Two weeks ago, this desk argued that the “bring your own power” question — turbines, SMRs, and fuel cells at the data center fence — was really a question about timelines: how long a hyperscaler is willing to wait for the grid. On October 22, Google offered a different answer entirely. Instead of generating power at the fence, it will buy flexibility across the grid: an expanded “Bring Your Own Capacity” agreement with Voltus to assemble a 1-gigawatt virtual power plant in PJM, announced Thursday morning in a joint release, adding 1,070 megawatts on top of the 100 megawatts the two companies contracted in June, under a deal that runs through 2040.
The premise is disarmingly simple. When PJM is under strain, Voltus signals enrolled sites — homes and businesses across the region — which respond with batteries, smart thermostats, industrial process adjustments, and other flexible assets to reduce energy use. Participants get bill savings or direct payments. Google, which funds the whole arrangement, gets contracted capacity to offset load from its data centers in PJM. The demand side becomes supply.
Two cautions belong up front, because they shape everything that follows. First, the headline arithmetic does not quite add up, and the discrepancy is instructive. The release says June’s 100 megawatts plus “an additional 1,070 MW” constitutes “a 1 gigawatt (GW) virtual power plant.” One hundred plus one thousand seventy is one thousand one hundred seventy — 1.17 gigawatts, not one. The dailyenergy’s trade coverage puts the total at roughly 1,170 MW. Google, for its part, calls the expansion tenfold. The numbers reconcile if you read “1 GW” as a round label for a 1.17-GW portfolio, but it is worth noting when a company puts a gigawatt in the headline: the actual contracted figure, per the release itself, is 1,070 additional megawatts.
Second, this is the counterparty’s press release, issued by Voltus, and the superlative — “one of the largest known distributed capacity portfolios deployed by a hyperscaler to date” — comes from the two parties to the contract. It is a first-hand account of the deal but not an independent one. Everything below takes the figures as filed and treats the framing as what it is: a sales document for a novel commercial structure, which is precisely what makes it interesting.
What “bring your own capacity” actually is
Strip away the branding and the mechanics are familiar. This is demand response with a hyperscaler paying the enrollment bill. Voltus, a San Francisco-based distributed energy resource platform founded in 2016, operates as the aggregator: it enrolls thousands of businesses and homes, dispatches their flexible assets into electricity markets when the grid needs relief, and shares the resulting value with participants. Google’s role is the funder — it pays Voltus to build and operate the portfolio, and in return Google can point to contracted flexible capacity backing its operations in PJM, which the release notes serves 67 million customers across 13 states and the District of Columbia.
The framework’s origin is the first clue to what this deal really is. Voltus launched BYOC last year, and per Latitude Media’s coverage, fast-tracked it when PJM’s record-breaking capacity prices — the $325-per-megawatt-day clearing price in the 2028/29 Base Residual Auction, at FERC’s approved cap — made flexibility worth a great deal of money. Google and Voltus were simply the first to formalize that logic into a bilateral contract: the hyperscaler funds the aggregation, the aggregator delivers the megawatts, and both sides treat the portfolio as capacity.
The release is explicit about the asset mix: batteries, smart thermostats, and industrial process adjustments, plus “other flexible assets.” Voltus CEO Dana Guernsey told Latitude Media that the 1,070-megawatt figure was not chosen for roundness — it is, she said, the result of an assessment of where Voltus could actually sign assets, the size of the market, and “most importantly, the real-world potential of batteries.” That last clause is the tell. Batteries are the anchor of this portfolio, the resource Voltus is most confident it can enroll at scale. Thermostats and industrial curtailment fill out the mix, but the gigawatt rests disproportionately on battery discharge — a two-to-four-hour product, typically, which matters when you start comparing it to generation.
The term is the story
The most consequential number in the announcement is not 1,070. It is 2040.
The June agreement was a three-year, 100-megawatt deal. That short duration drew exactly the questions Guernsey described to Latitude Media: “will it perform?” and “why is it only three years?” The expanded agreement answers both, in structure if not yet in evidence. A contract running through 2040 — roughly fourteen years from signing — turns a pilot into a durable commercial instrument: long enough for Voltus to invest in enrollment infrastructure, long enough for participants to count on the revenue, and long enough for Google to treat the portfolio as a planning resource rather than an experiment.
It is also a deliberate echo of Google’s earlier market-making. Guernsey framed the deal as designed to “catalyze the market, offering a durable commercial structure others can follow — much like Google has done with the creation of the clean transition tariff for emerging forms of clean firm power.” The pattern is consistent: Google uses its procurement muscle to create a template — a tariff, a contract form, a price signal — that other buyers and sellers can copy. Whether the template takes hold this time depends on whether the first installation works.
On that question, Guernsey offered one early data point: the 100-megawatt phase-one VPP, developed over the summer, has “already been fulfilled and was over-subscribed.” Treat that as the counterparty’s claim — Voltus has every reason to report strong enrollment — but if true, it suggests the larger ramp, which Voltus expects to build out over three to four years, is an enrollment execution problem rather than a demand problem. People will sign up; the question is how fast Voltus can find and wire them.
Capacity is capacity, if it performs
Here is the angle that matters for the grid, and the reason this desk is treating the announcement as more than a corporate sustainability press release. In capacity markets, the demand side has always been eligible in principle — demand response clears in PJM auctions. But it has historically been small, short-tenor, and treated as a second-class resource relative to steel in the ground. A hyperscaler signing a fourteen-year contract for over a gigawatt of flexible capacity is a different proposition: it prices flexibility like generation, with generation-like duration.
