Google's 1,070 MW Virtual Power Plant Turns PJM's $325 Capacity Cap Into an AI Capital Line insight cover
Industry NewsGOOG · CEG · VST•20 min read

Google's 1,070 MW Virtual Power Plant Turns PJM's $325 Capacity Cap Into an AI Capital Line

Google and power aggregator Voltus announced on October 22, 2026 an expansion of their PJM bring-your-own-capacity agreement adding 1,070 MW on top of the 100 MW tranche signed in June, with the contract running through 2040 — sixteen days after Constellation Energy agreed to supply Google 3,590 MW of new and existing nuclear power in the same market. PJM's July 2026 capacity auction cleared at the $325/MW-day cap and still came 6,831 MW short of reliability, so the demand-side leg is where hyperscaler power strategy now actually differentiates: Google buys the capacity obligation itself instead of waiting in an interconnection queue. For investors the read-through is that AI power cost is migrating off utility rate base onto hyperscaler balance sheets, with the money flowing to the aggregator and the device fleet — and with a performance risk that PJM's own 67% demand-response dispatch record quantifies.

Published Oct 23, 2026Updated Oct 23, 2026

Distributed capacity Google and Voltus added in

1,070 MW

Announced Oct 22, 2026, on top of a 100 MW tranche signed Jun 2, 2026

PJM capacity clearing price, 2028/29 delivery ye

$325/MW-day

Cleared at the FERC-approved cap, auction results published Jul 14, 2026

PJM capacity shortfall for 2028/29

6,831 MW

Auction results Jul 14, 2026; reserve margin 14.7%

Alphabet 2026 capital spending guidance

$195–205B

Raised from $180–190B on the Q2 2026 earnings call, Jul 22, 2026

Google and power aggregator Voltus announced on October 22, 2026 that they will fund 1,070 megawatts of distributed capacity in PJM — ten times the 100 MW they signed in June — on a contract that runs through 2040. It is small next to PJM's 168.2 GW record peak set on July 2, 2026, and large next to the 6,831 MW of reliability the region's July auction could not buy. The trade converts an interconnection queue into a financing decision on a hyperscaler's balance sheet instead of a residential ratepayer's bill.

Distributed capacity Google and Voltus added in PJM

1,070 MW

Announced Oct 22, 2026, on top of a 100 MW tranche signed Jun 2, 2026

PJM capacity clearing price, 2028/29 delivery year

$325/MW-day

Cleared at the FERC-approved cap, auction results published Jul 14, 2026

PJM capacity shortfall for 2028/29

6,831 MW

Auction results Jul 14, 2026; reserve margin 14.7%

Alphabet 2026 capital spending guidance

$195–205B

Raised from $180–190B on the Q2 2026 earnings call, Jul 22, 2026

Nuclear power Google contracted from Constellation Energy

3,590 MW

890 MW nuclear PPA plus 2,700 MW supply agreement, announced Oct 6, 2026

Voltus distributed capacity target for 2030

20 GW

Stated alongside a $225 million Series D financing, Oct 2026

PJM capacity market

The Auction That Cannot Clear

PJM's July 2026 Base Residual Auction cleared at $325 per MW-day — the cap set by the Federal Energy Regulatory Commission — and still came 6,831 MW short of its reliability requirement, with the reserve margin at 14.7% on 138,318 MW of unforced capacity. It was the third consecutive auction to clear at a cap, after $269.92 and $329.17 per MW-day for the two prior delivery years. When a market cannot price scarcity it rations it instead, and PJM is rationing megawatts through interconnection queues.

PJM capacity prices have cleared at the cap three auctions running

Clearing price by delivery year, dollars per MW-day of unforced capacity.

Unit: $/MW-day (UCAP)

2025/26

Cleared at the cap

269.9

2026/27

134,311 MW procured

329.2

2027/28

134,478 MW procured

333.4

2028/29

138,318 MW; 6,831 MW short

325

What each auction delivered
Delivery yearClearing priceUnforced capacity procuredResult
2026/27$329.17/MW-day134,311 MWCleared at the cap
2027/28$333.44/MW-day134,478 MWCleared at the cap
2028/29$325.00/MW-day138,318 MW6,831 MW short of requirement; 14.7% reserve margin
2028/29 including FRR$325.00/MW-day149,182 MW$16.4 billion of cleared capacity value
PJM bought 138,318 MW of unforced capacity for $16.4 billion in July and still finished 6,831 MW short of reliability. The cap is doing the rationing: every auction clears at the price ceiling while the interconnection queue gets longer.

