fix it, the company cut a deal.

What the Deal Actually Is

On June 2, 2026, Voltus and Google announced a three-year "Bring Your Own Capacity" agreement, according to a press release from Voltus via GlobeNewswire. Voltus will aggregate up to 100 megawatts of distributed energy resources — residential and commercial batteries, smart thermostats, electric vehicles — from customers across the PJM region.

Google funds the whole thing. Voltus runs it. Participating homeowners and businesses get paid.

When the grid gets stressed — think a brutal July evening with everyone blasting AC — Voltus's software automatically coordinates thousands of devices at once. A battery discharges here. A thermostat bumps up a degree there. Individual users barely notice. But combined, that's 100 MW of capacity that didn't require a single new power plant.

Why Google Is Paying Strangers Instead of Flexing Its Own Data Centers

Google has worked to make its own data centers flexible. Amanda Peterson Corio, Google's global head of data center energy, told Utility Dive directly: it's often faster and more cost effective to pay other customers to shift their usage than to curtail its own operations.

"The cost of capital in the data center, of our chips, can be billions and billions of dollars of hardware that only gets utilized to our customers if it's running," Corio said.

Google's data centers are running AI workloads that generate revenue every second they're on. Shutting them down to save grid capacity costs Google real money. Paying a homeowner $50 to let their thermostat adjust for two hours costs a fraction of that.

Michael Terrell, Google's global head of advanced energy, added that the VPP model provides a flexible mechanism to support data center load growth while contributing to grid reliability and affordability, according to TheEnergyMag.

The Scale of the Grid Problem

This 100 MW deal is a Band-Aid on a gaping wound.

A 2024 Duke University study found that if data centers agreed to cut demand for roughly 40 hours per year, approximately 100 gigawatts of new data center capacity could come online without requiring new power plants or transmission infrastructure, according to MIT Technology Review. That's 100,000 MW. The Google-Voltus deal covers 100 MW — or one-tenth of one percent of that figure.

The Brattle Group found that U.S. consumers could save more than $100 billion over the next decade by better utilizing existing grid resources through virtual power plants, according to Voltus's own press materials. That number comes from Voltus, so treat it as motivated — but even discounted, the underlying logic holds.

This Is a Market Solution

No federal mandate made this happen. No regulatory agency dictated this approach. Google looked at its energy problem, found a private company with a workable model, and wrote a check.

That's how infrastructure problems should get solved — by the people who have a financial stake in solving them, using their own capital, not taxpayer money.

The alternative is what Texas tried: the state passed a law this year requiring large power users to switch to backup power or curtail demand during grid emergencies, according to MIT Technology Review. Mandatory curtailment is one approach. Google's model — voluntary participation with direct payment — is another. One rewards people for flexibility. The other punishes them for existing.

What This Really Is

This is a capacity arbitrage play. Google has a mission-critical electricity problem. Building new gas plants or waiting for transmission upgrades takes years and billions. Aggregating existing distributed resources through software takes months and far less capital. Google is paying for reliability.

But a genuine question about scalability remains. Voltus CEO Dana Guernsey told GlobeNewswire that the Google partnership is "pioneering a model that large load customers can follow." Fine. But 100 MW of aggregated smart thermostats is NOT the same as a 1,000 MW power plant running 24/7. Virtual power plants provide capacity during peaks — they don't replace baseload generation.

Anyone claiming VPPs alone can power the AI economy is selling something.

The Reality

Google is spending its own money to solve a real problem creatively. Homeowners in the PJM region get paid for flexibility they probably weren't monetizing anyway. The grid gets breathing room. No taxpayer dollars involved.

But America's power grid is being stress-tested by AI demand in real time, and a 100 MW software-coordinated deal — however clever — is a stopgap, not a solution. The United States needs new generation capacity, and it needs it fast. That means nuclear, gas, and transmission investment at a scale that makes this Google deal look like a rounding error.

Smart thermostats can buy time. They cannot build civilization.