Do the comparison the market would do. PJM’s 2028/29 Base Residual Auction cleared 138,318 megawatts of unforced capacity at $325 per megawatt-day and still came up 6,831 megawatts short of the reliability requirement — the shortfall that prompted the Reliability Backstop Procurement, now suspended at FERC. Set the monitor’s corrected figure of about 3,055 megawatts against Voltus’s 1,070. The portfolio, at full build-out, would be roughly a sixth of PJM’s claimed gap, or a third of the monitor’s corrected one — if it were unforced capacity accredited the same way, which it is not. Nameplate megawatts of demand response are not unforced megawatts of generation; accreditation, dispatch duration, and performance during scarcity events are where the comparison gets made or broken.
And that is the honest limitation. A battery discharging for two hours is not a combined-cycle plant running through a three-day heat wave. PJM’s tightest hours are sustained peaks, and a portfolio anchored in batteries and thermostats has a duration ceiling that generation does not. The release calls it “flexible power capacity,” not firm capacity — language chosen carefully. What Google is buying is best understood as a hedge on the peakiest hours and a down payment on reliability, not a one-for-one substitute for the steel PJM’s backstop was meant to procure.
But there is a mirror-image error worth avoiding: dismissing it as greenwashing. One thousand seventy megawatts of contracted, dispatchable flexibility, enrolled over three to four years and held through 2040, is a real addition to PJM’s resource mix — and on a timeline that compares favorably with the alternatives: eighteen to thirty months for gas turbines, the better part of a decade for an SMR, three to four years for this portfolio. In a region where the capacity auction keeps missing its reliability requirement, a resource that can be built in the time it takes to enroll customers is the fastest firm-adjacent capacity available in PJM.
From phantom load to flex load
The timing of the announcement carries its own commentary. October 22 was also the day of PJM’s special Members Committee meeting, convened to consult members on the substance of the revised backstop proposal PJM intends to file by October 29 — the refile that will decide whether the procurement target reflects the monitor’s 4,016.9 megawatts of canceled and delayed data center load. On the same day PJM’s members debated how to count load that may never arrive, Google contracted to create capacity that exists only when called upon.
The juxtaposition is the story of this desk’s last two weeks, compressed into a single day. The October 14 piece documented load that was forecast, planned for, and then never showed up — phantom load that inflated the capacity gap and the case for emergency procurement. This deal is the inverse proposition: flexibility that is contracted, enrolled, and dispatched — load that can be made to vanish on command, which is to say, load that functions as supply. One is an artifact of bad forecasting. The other is a product you can buy, with a contract through 2040.
There is a real question underneath the symmetry: whether PJM’s capacity construct will give this kind of resource full credit. The October 9 FERC order in docket ER26-3515 accepted PJM’s Large Load Registry but rejected its Bring Your Own New Capacity rules and curtailment provisions — the commission found the curtailment scheme reached into retail rate regulation. The regulatory posture toward large loads bringing their own capacity is, at best, unsettled. A bilateral contract like Google’s does not need FERC’s permission to exist, but its value to the grid — and Google’s ability to claim reliability credit for it — will ultimately be adjudicated in PJM’s market rules and accreditation standards.
The economics, such as they are
Neither party disclosed the price. That is standard for a bilateral, and this desk’s standing rule applies: PPA prices — and by extension, capacity contract prices — are never inferred unless a primary source states them. What can be said is structural. Voltus raised $225 million in an oversubscribed Series D on October 8, led by Generation Investment Management with co-leads Activate Capital and Vitol, and says it will grow its platform to 20 gigawatts by 2030 — the war chest for exactly this kind of enrollment build-out.
The Brattle Group figure quoted in the release — more than $100 billion in U.S. consumer savings over the next decade through better grid utilization — belongs in the category of consultancy estimates deployed in press releases. Note it, attribute it, and move on.
Wood Mackenzie’s context is more useful, and more sobering. The consultancy’s estimate puts total North American virtual power plant capacity at about 37.5 gigawatts across roughly 2,000 VPPs. Google’s 1.17 gigawatts — call it one — would be nearly 3 percent of the continent’s existing total, contracted by a single buyer. That is a striking concentration for a resource class that is supposed to be distributed by definition. It is also a measure of how small the VPP world still is relative to the need: PJM alone was 6,831 megawatts short in one auction.
What to watch
Four things will determine whether this is a template or a trophy. The first is enrollment: whether Voltus actually signs 1,070 megawatts of batteries, thermostats, and industrial flexibility over three to four years, and whether the phase-one oversubscription claim holds up at ten times the scale. The second is performance: the first real scarcity event in PJM with this portfolio enrolled will be the test Guernsey’s “will it perform?” question was really about. Batteries discharge, thermostats cycle, industrial processes curtail — until the hour they don’t, and then the accreditation conversation changes.
The third is PJM’s market treatment. The backstop refile due October 29, the paper hearing on cost allocation, and the December auction for 2029/30 will all, in different ways, decide what a gigawatt of contracted flexibility is worth in the region’s reliability math. If the construct credits it, other hyperscalers will copy the structure — which is, by Guernsey’s own account, the point.
The fourth is the arithmetic the release declined to do: the cost. Fourteen years of contracted payments for 1,070 megawatts is a large number, whatever the per-megawatt price, and Google is betting it is smaller than the alternative — building generation, paying capacity prices, or waiting. That bet is the purest statement of the thesis: that the cheapest new capacity in PJM is not a power plant at all. It is a million small decisions to use less, coordinated, contracted, and paid for like the real thing. The grid has always known the demand side exists. This week, for the first time at this scale and duration, somebody wrote it a fourteen-year check.