The two-sided trade

Sixteen Days Apart, Two Ways to Buy the Same Grid

On October 6, 2026, Constellation Energy announced a 20-year power purchase agreement with Google for 890 MW of new nuclear capacity at 11 reactors in Illinois, Pennsylvania and New Jersey, plus a 15-year supply agreement for another 2,700 MW — 3,590 MW in all, backed by more than $4.3 billion of investment and a first uprate due in 2028. Sixteen days later, Google and Voltus expanded the PJM virtual power plant from 100 MW to 1,070 MW, running through 2040 — buying the megawatts and the right to be called on in the same market.

Google's PJM position after two October announcements
InstrumentSizeTermCounterpartyAnnounced
Bring Your Own Capacity, June tranche100 MW3 yearsVoltusJun 2, 2026
Bring Your Own Capacity, expansion1,070 MWThrough 2040VoltusOct 22, 2026
Nuclear uprate power purchase agreement890 MW20 yearsConstellation EnergyOct 6, 2026
Energy supply agreement2,700 MW15 yearsConstellation EnergyOct 6, 2026
Data center and energy developerIntersect portfolion/aAlphabetAgreed Dec 22, 2025 at $4.75B
  • Constellation Energy sells 3,590 MW — 890 MW of new nuclear uprates plus 2,700 MW from its existing fleet — on 20-year and 15-year terms.
  • Voltus carries Google's PJM capacity obligation and pays households and businesses to curtail batteries, thermostats and industrial load at peak.
  • Google now buys the megawatt twice over: once firm for twenty years, once flexible for fourteen, with a ramp that runs three to four years.

Unit economics

What a Demand-Side Megawatt Is Worth

At the 2028/29 clearing price, 1,070 MW of accredited capacity is worth about $127 million a year gross, before any share to the aggregator and before the cost of the assets standing behind it. Google's payment is not disclosed, so the only defensible read is comparative: the same 1,070 MW of new nuclear capacity would cost on the order of $5.2 billion at the rate Constellation Energy is investing per megawatt of uprate. The demand-side portfolio buys queue position and speed, not cheap power.

The arithmetic behind1,070 MW
MeasureFigureBasis
Gross annual value of 1,070 MW at the $325 cap~$127 million1,070 MW × $325/MW-day × 365 days
Cost of equivalent new nuclear capacity~$4.8 million per MWConstellation's $4.3B+ for 890 MW of uprates
Google's payment to VoltusNot disclosedDeal specifics have not been published
Share of PJM's 2028/29 reliability shortfall covered15.7%1,070 MW against a 6,831 MW gap
Share of PJM's record peak load0.6%Against the 168.2 GW peak set Jul 2, 2026
Whole PJM capacity market, 2028/29$16.4 billion138,318 MW at the cap

1,070 MW against the scale of the PJM problem

Megawatts. Google's new portfolio is a rounding error against the region's load and a visible slice of its shortfall.

Unit: MW

PJM record peak load, Jul 2, 2026

168,200

2028/29 capacity available incl. FRR

149,182

2028/29 capacity procured

138,318

2028/29 reliability shortfall

6,831

Google and Voltus expansion

1,070

Constellation nuclear uprates for Google

890

What new PJM peak load costs per year at the cap

Annual capacity charges at $325 per MW-day for additional unforced capacity serving load, before energy costs.

Unit: $ million per year

1 GW of new peak load

118.6

5 GW

593.1

10 GW

1,186.3

20 GW

2,372.5

What Google and Voltus have not disclosed

Price paid by Google

Not disclosed

Deal economics have not been published

Accreditation ratio for the portfolio

Not disclosed

Determines how much nameplate MW clears as firm capacity

Who holds the obligation

Voltus

Reported to be responsible for Google's PJM capacity obligation

Participant payment mechanics

Bill savings plus cash

Homes and businesses are paid to curtail at peak

Energy and ancillary settlement

Not disclosed

Only the capacity leg has been described

Upstream

Who Actually Builds the Volt

The 1,070 MW does not exist yet: Voltus expects to ramp the plant over three to four years and then run it under contract with Google through 2040. The hardware behind it is a mix of home batteries, smart thermostats and industrial assets assembled by third parties rather than built by a utility. That is the structural change — the supply of flexibility now comes from fleets that can be paid to switch off, and the party that signs the contract owns the option.

