The smart panel company Span will roll out a high-performance edge data center later this year that can be installed in homes and be powered by excess grid capacity, Latitude Media has learned.
Called the XFRA Node, the mini distributed compute system is paired with Span’s smart panel, which monitors a home’s electricity use in real time, as well as a whole-home battery system. XFRA’s orchestration software can schedule and route artificial intelligence workloads across different nodes based on latency requirements and available energy capacity, the company said.
“Our hypothesis at Span has been that the existing distribution network operates at only 40 to 45% utilization, nominally,” Arch Rao, founder and CEO of Span, told Latitude Media. “So there’s plenty of headroom on the existing system that can be used in a more effective way. And then we’re able to deliver compute much faster.”
This comes as Span raises its Series C round, with $163 million in equity sold as of February. The company in 2025 expanded its business with a turn toward utilities looking to meet load growth, rather than selling its smart panel to the individual consumer market alone.
The new compute product will be announced at Latitude Media’s Transition-AI conference this week.
It arrives as demand for AI computing power outstrips the pace of deploying new power infrastructure in the U.S., due to years-long interconnection queues for new load and generation, equipment shortages, and local opposition. Up to half the data centers worldwide may be delayed this year, according to research by Sightline Climate.
The crunch has created momentum around grid utilization, a broad range of initiatives to unlock capacity via, for instance, shifting large load energy use via demand response, or else building out behind-the-meter resources like battery storage. Distributed computing falls under that grid utilization umbrella: Hyperscalers that buy the resulting compute could avoid the need for expensive new poles and wires — infrastructure that utility customers often end up paying for.
Rao noted that it currently takes between three and five years to build a 100-megawatt data center, and it costs upwards of $15 million per MW. Span can supply the same amount of compute by installing the XFRA Node at 8,000 new homes, which would take roughly six months, he said, at a cost of $3 million per megawatt.
An XFRA pilot is expected to roll out this year in 100 newly constructed homes, or about 1.25 MW of compute capacity across 1,600 direct liquid cooled inference GPUs. Span has several partnerships, including with Pulte Homes, the third-largest homebuilder in the country, which has already been installing Span’s smart panels.
“They’re able to use Span’s panel technology to stay at 200 amps per home, which reduces the size of interconnection with the utility and the amount of copper in the home,” Rao said. “Now we’re going one step further in our partnership to say, ‘If you also deploy XFRA Nodes in a subset of homes, that has the added benefit of making the home more affordable to their buyers in terms of heavily discounted energy and Internet.’”
Installation is at no cost to homeowners, but they will pay a $150 monthly fee that covers energy and internet, which is a discount on typical monthly costs. Span then sells computing power to customers such as hyperscalers, neocloud companies, and AI companies. Rao said the revenue from selling compute will exceed the cost of the XFRA system itself, and that customers will end up paying less month-to-month. The model is similar to homeowners paying a fixed rate for solar panels on their roofs that are owned by a third party, Rao said — except in this case, Span will own computing assets and backup battery storage.
These small “nodes” can’t replace large, centralized AI data centers, of course, which are needed for training AI models. The nodes are complementary, Rao said, especially for inference, meaning the actual use of AI models such as chatbots, autonomous driving, and medical image diagnostics. As inference becomes the primary source of revenue for AI companies, it will be ideal to locate computing resources closer to users to reduce latency, he added.
Span is also working on a commercial XFRA Node for early 2027 that can be sited at businesses and office buildings.
