Google has signed an agreement with Voltus, a virtual power plant operator, to finance a 100 megawatt distributed energy program across PJM (the regional transmission organization serving 65 million people from Illinois to New Jersey). Starting in 2027, Voltus will pay homeowners and businesses to let the company remotely adjust their electric vehicle charging schedules, smart thermostats, and other connected devices during the roughly 40 hours per year when the grid is maxed out. The freed up electricity goes straight to Google's data centers in the region. Neither company will say how much they're paying participants, which tells you everything about whether they think the number is impressive. This is the first named customer for Voltus's "Bring your own capacity" program, launched last September. The pitch to tech giants is elegant: instead of waiting years for new transmission lines or power plants, just pay other people to be flexible so you don't have to be. Google has been curtailing its own data center loads for years, but as AI workloads explode (training runs can't easily be paused mid model), the company needs flexibility from someone else's plug. Virtual power plants aggregate thousands of small energy sources into one controllable block, turning a neighborhood of smart devices into a makeshift battery. The Duke University study that sparked last year's data center flexibility craze found that if new facilities agreed to reduce demand during peak stress, the grid could absorb 100 gigawatts of new load without infrastructure upgrades. That's the equivalent of powering 75 million homes. The math works because grids are engineered for the absolute worst case scenario (think August heat wave at 6pm when everyone's home running air conditioning), not average use. Data centers that voluntarily go dark during those needle peak hours could theoretically slot in everywhere else. Texas already passed a law requiring large users to switch to backup power during emergencies. A federal proposal would let data centers skip years long interconnection queues if they commit to demand response. But there's a credibility problem. A recent California study on managed EV (electric vehicle) charging found that only 1% of owners enrolled with no incentive. At $40 per month (roughly 15% of a typical electric bill), participation crept up to 4.6%. That's abysmal adoption for a program that requires almost zero effort, you literally just let the utility delay your car charging by a few hours. If Californians won't bite at $40 monthly for EV management, what will it take to get them to hand over thermostat control to help power Google's Gemini training runs? And that's before considering that 70% of Americans oppose AI data centers in their area, according to Gallup polling from earlier this year. The incentive problem cuts both ways. Data centers themselves have limited flexibility once AI workloads dominate their racks. Training a large language model can be paused or shifted to off peak hours, but inference (answering user queries in real time) cannot. Every minute of downtime is lost revenue. Google's blog post from earlier this year acknowledged that not every facility can ramp down power demand. So the company is outsourcing flexibility to your living room. Whether your living room wants the job is the $100 million question (Voltus raised $68 million in 2024 and is valued north of $400 million, so there's serious money riding on this working). PJM's capacity market already pays generators and demand response providers for being available during peak hours, but this Google deal represents a new model: private tech financing public grid flexibility to serve private data center growth.
💻 technology
Google Bribes Your Thermostat to Feed AI Hunger
Google just inked a deal to pay American households real money to dial back their power use during peak hours not for the planet not for the grid but so its data centers can keep training AI models without building new power plants. The catch? Most people won't play along.
My Take
This is demand response theater dressed up as climate virtue. Google doesn't want to build power plants or wait for transmission upgrades, so it's paying Voltus to ask you to turn down your AC while its data centers churn out another chatbot that hallucinates legal citations. The company frames this as grid modernization, but it's really grid arbitrage: lock in cheap flexibility from thousands of households who don't realize they're subsidizing AI training. The California EV study should terrify Voltus investors. If fewer than 5% of people will accept $40 monthly to delay car charging by a few hours, what adoption rate do you expect when the pitch is "let us control your thermostat so Google can keep Bard running"? Americans already distrust Big Tech with their data; now we're asking them to hand over control of their physical comfort. The anti data center sentiment is real and growing. When 70% of the country opposes these facilities in their backyards, good luck convincing them to become unpaid grid operators for those same facilities. The bigger issue is pretending this solves the data center energy crisis. It doesn't. It delays the reckoning by maybe two years while AI power demand doubles again. Virtual power plants are a temporary fix on a structural problem. The real question is whether we actually need every AI feature these companies are racing to ship, but that would require some uncomfortable honesty about priorities. Instead we get programs that pay households to reduce consumption during peak hours while data centers expand their footprint training models for features nobody asked for.
What Happens Next
PJM's capacity auction in 2027 will be the first real test of whether Google's bet pays off. If Voltus can't aggregate the promised 100 megawatts because participation is anemic, Google will quietly pivot to co locating data centers next to nuclear plants (watch for announcements in Pennsylvania and Ohio by late 2026). The company has already signed deals with Kairos Power for small modular reactors; this VPP experiment is the backup plan, not the main strategy. Here's the scenario nobody's pricing in: what if utilities start blocking these private VPP deals? PJM's independent market monitor has already raised concerns about tech companies paying for demand response that competes with utility run programs. If regulators decide that Google is essentially buying its way to the front of the interconnection queue by financing flexibility, expect lawsuits from utilities who've spent billions on transmission upgrades. The Virginia legislature, where Amazon owns 40% of all data center capacity in the state, will be ground zero for this fight. The real action will be in managed charging adoption curves over the next 18 months. If Voltus can't get participation above 10% in PJM (where electricity is cheaper and people are less environmentally motivated than California), every other tech giant will abandon the VPP approach and go nuclear, literally. Microsoft, Meta, and Amazon are all watching this Google pilot. A failure here accelerates the data center nuclear renaissance by 2 to 3 years and potentially saves the struggling small modular reactor industry. Success means we'll see VPP deals in every grid region by 2028, with participation incentives climbing toward $100 monthly as tech companies bid against each other for the same pool of flexible load. Either way, your thermostat is about to become a commodity.
Market Impact
Voltus is privately held (last valued around $400 million after raising $68 million in 2024), so there's no direct ticker play, but Stem Inc. (STEM), currently trading around $0.45 after cratering 90% from its 2021 SPAC peak of $5, could see a speculative bounce if Google's program validates the VPP business model. STEM operates AI driven energy storage and has pivoted hard toward data center clients after its residential business flopped. A successful Google Voltus pilot would prove that tech giants will pay real money for distributed flexibility, potentially rescuing STEM's bleeding balance sheet. Watch for any earnings mention of new enterprise VPP contracts. On the utility side, NextEra Energy (NEE), trading around $72 (up 8% YTD but down from its $90 pre rate hike peak), has the largest VPP footprint in the US through its subsidiary. If demand response becomes table stakes for data center interconnection approvals, NEE's existing programs give it a structural advantage in winning hosting agreements with tech companies. The company's Q2 earnings call in July will likely address whether they're seeing increased interest from hyperscalers. Nuclear plays are the contrarian bet here. If this VPP experiment fails to scale, Constellation Energy (CEG), currently at $185 (up 45% over the past year as data center nuclear deals proliferated), becomes the cleaner path for tech companies needing guaranteed baseload power. CEG already has agreements with Microsoft for the Three Mile Island restart. A Voltus stumble would accelerate similar Google and Amazon deals, potentially pushing CEG toward $220 by year end. The anti correlation is stark: VPP success caps nuclear upside, VPP failure sends CEG parabolic.