Although some distributed energy companies are starting to get it…
We tend to focus on capacity shortages, interconnection delays, and rate impacts when we talk about the power industry/AI relationship.
But it’s even bigger than that.
Ezra Klein recently interviewed Jensen Huang, founder and CEO of Nvidia, about the potential existential threat of recursive AI. Huang, who was arguing against humanizing AI, said: “There is no will power here; there is just electrical power.”
The statement is telling. At its most basic, AI is not human; it’s electrons. Pull the plug, and it’s gone.
Obviously, no one is pulling any plugs. But the power industry can — and is — starting to pull some levers. A few distributed energy companies are realizing they can leverage the needs of the $2.7 trillion global AI industry to reinvent themselves and tackle market and societal issues at their doorstep.
A new model for distributed energy and compute
It’s been a tough couple of years for segments of distributed energy, rooftop solar in particular, with high interest rates, loss of federal incentives and its biggest market — California — shifting to a less generous net metering structure.
That’s left solar companies looking for new models, particularly ways to scale beyond the one-off home installation. This is where the power industry’s marriage with AI gets interesting.
Consider what solar company Sunrun and electric panel maker SPAN have teamed up to do. The companies are marrying distributed energy with distributed edge computing as a home product. Sunrun first unveiled the pilot in July, and last week named SPAN as a technology partner.
Sunrun plans to launch the product in Texas, and possibly other states, by working with homebuilders. Sunrun will install solar and batteries in the home, and SPAN will provide its new XFRA compute node.
The node, which sits outside the house, operates under principles similar to distributed energy. It provides compute power close to the user and can be aggregated virtually to support the larger system.
Because the compute happens in the home, processing doesn’t need to travel long distances to a hyperscale data center. This reduces latency.
Support our work with a free subscription or premium membership
Meanwhile, Sunrun tackles the energy side, providing solar and storage to power the home and the mini-data center and possibly serve the grid via the battery.
Why would a homeowner want this? The companies say households will get revenue, backup power, and faster compute. Because the system is provided via a third-party funded prepaid lease, the household pays nothing for the smart panel, solar and battery.
So if it tests out, the Sunrun/SPAN system will bring a new value proposition to the home solar and storage industry, new benefits to households, and a way for data centers to secure energy without waiting to interconnect to the grid.
Looking deeper into the home’s value
All of this stems from Sunrun’s view of the home as a place where energy and compute can be jointly monetized.
“Allocating those electrons to the highest return is something that we’re constantly thinking about. And when you look at the value of using those electrons to power a GPU in someone’s home versus the alternative, the returns to us are really, really attractive,” said Paul Dickson, Sunrun president and chief revenue officer during its second quarter earnings call August 5.
But the concept offers more than a new revenue path for distributed energy companies. SPAN says it can aggregate multiple home nodes virtually to coordinate capacity at a gigawatt scale. In an era when building new data centers is often difficult because of interconnection delays and community opposition, SPAN instead leverages existing infrastructure.
Inside SPAN’s Home Data Center, XFRA
- 16 NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs
- 4 AMD EPYC CPUs
- 3 TB of memory
- 15-kWh battery
- 35,000-BTU heat pump for liquid cooling
- SPAN smart electrical panel
- XSOL orchestration software
Scale: SPAN says XFRA is designed to reach more than 1 GW of annual AI-inference compute capacity. At roughly 12.5 kW of compute capacity per node, 1 GW would require about 80,000 XFRA nodes.
Second, the model addresses energy equity issues, offering households a way to capture some piece of the wealth being generated by the AI boom. Sunrun said in July that it is testing the economics in a pilot program.
AI as a catalyst for a different power industry
The Sunrun/SPAN concept isn’t the only new model that leverages AI’s need for power to benefit the household.
- Earlier this month, Energy Changemakers reported that startup Resilience Energy plans to use solar- and battery-powered homes to provide capacity for data centers. The capacity contract with the data center finances the solar and storage; homeowners pay nothing for the equipment, while seeing likely reductions in their electricity bills and getting backup power. In return, they agree to dispatch energy from their batteries when the grid is under strain, roughly 15-20 days per year.
- In California, PG&E is working with several energy vendors to build an ecosystem of distributed energy companies that install new home energy systems and aggregate existing devices. The devices will be brought together to create a virtual power plant. The households become an energy resource and receive compensation. In some cases, the program works with existing devices; in others, Google pays for installation in homes.
- Google is also working with virtual power plant company Voltus on a similar program in PJM’s territory. The 100-MW, three-year program assembles flexible assets in homes and businesses, such as batteries, smart thermostats, solar panels, electric vehicles, and backup generators, to serve the grid.
What’s emerging here may be bigger than a new business model for rooftop solar or a workaround for data center interconnection delays. The AI boom creates an opening to rethink who provides energy and capacity, how they are compensated, and who gets to participate in the value being created.
While much of the market is focused on how the power industry will serve AI, these companies are raising a bigger question: How can this moment create a different kind of power industry?
