Home solar company Sunrun is recruiting subscribers to host small AI compute nodes that resell workloads to outside AI buyers, a pilot that arrives as more than 70% of Americans oppose new local data centers.
A small box, tucked next to your home battery, running AI workloads for a stranger you would never meet, and paying you for the privilege: that is what home solar company Sunrun is now pitching to its residential customers, in what it calls a "distributed AI compute" pilot.
Sunrun already places solar panels and battery storage on houses across the country; it wants to add a small compute node at homes that already have both, aggregate their output, and sell the cycles to "enterprise compute buyers" such as AI companies. Homeowners would be "compensated," per Sunrun, though the company has not disclosed the dollar amount, the contract length, the exit terms, the hardware involved, the workloads, or the heat and noise footprint.
A 2026 Gallup survey, as reported by The Verge, found more than 70% of Americans oppose construction of new data centers in their local area. The drivers (electricity demand, water use, noise, and pollution) are not going away. They amount to a siting wall.
Sunrun's pilot is the first concrete commercial response from a company with a distributed energy-asset base and a customer-recruitment channel that does not require a single new construction permit. As a signal of where AI infrastructure is now trying to be built, the move is hard to miss.
The 70% figure matters because it converts a political complaint into a market pressure. If more than seven in ten Americans will fight a data center in their town, the AI industry's plan to keep building centralized campuses at the rate its model roadmaps assume starts to look arithmetic, not aspirational. The utilities, regulators, and county commissions that gate those builds have heard the same message.
Distributed residential compute is one workaround. On paper, the arithmetic is appealing: a box on thousands of existing rooftops, drawing from existing solar surplus and battery state of charge, with the hardware already financed through an electricity subscription. Sunrun's press release frames the aggregation as a "nationwide compute network," running AI from small nodes in customers' homes.
The technical gap in the planning is wide, and Sunrun has not closed it publicly. A residential node is not a hyperscale training cluster. Inference workloads (the kind that run an already-trained model against a single user prompt) are the most plausible fit, because they tolerate latency, do not require rack-scale GPU bandwidth, and let a single box be useful on its own. Sunrun has not said what its nodes will actually run.
What does matter is the contract the homeowner would sign. The disclosure so far does not include how much the box pays, in dollars per month or per kilowatt-hour; how long the commitment runs, and whether the homeowner can exit; the noise, heat, and fire-safety profile of an always-on compute box in a typical garage or basement; whether homeowner's insurance treats the node as a business activity; or whether utilities will let a home battery schedule compute load against the grid, and at what tariff. Without that disclosure, neither Sunrun nor the residents can tell whether the pilot is a working product or a cautionary tale.
The pushback that produced this moment is not going to disappear because the box sits behind a closed garage door. The Verge's coverage of the pilot and Electrek's reporting both flag that residents already organize against nearby compute, the same NIMBYism that blocks centralized builds. A pilot that places compute behind closed doors in private homes does not make that friction disappear; it relocates it.
Sunrun has not said how many homes it plans to recruit for the pilot, or what counts as success beyond its own "nationwide compute network" framing. That language is the company's, not an analyst's. The next test will be the homeowner contract: when the first draft terms appear, the gap between "distributed AI" as a press-release frame and "distributed AI" as something a homeowner would actually sign will become legible.