  • Voltus raised $225 million in Series D, targets 20 GW of distributed capacity by 2030, and acquired storage software startup Brightfield to shorten deployment time.
  • Sunrun began supplying capacity from thousands of residential storage-plus-solar systems in PJM and MISO on August 17, 2026, giving Voltus a bank of household megawatts it can dispatch on command.
  • Sunrun, Renew Home and Tesla set a 16.8 GW framework on June 24, 2026 drawing on more than 8 million smart thermostats, with over 300 MW deployable in Virginia and at least 500 MW targeted by 2030.
  • Firm supply answers instead: Doosan Enerbility shipped its first 380 MW turbine to a U.S. data center and holds 12 U.S. turbine contracts, while GE Vernova's gas backlog reached 116 GW.

Downstream

Who Absorbs the Capacity, and Who Pays for It

Downstream, the capacity has to land somewhere. Generators sell into the capped auction, utilities build the wires and the rate base, and ratepayers eventually carry the bill for the capacity that was not built. FERC's October 2026 rejection of most of PJM's bring-your-own-power plan removed the rule that would have forced large new loads to bring equivalent generation, leaving hyperscalers to fund their own — which is exactly what Google is now doing on both sides of the meter. The effect moves the cost of data-center load off the rate base and onto corporate balance sheets.

Who owns the capacity: trailing twelve months to June 30, 2026, from each company's Form 10-Q
CompanyRevenueEBITDACapital expenditureFree cash flow
Alphabet$445.9B$326.3B$132.4B$53.3B
Constellation Energy$31.3B$8.0B$3.9B$0.3B
Vistra$19.2B$6.6B$3.7B$1.4B
Exelon$25.3B$9.1B$9.1B-$1.9B
GE Vernova$41.4B$9.1B$1.7B$12.4B
At the $325 cap, each 10 GW of new PJM peak load adds about $1.19 billion a year to the capacity bill. If a hyperscaler funds its own capacity rather than putting it in the rate base, the cost of data-center load leaves utility customers entirely.

Risk

The Performance Problem

The open question is whether a megawatt bought from a household behaves like a megawatt bought from a power plant. PJM's own materials record a 67% weighted average performance across demand response dispatches in summer 2025, and emergency programs supplied 88% of the region's $523.5 million of 2025 demand response revenue, which rose 240% from 2024 — revenue that is high and volatile precisely because it is scarce. A hyperscaler paying for flexibilityneeds it to perform when PJM calls, not when it is convenient.

  • PJM recorded 67% weighted average performance across summer 2025 demand response dispatches, a figure cited in Voltus's own January 2026 petition to the Market Implementation Committee.
  • The counterparty to this deal is lobbying for the rules that set its value: Voltus asked PJM to reform performance incentives in the same forum where that 67% figure appears.
  • The one comparable Google-funded program, PG&E's SHARE pilot enrolling more than 20,000 customer-owned devices plus battery-enabled heat pumps, has been called not a success story by flexibility adviser Jigar Shah.
  • A 2025 Department of Energy study put distributed resources at up to 20% of U.S. peak load by 2030 — the bull case for scale, and the reason regulators are moving on this.

Horizons

Two Horizons for the Same Trade

What moves when, and who is exposed
HorizonWhat movesWho is exposedMarker to watch
Days to quartersAlphabet leans on third-party capacity as a bridge while its own build-out rampsAlphabet margin; PJM generators selling at the capQ3 2026 commentary on the modest margin pressure management flagged
Days to quartersVoltus ramps toward the plant's full capacity over three to four yearsSunrun and Tesla battery volumesDeployment pace and the first PJM accreditation filings
One to three yearsFirst Constellation nuclear uprate lands, with 890 MW of new firm supplyConstellation Energy volumes; merchant scarcity premiumUprate schedule across 11 reactors
One to three yearsDemand-side capacity scales as Voltus pursues 20 GW by 2030 against PJM peak load rising to 222,106 MW by 2036Vistra and other merchant capacity revenueWhether the cap binds again at the next auction
  • Watch the next capacity auction: three straight cap-clears mean the price signal for new firm supply is broken, so a fourth is the base case.
  • Watch how PJM prices distributed capacity in its own auctions. If it clears below the cap, Google has built a cheaper megawatt than the one it is contracting from Constellation Energy.
  • Over one to three years the demand-side leg becomes the marginal megawatt in PJM, and whoever owns the fleet of batteries and thermostats collects the spread.

