Google has agreed with virtual power plant operator Voltus to increase leased capacity in the PJM grid, which serves 67 million consumers in the US, tenfold: the volume has grown from 100 MW to 1 GW. Under the Bring Your Own Capacity (BYOC) scheme, the company pays homes and businesses to reduce electricity consumption during peak load hours, and the freed-up grid power goes to Google's data centers—no new power plants are built for this.

!image.jpg)

What happened

On October 22, 2026, Google and Voltus signed an expanded agreement under the BYOC program: the June contract for 100 MW, concluded for three years, grew to 1 GW in four months. The mechanism works as follows: Google pays Voltus, which aggregates consent from home and business owners for temporary consumption reduction. During peak load hours, smart thermostats briefly dim air conditioners, home batteries take over home power supply, and industrial consumers shift energy-intensive processes. No new generators are built for data centers—the grid power freed up during peaks is allocated to Google. Voltus itself already manages over 9 GW of distributed capacity nationwide, and after raising $225 million in August 2026 in Series D, it aims for a 20 GW portfolio by 2030.

Context

PJM is one of the largest US grid operators, covering 13 states plus the District of Columbia, and it is in its zone that the new agreement operates. The background to the deal is a sharpened capacity balance: peak power prices in PJM have risen more than eightfold, and by the early 2030s, an additional 30–34 GW of demand from large consumers, including data centers, is expected. Demand management, i.e., agreements where consumers temporarily reduce consumption for compensation, is an established practice with a long history: smart thermostats and aggregators of such services have long been working in the market. The novelty here is in the class of buyer: according to available sources, a hyperscaler is purchasing distributed household-scale flexibility in a volume of 1 GW for the first time as a replacement for generation for its data centers. However, load reduction is not equivalent to a full power plant: such capacity has limits regarding event duration, frequency of occurrence, and correlation with weather.

Why this matters for the industry

The deal changes the economics of connecting data centers to grids: peak power becomes a scarce resource, and hyperscalers begin to close its deficit not by building generation, but by purchasing distributed flexible load through virtual power plant operators. The tenfold expansion of the contract in a few months shows that the architecture of "consent collection, resource automation, payments" is perceived as a proven channel, not a pilot: what is being sold is not generation or hardware, but the orchestration of distributed resources. If the 1 GW plan is executed, similar BYOC deals are likely among other hyperscalers and in other grids, and the purchase of flexible power risks becoming a standard line item in data center placement planning. A practical class of tasks is maturing for ML teams: energy-aware scheduling, shifting training and inference in time, elastic inference, and co-optimization of computations with price and grid availability. The growing flexibility market creates demand for integration and telemetry layers that combine household devices into a managed fleet.

Why this matters for users

The practical effect is currently local: payments or discounts on electricity bills are available to residents and businesses in the PJM zone—this is 13 US states and Washington. Participation means that during peak load hours, the participant's air conditioners, batteries, or production processes are briefly adjusted, and the aggregator compensates for this with money. For developers and product builders, the contract does not change anything in their own products today: this is a grid agreement, not an API, and it does not provide new computational or software tools. Outside the PJM region, there are no available actions yet: participation is limited by the program's geography, and no public API has been presented in published materials.

What is currently unknown / limitations

There are no public metrics of actual triggers for the 1 GW contract. The portfolio of over 9 GW under Voltus management and the 20 GW target are commercial indicators of collected agreements, not measured reliability of power supply: sources do not disclose the verification methodology, including the calculation of the consumption baseline and the share of devices actually participating in outages. Load reduction is not equivalent to firm generation: it is limited to events lasting hours, not days, the allowable frequency of interventions, and dependence on weather. The rapid growth of the target from 100 MW to 1 GW indirectly indicates that the pilot provided data on the scheme's operability, but this is an interpretation: the corresponding figures were not published. The sizes of payments to specific households, as well as the degree of their actual participation in events, are also not disclosed.

Sources

Author

Look at AI, editorial team