Sunrun is expanding a pilot program that installs computing hardware in homes already equipped with its solar and battery systems, the company announced on July 8. The distributed data center initiative aims to help AI companies avoid years-long waits for traditional data center capacity while reducing strain on congested transmission and distribution networks. Unlike centralized data centers that can consume hundreds of megawatts in a single location, Sunrun's approach places computing nodes about the size of small desktop computers across its network of more than 1.1 million existing solar and battery customers.
The company says its customer base represents "an addressable deployment base" that gives it "a structural advantage hyperscalers can't quickly replicate," referring to the multibillion-dollar tech firms that develop and lease computing capacity in massive data centers. Sunrun will manage the computing nodes around hosts' energy consumption patterns, electricity rate structures, and available grid service programs, with customers receiving unspecified compensation in exchange. The company hasn't disclosed how much it will spend on the expanded pilot or how many customers it hopes to enroll, though a Sunrun webpage invites prospective hosts to sign up for a waitlist. Sunrun expects to complete the pilot "over the coming months" and assess results to determine how to proceed with a wider rollout.
"AI companies are scrambling to secure greater access to energy and computing power," Sunrun President and Chief Revenue Officer Paul Dickson said in a statement. "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." The company argues that behind-the-meter computing nodes are more resilient to "regional threats" like rising utility rates, overloaded grids, and power supply shortages. Sunrun is "actively in discussions with enterprise compute offtakers, homebuilders, and utility partners to structure the commercial and deployment frameworks that would support expansion," the company said.
The announcement comes about three months after electrical hardware startup SPAN unveiled its own distributed compute initiative with NVIDIA, citing the ability to sidestep grid power constraints delaying traditional data centers while improving AI model performance. SPAN CEO Arch Rao told Latitude Media in April that building a 100-megawatt data center can take up to five years at a cost of $15 million per megawatt or more, while SPAN can deploy the same computing capacity across 8,000 residential nodes in six months at $3 million per megawatt. Computer architect Benjamin Lee of the University of Pennsylvania said distributed nodes could take off as inference overtakes training as the dominant AI workload, since inference can run on just a few chips rather than "thousands of [chips] working in concert." But Lee questioned whether the approach needs to scale down to individual homes, suggesting a 20-megawatt data center could provide similar grid benefits with better security for valuable AI chips than private residences offer.