Residential solar and battery storage company Sunrun is joining the emerging market for edge computing by turning homes into decentralized data centers.
The company last week announced a pilot program to install “nodes” equipped with Nvidia chips in homes equipped with Sunrun’s solar and storage systems. The pilot is similar to one rolled out by the smart electric panel maker Span in April. Span said it is testing its approach in 100 newly constructed homes this year.
The edge computing trend comes as the electricity demands of artificial intelligence are outstripping the pace of new power generation and transmission lines added to the grid. That gap is due in part to years-long interconnection queues and equipment shortages. Around the world, up to half of the planned data centers scheduled to come online this year may be delayed, according to research by Sightline Climate.
“There’s aging poles and wires and ever increasing complexities and timelines to build a new generation,” Paul Dickson, Sunrun’s president and chief revenue officer, told Latitude Media. He noted that it can take a decade or more to build a one gigawatt nuclear power plant. By contrast, Sunrun builds the same amount of capacity every year, with more than 1 million residential customers already using solar and storage.
“Energy is the main bottleneck to compute,” Dickson said. “So rather than trying to put 100,000 GPUs in one big building, let’s put one GPU in someone’s house and essentially build a distributed data center.”
Testing edge compute
Scaling up to that many homes is likely a long way off. Dickson said Sunrun already has some homes enrolled in the pilot, but declined to say how many nor the number Sunrun is targeting for the initial test.
The goal of the pilot is to learn more about how the hardware performs, how much compensation will entice customers, and the types of AI workloads that edge computing is best suited for.
Dickson said inference — the actual use of chatbots or other AI programs — is the focus as opposed to training. That’s because placing compute closer to workload requests is crucial for speed. Hyperscalers also pay a lot of money for inference compute, because the actual application of AI is what generates revenue.
Sunrun doesn’t want those workload requests to further strain the grid. So the ideal homeowner has solar and storage systems with more than enough generation to cover their own use, plus run GPUs.
The company’s Flex program, launched last year, is designed for this, Dickson said.
“When you install a GPU, all of a sudden a lot more energy is being used and your energy bill would go way up,” he explained. “We install a solar and storage system that covers all of that.”
Sunrun will cover the cost of the energy it takes to run GPUs with the money it earns from AI compute customers, and also pays homeowners for hosting the node that’s roughly the size of a mini fridge but uses a lot more power.
That would be in addition to whatever financial contract a homeowner has with Sunrun to install their solar and storage systems, such as fixed monthly payments for the hardware or power purchase agreements for the electricity at lower prices than a utility.
Dickson said edge computing complements Sunrun’s recently announced agreement with Renew Home and Tesla to offer a combined 16 GWs of capacity to hyperscalers via a virtual power plant. That program eventually is aimed at selling power to large data centers during periods of grid stress so they don’t have to curtail workloads.
“We’re approaching the energy crisis from two angles,” Dickson said. “With Renew and Tesla, we’re sitting on more than 16 gigawatts of power that data centers can tap into. We’ll also have GPUs in homes to produce the compute ourselves.”
