Let Utility Dive’s free newsletter keep you informed, straight from your inbox.
In partnership with
In partnership with
The announcement comes as ratepayer and clean energy advocates push capacity-hungry hyperscalers to subsidize residential solar, batteries and energy efficiency.
Sunrun frames its distributed computing pilot as a way for the fast-growing firms behind the leading consumer and enterprise AI models to avoid queuing up for computing capacity in data centers that can take years to interconnect.
Behind-the-meter computing nodes are more resilient to “regional threats” like rising utility rates, overloaded grids, and power supply shortages, the company said last week.
“AI companies are scrambling to secure greater access to energy and computing power. We are now using our leadership position in distributed home energy and proven infrastructure to bring compute closer to the sources of energy and inference,” Sunrun President and Chief Revenue Officer Paul Dickson said in a statement.
Sunrun’s more than 1.1 million existing solar and battery customers “represent an addressable deployment base and [give us] a structural advantage hyperscalers can’t quickly replicate,” the company said last week, referring to the multibillion-dollar tech companies that develop and lease computing capacity in centralized data centers that can consume hundreds of megawatts of power or more in a single location.
Sunrun says it will manage computing nodes about the size of a small desktop computer around hosts’ energy consumption patterns, electricity rate structure and available grid service programs. In exchange, customers will receive unspecified compensation, the company said last week.
Sunrun has not said how much it will spend on the expanded pilot nor how many customers it hopes to enroll. A Sunrun webpage on the initiative invites prospective computing hosts to sign up for a waitlist. Sunrun did not respond to a Utility Dive request for more information about the pilot by press time.
The company said last week that it “expects to complete the pilot over the coming months” and assess its results to determine how to proceed with a wider rollout. In the meantime, it’s “actively in discussions with enterprise compute offtakers, homebuilders, and utility partners to structure the commercial and deployment frameworks that would support expansion.”
Benjamin Lee, a computer architect who teaches at the University of Pennsylvania, told Ars Technica in May that distributed computing nodes could take off as inference overtakes training as the dominant type of AI-related processing. Unlike training workloads, which require “thousands of [chips] working in concert,” inference workloads can be scaled down to run just a few chips, Lee said.
“Computation for AI inference can and should be distributed at the ‘edge,’ deployed on smaller platforms closer to population centers and users. The strategy could impose much smaller impacts on the grid,” he said.
But Lee said it’s not clear that the downscaling needs to go as far as Sunrun and SPAN propose. A 20-MW data center could provide similar grid benefits, he said. And valuable AI chips may be more vulnerable to both digital and physical security threats in private homes than in well-defended data centers, he added.
In an April interview with Latitude Media, SPAN CEO Arch Rao said it can take up to five years to build a 100-MW data center at a cost of $15 million/MW or more. SPAN can deploy the same amount of compute across 8,000 residential nodes in six months at a cost of $3 million/MW, he said.
SPAN is starting smaller, however. The pilot it announced in April will install about 1.25 MW of computing capacity at 100 new-build homes, Latitude reported.
Get the free daily newsletter read by industry experts
The grid operator urged states to develop rules to shield other ratepayers from data center-driven costs, but analysts said it remains unclear how a reliability auction’s costs could be allocated only to hyperscalers.
Some of the generating units have operated during tight grid conditions and may have contributed to reliability. Keeping them running costs about $550 million a year, the Sierra Club says.
Subscribe to Utility Dive for top news, trends & analysis
Sign up for the free newsletter.
Interested? Explore more of what has to offer.
Thanks for signing up! Please keep an eye out for a confirmation email from [email protected] To ensure we make it into your inbox regularly, add us to your allow list, mark us as a safe sender, or add us to your address book. Check out more from
Get the free daily newsletter read by industry experts
The grid operator urged states to develop rules to shield other ratepayers from data center-driven costs, but analysts said it remains unclear how a reliability auction’s costs could be allocated only to hyperscalers.
Some of the generating units have operated during tight grid conditions and may have contributed to reliability. Keeping them running costs about $550 million a year, the Sierra Club says.
The free newsletter covering the top industry headlines