Synthesis

Capacity Has Become a Corporate Cost Line

The demand-side leg of AI power is now a financing decision on a hyperscaler's balance sheet rather than a line on a utility bill. Google has shown the template: fund the aggregator, take the capacity obligation, and let other people's batteries and thermostats stand behind 1,070 MW of firm claims. The money follows the obligation — to Constellation Energy on the supply side, to Voltus and the device owners on the demand side — and the rate base gets smaller. Hyperscalers now set the terms of AI power rather than accept them.

Expand the use of third-party capacity in Q3 as a bridging strategy while we build out more internal capacity.

Alphabet, Q2 2026 earnings call, Jul 22, 2026

The stocks this trade actually touches

GAlphabetGOOG--
--Vol --
-
Mixed
  • Owns the obligation rather than the assets: Voltus is responsible for Google's PJM capacity obligation, so the cost is a contract rather than a rate case.
  • 2026 capital spending guidance of $195–205 billion against $53.3 billion of trailing free cash flow means power cost competes with compute for every incremental dollar, and energy becomes a first-order AI cost line.
  • Horizon: near term, third-party capacity as a Q3 bridge adds the modest margin pressure management flagged; long term, 1,070 MW of self-funded flexibility shortens the path from capital plan to operating capacity.
CConstellation EnergyCEG--
--Vol --
-
Bullish
  • 3,590 MW contracted with Google, including an 890 MW nuclear PPA and a 2,700 MW supply agreement, at a market clearing at the $325/MW-day cap.
  • Trailing revenue of $31.3 billion and EBITDA of $8.0 billion give the balance sheet to fund more than $4.3 billion of uprates without equity.
  • Horizon: the first uprate due in 2028 adds firm baseload into a grid short 6,831 MW, while the cap keeps realized capacity prices elevated.
VVistraVST--
--Vol --
-
Mixed
  • Sells PJM capacity into an auction clearing at $325/MW-day for 2028/29, with trailing revenue of $19.2 billion and EBITDA of $6.6 billion.
  • Trailing capital expenditure of $3.7 billion against $1.4 billion of free cash flow limits how fast merchant capacity can chase a 6,831 MW gap.
  • Horizon: 1,070 MW of Google-funded demand-side supply, and a 20 GW aggregator ambition behind it, caps the scarcity premium merchant capacity depends on over one to three years.
EExelonEXC--
--Vol --
-
Bearish
  • Trailing capital expenditure of $9.1 billion against $7.2 billion of operating cash flow leaves free cash flow at negative $1.9 billion while data-center load arrives.
  • FERC's rejection of most of PJM's bring-your-own-power plan leaves hyperscalers free to fund their own capacity, bypassing the rate base and the cost recovery that goes with it.
  • Horizon: PJM peak load growth to 222,106 MW by 2036 is the offset, and the reason a bearish view on cost recovery is not a bearish view on volumes.
RSunrunRUN--
--Vol --
-
Mixed
  • Supplies capacity from thousands of residential storage-plus-solar systems in PJM and MISO into Voltus's bring-your-own-capacity programs for AI hyperscalers.
  • Part of the 16.8 GW framework with Renew Home and Tesla, with more than 300 MW deployable in Virginia now and at least 500 MW targeted by 2030.
  • Trailing free cash flow of negative $1.4 billion against $1.05 billion of annual interest expense makes every contracted megawatt matter more than the margin.
TTeslaTSLA--
--Vol --
-
Mixed
  • Home batteries are one input to the 16.8 GW Sunrun-Renew Home-Tesla framework, which committed capacity to PJM's proposed Reliability Backstop Process.
  • Trailing revenue of $103.6 billion and net income of $3.2 billion mean grid flexibility stays optionality until PJM prices demand-side capacity at scale.
GGE VernovaGEV--
--Vol --
-
Bullish
  • Gas power backlog and slot reservations reached 116 GW at the end of Q2 2026, up from 100 GW three months earlier.
  • $2.4 billion of AI data-center orders in the first quarter shows who gets paid when a hyperscaler needs firm megawatts rather than flexible ones.
  • Horizon: demand-side capacity defers but does not cancel turbine demand as PJM peak load grows 3.6% a year to 222,106 MW by 2036.
0Doosan Enerbility034020.KS--
--Vol --
-
Bullish
  • Shipped its first 380 MW gas turbine to a U.S. data center project in October 2026 and now holds 12 U.S. turbine contracts.
  • Agreed three 380 MW units for a leading U.S. technology company in December 2025, ahead of that first export.
  • Horizon: U.S. data-center demand pulls Korean turbine orders forward as PJM's firm-capacity gap persists into 2028.