Our Take
This is a real financing structure (Voltus gets paid to orchestrate, Google pays for the capacity), but the participation question is unsolved: a California study found that even $40/month persuaded only 4.6% of EV owners to cede control.
Why it matters
Data centers consume enormous power during peak hours, forcing grid operators to build overcapacity. If tech giants can pay households and businesses to shift demand instead, they bypass years of new transmission infrastructure—but only if people actually sign up.
Do this week
Data center operators: model your own demand flexibility (training vs. inference, batch timing) and compare the cost of buying external flexibility through Voltus-style programs versus absorbing grid curtailment penalties when they arrive.
Google backs a 100-megawatt virtual power plant
Google has agreed to finance a virtual power plant operated by Voltus in PJM, the electricity grid covering much of the US East Coast. Voltus will aggregate distributed energy resources—electric vehicles, smart thermostats, and other controllable loads—and pay participating households and businesses to reduce or shift their power consumption during grid stress periods. Google foots the setup cost; the freed capacity will supply its data centers in the region. The plant aims to be operational in 2027 and could aggregate up to 100 megawatts of distributed resources annually (per Voltus).
This is the first named customer in Voltus's "Bring your own capacity" program, launched in September, which allows data center operators to finance flexibility on their local grids rather than relying solely on their own load flexibility or negotiating with utilities.
The grid was designed for peak demand, not average demand
Power grids are engineered to handle absolute maximum load: a brutally hot summer evening when everyone is running air conditioners simultaneously. A Duke University study (cited in the reporting) found that if 100 gigawatts of data center capacity agreed to curtail demand for just 40 hours per year during those peak windows, new power plants and transmission infrastructure could be avoided entirely.
Google is not new to this play. The company already has agreements with utilities across the US to limit or shift its own data center demand during peak periods. But Google acknowledges a hard constraint: not every facility can ramp down flexibly, especially as AI training workloads (which can be delayed) coexist with real-time customer serving (which cannot). This is why Google is paying others to be flexible instead.
The missing piece is adoption. A California study on managed EV charging found that with no financial incentive, only 1% of EV owners enrolled. At $40 per month (roughly 15% of their power bill), enrollment rose to only 4.6% (independent research cited by MIT). Google and Voltus are not disclosing their payment rates, which will be the decisive factor. Separately, Gallup polling shows roughly 70% of Americans oppose AI data centers in their area, adding friction to a program that asks local households to support them.
Regulation and incentive design will determine viability
Two pathways are emerging. The regulatory approach: some US states (notably Texas) now require large power consumers to curtail demand or switch to backup power during grid emergencies. The incentive approach: tech companies finance voluntary flexibility programs and hope payment is sufficient to overcome inertia and opposition.
For data center operators, the practical question is simple: model your own curtailment cost (lost revenue, SLA penalties) against the cost of buying external capacity through Voltus-style vendors. If your facility cannot flexibly defer load, expect to pay market prices for that option. If it can, measure the spread between internal flexibility (retrain, reschedule batch jobs) and external programs. Texas law signals that mandatory flexibility is coming; negotiating it voluntarily now may offer better terms than reactive compliance later.