Span announced the XFRA Node, a distributed edge data center installed in homes and powered by excess grid capacity. The system costs $3 million per megawatt, a fifth of the $15 million per MW for traditional data centers.
Key facts
- XFRA Node costs $3M/MW vs $15M+ per MW for traditional data centers
- 100-home pilot this year provides 1.25 MW across 1,600 GPUs
- Grid operates at 40-45% utilization, Span CEO says
- Span raised $163M in Series C equity as of February
- 8,000 XFRA Nodes can match 100-MW data center in 6 months
The smart panel company Span will launch the XFRA Node later this year, a mini distributed compute system paired with its smart panel and whole-home battery. The orchestration software schedules AI inference workloads across nodes based on latency and available energy [According to Latitude Media].
The math behind distributed compute
Span CEO Arch Rao told Latitude Media the existing distribution network operates at only 40 to 45% utilization. Building a 100-MW data center takes three to five years and costs upwards of $15 million per MW. Span can supply equivalent compute by installing XFRA Nodes at 8,000 homes in roughly six months at $3 million per MW.
A pilot will deploy the XFRA Node in 100 newly constructed homes this year, providing 1.25 MW of compute across 1,600 direct liquid cooled inference GPUs. Span has partnerships with Pulte Homes, the third-largest U.S. homebuilder, which already installs Span's smart panels.
Why this matters more than the press release suggests
The unique take: Span is not selling compute to consumers — it's selling grid-utilization to hyperscalers. Up to half of data centers worldwide may be delayed this year, per Sightline Climate research. Distributed computing lets hyperscalers avoid expensive new poles and wires, costs utility customers often bear. Span raised $163 million in Series C equity as of February, pivoting from consumer smart panels to utility partnerships.
The XFRA Node arrives as AI inference workloads shift toward edge deployment, a trend noted in our previous coverage of inference startups challenging Nvidia [per gentic.news, 'Inference shift opens door for AI chip startups to challenge Nvidia']. Span's model could accelerate this shift by providing cheap, distributed compute without new grid infrastructure.
What to watch
Watch for the 100-home pilot results in late 2026, including GPU utilization rates and latency benchmarks. If Span secures a hyperscaler offtake agreement, it could validate the distributed model and trigger copycats from other smart panel or battery companies.
