Sunrun, the San Francisco-based residential solar provider, is testing whether homes with solar panels and batteries can become part of the AI infrastructure stack.
The company launched a distributed AI compute pilot July 8 to place compute nodes in homes equipped with Sunrun solar and battery systems, coordinate the sale of inference capacity to enterprise buyers and compensate participating homeowners.
Sunrun President and Chief Revenue Officer Paul Dickson described the pilot as a response to constraints in both power and compute. “AI companies are scrambling to secure greater access to energy and computing power,” he said in the announcement.
Sunrun disclosed the pilot after a proof of concept and said it will test nodes under different conditions and rate structures before deciding whether to expand.
Building on flexible capacity agreements
The pilot follows Sunrun’s June 24 agreement with Renew Home and Tesla to aggregate more than 16 GW of flexible home energy capacity for hyperscalers and utilities. The companies said the framework would combine home batteries, smart thermostats and vehicle-to-grid systems into local capacity without requiring additional hardware, interconnection, water or land from offtakers.
Sunrun tied the two efforts directly. The compute pilot is separate from the Tesla and Renew Home agreement, but Sunrun said in-home compute capacity can serve the same AI demand pushing hyperscalers to seek new energy capacity.
Sunrun CEO Mary Powell made the ratepayer argument in the June announcement: “The grid of the 1800s cannot power the innovation of 2026.”
Rising data center power projections
Federal and international projections show data-center electricity demand is already becoming a planning issue. In its Annual Energy Outlook 2026, the U.S. Energy Information Administration said U.S. electricity demand has risen 2.1% per year on average over the past five years and projected 0.9% to 1.6% annual growth through 2050, with data center server energy use a major factor.
The International Energy Agency projected global data center electricity consumption to double to about 945 TWh by 2030, growing about 15% per year from 2024 to 2030.
Targeting atomized inference workloads
The pilot’s focus on inference tracks McKinsey’s projection that inference will become the dominant AI data-center workload by 2030. McKinsey research projects inference will surpass training by 2030, represent more than half of AI compute and grow at a 35% compound annual rate from 2025 to 2030.
The same research describes inference as more atomized than training, which relies on large synchronized clusters.
Regulatory scrutiny and ratepayer impact
Grid regulators are already treating large-load connection as a policy issue. On June 18, the Federal Energy Regulatory Commission ordered six regional grid operators and their transmission owners to justify within 60 days why their tariffs remain just and reasonable without clear provisions for large-load customers, or propose changes.
In a Jan. 14 FERC statement, the agency said Energy Secretary Chris Wright had asked it to consider large-load reforms that ensure “families and small businesses do not foot the bill” for grid upgrades needed to support large consumers, including data centers and industrial facilities.
The ratepayer argument is less settled. A Brattle Group study said better system utilization could put downward pressure on customer rates and help new loads connect faster, but it also said outcomes depend on utility-specific conditions, cost allocation, regulatory design and whether distributed resources are cheaper than conventional grid upgrades.
Pilot unknowns and operational risks
The announcement does not disclose the number of compute nodes, hardware specifications, pricing, service-level commitments, enterprise customers or detailed security architecture for the pilot.
Its own forward-looking statement lists customer authorization, homeowner participation, data security, cybersecurity, utility rate structures, outages, equipment failures and local, state, federal, utility and building-code requirements among the risks. Sunrun expects to complete the pilot over the coming months and assess defined milestones, compute performance and homeowner experience before deciding the scale, speed and customer offer for any broader rollout.