A San Francisco startup thinks the answer to AI’s enormous power appetite isn’t another warehouse-sized facility somewhere out by the freeway. It’s your house.
Key Takeaways:
- SPAN’s XFRA nodes pack 16 Nvidia RTX Pro 6000 Blackwell GPUs and 4 AMD EPYC CPUs into a single unit attached to the side of a home.
- The company plans to scale from a 100-home pilot to 80,000 nodes across the US by 2027, providing over 1 gigawatt of distributed compute.
- Homeowners get paid utility bills (or a flat $150 fee), free internet, and a 16 kWh backup battery – but experts flag concerns about grid strain, physical theft, and side-channel attacks.
SPAN has begun pilot testing a plan to install compact, liquid-cooled GPU nodes on the exterior of newly built homes, with the homeowner getting subsidized electricity, internet, and a backup battery in return. A 100-home trial is set for this year, with much larger ambitions to follow.
How the home-attached node would work
SPAN calls the approach a “distributed data center solution.” Each XFRA node sits beside a house, runs almost silently thanks to liquid cooling, and taps into the spare electrical capacity that most modern American homes already have. The pitch to utilities and AI companies is straightforward: instead of waiting years for a hyperscale facility to be sited, permitted, and built, deploy thousands of small nodes inside the existing electrical grid.
“Data centers are loud, ugly, and often drive up local electricity bills,” Chris Lander, vice president of XFRA at SPAN, told Ars Technica. “[This] is quiet, discreet, and makes energy more affordable for the host and community.”
According to a CNBC interview, SPAN claims 8,000 XFRA units can be installed for one-fifth the cost of building a typical 100-megawatt data center with the same compute output. The first wave focuses on new construction – SPAN pays for and runs all the equipment – though a company whitepaper hints at later retrofits and larger commercial node configurations.
What’s inside the box
A SPAN animation shows each XFRA node housing 16 Nvidia RTX Pro 6000 Blackwell Server Edition GPUs, 4 AMD EPYC Server CPUs, and 3 terabytes of memory. Alongside the node, every participating home gets a wall-mounted SPAN smart panel and a 16 kilowatt-hour battery. The company’s proprietary PowerUp software handles energy management across the whole setup. Rooftop solar may be added in suitable locations.
Power capacity is the linchpin. Most US homes built in the last three decades have 200-amp utility service, and Lander said SPAN’s analysis shows about 80 amps are typically idle. That headroom defines the ceiling for a single node. “Virtually all homes with 200-amp utility services have 80 amps available at all times, so we set that as the maximum power consumption for a single XFRA node,” Lander said. The nodes “operate as always-on loads within verified residential capacity,” running continuously under normal conditions.
What homeowners actually get
For people who agree to live with a data center bolted to their wall, the financial arrangement looks generous on paper. SPAN takes over both the electricity and internet bills, then either charges a flat fee – the example mentioned to Realtor.com was $150 per month – or nothing at all. Internet service plan details are still being worked out.
The company insists daily life inside the house would feel normal. Appliances would run as expected because the node uses only the slack in the home’s electrical service. If consumption ever spikes, the home battery kicks in first to keep the GPUs fed. Should that not be enough, the system temporarily throttles “non-critical flexible loads” like EV charging. Homeowners can use the PowerUp app to set their own priorities for which loads get curtailed and in what order. Lander said those interventions would be “rare and brief.”
Full node shutdowns only happen during power outages, utility demand response events, or safety triggers. When that occurs, the affected workload shifts elsewhere on SPAN’s network and the household keeps using the backup battery for its own circuits. “This home backup is provided to the host at no cost to them, contributing to greater energy resilience in addition to affordability,” Lander said.
Why utilities might like this idea
Power companies across the country are wrestling with AI-driven demand growth that often requires expensive grid upgrades, with the costs eventually showing up on customer bills. SPAN argues distributed nodes use grid capacity that already exists. “Networks of XFRA nodes make electricity more affordable for the entire community because they increase sales over grid infrastructure that already exists, saving utilities from costly upgrades to support big data centers,” Lander said.
Ari Peskoe, director of the Electricity Law Initiative at Harvard Law School, called the homeowner subsidy “fascinating”, but raised a practical question about clustering. “If there’s a block that has several homes with these devices, maxing out compute and energy would force a lot of power to that local area,” Peskoe said. Local distribution networks weren’t designed for sustained heavy industrial loads dropped into residential streets.
The case for edge inference
Benjamin Lee, a computer architect and engineer at the University of Pennsylvania, sees technical logic in the distributed model – at least for the right workloads. “Computation for AI inference can and should be distributed at the ‘edge,’ deployed on smaller platforms closer to population centers and users,” he said. “The strategy could impose much smaller impacts on the grid because inference requires a few GPUs, unlike training which requires thousands of them working in concert.”
Training the largest AI models will keep happening at the centralized hyperscale facilities being built by Google, Microsoft, and similar companies. SPAN’s nodes would handle the downstream work: cloud gaming, video streaming, and applying already-trained models to user queries. But inference tasks vary widely – answering questions about a document is a different beast from generating code or holding a long conversation – so each node will need enough horsepower for whatever lands on it, plus reliable connectivity to the rest of the network.
Lee also wondered whether shrinking compute all the way down to “the granularity of a few GPUs” is even necessary. Building 20-megawatt data centers rather than 1-gigawatt megastructures might solve much of the same problem with fewer logistical headaches.
Security, theft, and the curious neighbor
Distributing pricey hardware into thousands of suburban backyards creates risks that walled, guarded facilities don’t have. Side-channel attacks – the class of exploits that pull data out of a chip by measuring power draw, electromagnetic emissions, or timing – generally need close physical access. “Many side-channel attacks require physical proximity to the machine, which data centers can guard against,” Lee said. “Distributed GPUs in individual homes are much more difficult to protect.”
Then there’s the simpler problem of someone walking off with the hardware. Each Nvidia RTX Pro 6000 Blackwell sells for roughly $10,000, and a node contains sixteen of them. Reddit threads on the announcement have already speculated about theft, with a few commenters joking that the host might be the first suspect. “Of course, there is the risk of losing the actual hardware itself to theft,” Lee said.
What comes next
SPAN plans the 100-home pilot in 2026 and aims to ramp to 80,000 nodes nationwide starting in 2027, delivering more than 1 gigawatt of distributed compute capacity. The pilot will produce the data needed to evaluate whether the economics, reliability, and security claims hold up outside a slide deck.
While other parts of the AI industry chase ideas like orbital data centers and floating offshore facilities, the suburban version might be the most grounded of the bunch – assuming homeowners, neighbors, and HOAs don’t push back once the first nodes start humming next to someone’s fence.
Written by Alius Noreika
