> ## Content Index
> Fetch the complete content index at: https://www.aixenergy.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# Google’s 100-Megawatt Capacity Deal Signals a New Phase in AI Power Procurement
- URL: https://www.aixenergy.io/googles-100-megawatt-capacity-deal-signals-a-new-phase-in-ai-power-procurement/
- Published: 2026-09-05T03:10:00.000Z
- Updated: 2026-10-22T00:59:37.000Z
- Description: The strategic change is not that a virtual power plant can provide demand response. It is that an AI company can now procure accredited capacity through a distributed portfolio, turning flexibility from a utility program into a corporate infrastructure contract.
- Author: Brandon Owens
- Tags: LoadFlex

For most of the data-center boom, the power conversation has focused on energy: how many megawatt-hours a campus will consume, which generator will supply them, and whether the resulting portfolio is renewable, firm, or merely matched on an annual basis. That framing is no longer sufficient. In constrained power systems, the scarce product is increasingly not energy but dependable capacity at the right location and time.

Article Summary

0:00

/61.759184

1×

Google’s June 2026 agreement with Voltus makes that distinction concrete. Under the three-year arrangement, Voltus will aggregate up to 100 megawatts of accredited distributed energy resources across PJM—batteries, thermostats and flexible commercial loads—into a Google-funded virtual power plant. The resources can be called when the system is stressed, and their accredited contribution can help meet the capacity requirement associated with Google’s load.

The deal is modest beside a hyperscale campus. One hundred megawatts would cover only one-tenth of the capacity associated with a one-gigawatt load, before accounting for accreditation and contractual details. Yet its commercial architecture matters more than its initial scale. Google is not simply enrolling its own data centers in a utility demand-response tariff. It is paying an aggregator to assemble capacity from thousands of other customers and directing that capacity toward an obligation created by Google’s own growth.

## Why the economics changed

PJM’s capacity market has repriced the value of reliability. The RTO-wide clearing price rose from $28.92 per megawatt-day for the 2024/25 delivery year to $269.92 for 2025/26, $329.17 for 2026/27 and $333.44 for 2027/28\. The July 2026 auction for 2028/29 cleared at the new $325 price cap. PJM also procured less capacity than its reliability requirement for the second consecutive auction, falling short by 6,831 megawatts of unforced capacity for 2028/29.

For a continuously operating 100-megawatt load, the simple gross capacity-value arithmetic rises from roughly $1.1 million a year at $28.92 per megawatt-day to $11.9 million at $325\. For a one-gigawatt campus, it approaches $119 million. These figures are not a forecast of any customer’s bill: actual exposure depends on the utility tariff, coincident load, capacity-zone rules, hedges, bilateral arrangements and the load-serving entity’s allocation method. They do show why capacity procurement has moved from a specialist concern to an executive one.

PJM capacity market

## The capacity-price signal 

| Delivery year | RTO price($/MW-day) | 100 MW annual value\* | System signal                                          |
| ------------- | ------------------- | --------------------- | ------------------------------------------------------ |
| 2024/25       | $28.92              | $1.06 million         | Reference point before the repricing                   |
| 2025/26       | $269.92             | $9.85 million         | Large step-change in reliability value                 |
| 2026/27       | $329.17             | $12.02 million        | Auction cleared at the approved cap                    |
| 2027/28       | $333.44             | $12.17 million        | First PJM auction short of its reliability requirement |
| 2028/29       | $325.00             | $11.86 million        | Cleared at the extended price cap; 6,831 MW UCAP short |

**\*Illustrative gross value** \= price × 365 days × 100 MW. This is not a customer bill estimate. Sources: [PJM 2027/28 auction results](https://www.pjm.com/-/media/DotCom/committees-groups/committees/mrc/2025/20251219-special/item-01---review-of-2027-2028-base-residual-auction-results---presentation.pdf?ref=aixenergy.io) and [PJM 2028/29 auction report](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2028-2029/2028-2029-bra-results-report.pdf?ref=aixenergy.io). 

## What the deal changes

Historically, utilities have been the primary buyers in virtual power plant programs. They recruit or contract for customer flexibility to reduce peak demand, defer infrastructure, meet capacity obligations or improve reliability. Customers participate at the edge of the system; the utility or its contractor controls the program design and captures the portfolio value.

The Google–Voltus structure introduces a different buyer. A large load can sponsor the aggregation itself, while the utility remains the load-serving entity and PJM remains the market operator. This does not remove the utility from settlement, planning or reliability responsibility. It does change the source of capital and the commercial logic. The value stream begins with a corporate buyer trying to manage its own capacity exposure, not with a utility budget trying to achieve a program target.

That shift can matter in four ways. First, creditworthy technology companies can support multi-year contracts that give aggregators more certainty to recruit customers, install controls and accept performance risk. Second, corporate procurement teams are accustomed to fixed-price infrastructure contracts; capacity can be packaged in a form resembling the power-purchase and supply agreements they already understand. Third, the arrangement creates a direct line between data-center growth and payments to households and businesses that provide flexibility. Fourth, it offers a practical interpretation of the emerging “build, bring or buy” principle: a large load can bring accredited capability to the system rather than simply asking the utility to find it.

