Google Funds a 100-Megawatt Voltus Virtual Power Plant in PJM to Cover Data Center Demand

Google has agreed to pay for a virtual power plant it will never own.

On June 2 the company signed a three-year agreement with Voltus to fund up to 100 megawatts of distributed capacity per year inside PJM, the largest grid operator in the United States. Voltus will aggregate the capacity from batteries, electric vehicles, smart thermostats, and other flexible assets across residential, commercial, and industrial customers, then deliver it to a load-serving entity as accredited capacity. Google supplies the money. The participating customers collect the payments. Google keeps none of the hardware.

It is the first time a hyperscaler has directly financed a virtual power plant. The megawatt figure is modest against Google’s load. The structure is the part worth reading twice.

The contract, not the capacity. Voltus launched its Bring Your Own Capacity product in September 2025, pitched at large buyers who wanted to underwrite distributed capacity in a specific region the way they would sign a power purchase agreement for a solar farm. Google is the first name on it.

The obstacle such a deal has always faced is not engineering. It is accreditation risk. Aggregated distributed resources must be certified by the grid operator before they count as capacity, and that certification can be revised downward if the fleet underperforms. A revenue stream that the grid can mark down at will is difficult to put on a corporate balance sheet.

Caroline Golin, who led global energy and market development at Google, said Voltus was the only provider in the region willing to take that exposure off Google’s books. Voltus, she said, was “up for absorbing the accreditation risk” and “able to provide a contract that looked like every other PPA.” That is the whole transaction in one sentence. The thing Google bought is not electrons. It is a financial instrument shaped like one the company already knows how to sign.

Why pay someone else to flex. Google has spent two years making its own data centers more flexible and has said it is pursuing roughly a gigawatt of demand response across US power systems through utility agreements. The Voltus deal runs the other direction. Rather than curtail its own load, Google is paying other people to curtail theirs.

The reason is capital efficiency. A data center is, in the words of Google’s head of data center energy, “billions and billions of dollars of hardware that only gets utilized if it’s running.” Idling that hardware during grid peaks strands the most expensive asset on the site. Paying a warehouse battery or a neighborhood of thermostats to stand down for the same few hours costs less per avoided kilowatt. The arbitrage runs between Google’s cost of idle capital and everyone else’s cost of flexibility, and today everyone else is cheaper.

What a battery is worth. To see why this matters beyond one data center, it helps to look at what a behind-the-meter battery actually earns. The value arrives in layers. At the base is demand-charge reduction, the customer shaving its own monthly peak, a saving it controls and can count on regardless of what the grid does. Above that sit tax credits and state incentives, the investment tax credit, programs such as California’s SGIP and New York’s Retail Storage Incentive, one-time benefits that lower the cost of the asset. Above those sits grid-services revenue: capacity payments, demand response, the occasional ancillary product, money earned for making the battery available to the system.

The economics of commercial storage have always rested on the bottom of that stack. Bill savings and tax credits are bankable because they are either customer-controlled or paid upfront. The top layer has been treated as upside. Capacity and demand-response revenue depended on utility programs that changed every cycle, on accreditation that could be cut, and on ratepayer-funded budgets set by regulators rather than contracts. A developer could model that revenue, but few financiers would lend against it. So the standard pitch anchored on demand charges and counted grid revenue as a maybe.

The Google deal is the first clear sign that the maybe is becoming a contract. By wrapping distributed capacity in a PPA-shaped instrument and handing the accreditation risk to the aggregator, it converts the most speculative layer of the stack into something that looks like the most secure layer. That is a re-rating, not a new revenue stream. The capacity always existed. What changes is that an asset whose grid revenue is bankable is worth more, and finances more cheaply, than the same asset whose grid revenue is a hope.

Who pays for flexibility. The second shift is in the source of the money. Distributed capacity has historically been funded by ratepayers, assembled slowly through utility demand-response programs and state incentives. Here the funder is the load itself. The company whose growth is straining PJM is paying for the distributed capacity that relieves it, because doing so is cheaper than the alternatives, not because a regulator required it.

That places data center demand on both ends of the value stack at once. The same load growth that tightens reserve margins and pushes commercial demand charges higher, which is what makes the battery’s anchor saving larger, is now also the thing funding the battery’s grid-services layer. One force underwrites the bottom of the stack and the top of it.

PJM is where the arithmetic clears. Its recent capacity auctions have settled at record prices, reserve margins have narrowed, and the load forecast has been bent upward by data center construction across Virginia, Ohio, and Pennsylvania. Capacity is scarce enough, and dear enough, that funding new distributed supply now competes favorably with paying auction prices or waiting in an interconnection queue measured in years.

Where the evidence stops. Two limits sit in plain sight. The first is supply of counterparties. The deal works because Voltus could warehouse accreditation risk and write a bankable contract against it. That is a narrow capability, and it is not clear how many aggregators have the balance sheet to offer it at the scale every hyperscaler in PJM would want. Latitude Media called the model a potential unlock for virtual power plants, with the weight on potential.

The second is self-limitation. The value of capacity rests on its scarcity. If hyperscalers fund enough distributed supply, the same auction prices that justify the spending will soften, and the blueprint will begin to compete with itself. A model that scales until it erodes its own premise is still a real model. It is simply one with a ceiling built in.

The asset in the commercial building did not change this week. Its ability to earn from bill savings, from incentives, and from grid services was always there. What changed is that someone with a balance sheet and a reason finally made the top of that stack something a lender would recognize, and the check came from the load down the road rather than the ratepayer next door.

Sources