## What it does not change

The agreement should not be mistaken for a complete solution to the data-center power constraint. Capacity, energy, transmission and distribution are distinct products. A 100-megawatt virtual power plant does not deliver 100 megawatts of continuous electricity to a campus. It does not eliminate the need for generation, network upgrades or a feasible interconnection. It creates an accredited resource that can reduce system demand or inject power during defined periods, subject to PJM rules and performance requirements.

Nor is the transaction necessarily a clean avoidance of the capacity market price. Bilateral capacity has an opportunity cost shaped by the same market. When auction prices rise, aggregators and participating customers know the accredited product is more valuable. The advantage is not free capacity; it is the ability to replace uncertain, tariff-mediated exposure with a negotiated contract and to choose how performance, term and price risk are allocated.

Scalability is the hardest question. A 100-megawatt commitment is credible; replicating it many times is difficult. Distributed resources must be recruited in the correct territory, metered, telemetered where required, available during system stress, and converted from nameplate capability into accredited unforced capacity. Customer attrition, device failure, overlapping program claims and non-performance penalties all matter. As more buyers chase the same flexible load, the supply curve will steepen. The limiting factor may become not software, but the inventory of customers willing and able to surrender control when the grid needs it most.

## A new stack of power procurement

The deeper implication is that data-center power procurement is becoming layered. The old model treated electricity service as a utility input delivered after site selection. The emerging model requires developers and hyperscalers to assemble a portfolio: land and interconnection rights; deliverable energy; accredited capacity; transmission and distribution upgrades; on-site resilience; and controllable load. Each layer has different counterparties, timelines and failure modes.

Virtual power plants fit into this stack as a capacity and flexibility instrument. They can be faster to assemble than new central generation, distribute economic benefits across many customers, and improve use of assets already connected to the grid. They may also help bridge the interval between a data center’s commissioning and the arrival of new generation or transmission. But “bridge” is the operative word. A resource built around episodic load reduction cannot substitute indefinitely for the firm supply needed by a rapidly growing, high-load-factor industry.

## The strategic test for the next deals

The Google–Voltus agreement will be important if it becomes a repeatable procurement category rather than a singular transaction. The next contracts should be judged against five questions:

- **Additionality:** Does the contract enroll new flexible capacity, finance new storage, or merely redirect resources already counted elsewhere?
- **Deliverability:** Are the resources located where they can address the capacity obligation and relevant system constraints?
- **Accreditation:** How much nameplate capability becomes recognized UCAP, and who bears changes in accreditation rules?
- **Performance risk:** Who pays when the portfolio fails during a PJM performance interval—the aggregator, the buyer, or participating customers?
- **System value:** Does the arrangement reduce total system cost and reliability risk, or mainly hedge one buyer while shifting costs to others?

Those tests distinguish genuine resource creation from financial engineering. They also matter for regulators. If large-load customers can procure capacity directly, tariff design must preserve transparent cost causation, prevent double counting and ensure that bilateral arrangements do not weaken the utility’s ability to plan the local network. A contract can be privately rational and still produce poor system outcomes if accreditation, dispatch and cost allocation are misaligned.

## The larger signal

The most important fact about Google’s deal is not the cleverness of the virtual power plant. VPPs are established technology, and demand response has participated in organized markets for years. The important fact is that the buyer has changed because the scale and economics of load have changed.

AI companies are becoming power-system actors. They are no longer only choosing utilities, signing renewable contracts and negotiating interconnection schedules. They are beginning to procure the reliability products their growth requires. That will pull aggregators, storage developers, equipment vendors, utilities and market operators into new contractual relationships—and expose the difference between announcing megawatts and delivering accredited, locationally useful capacity.

At 100 megawatts, the Google–Voltus agreement will not rebalance PJM. But it provides a working commercial template at the precise moment when PJM’s capacity auctions are clearing at their caps and below the system’s reliability requirement. If the model scales with credible additionality and performance, it could turn millions of small flexible devices into part of the infrastructure financing the AI buildout. If it does not, it will remain an elegant hedge attached to a much larger physical shortage.

The distinction will be settled not by the announcement, but by the next several hundred megawatts.

## Sources

- [Voltus and Google, “Bring Your Own Capacity” announcement, June 2, 2026](https://www.voltus.co/press/voltus-google-bring-your-own-capacity-pjm?ref=aixenergy.io)
- [PJM, 2028/2029 Base Residual Auction Report, July 14, 2026](https://www.pjm.com/-/media/DotCom/markets-ops/rpm/rpm-auction-info/2028-2029/2028-2029-bra-results-report.pdf?ref=aixenergy.io)
- [PJM, 2027/2028 Base Residual Auction results](https://www.pjm.com/-/media/DotCom/committees-groups/committees/mrc/2025/20251219-special/item-01---review-of-2027-2028-base-residual-auction-results---presentation.pdf?ref=aixenergy.io)
- [FERC, Order No. 2222 explainer on distributed energy resource aggregation](https://www.ferc.gov/ferc-order-no-2222-explainer-facilitating-participation-electricity-markets-distributed-energy?ref=aixenergy.io)
- [White House, Ratepayer Protection Pledge, March 4, 2026](https://www.whitehouse.gov/releases/2026/03/ratepayer-protection-pledge/?ref=aixenergy.io)