California Company SPAN Partners with NVIDIA to Install $250,000 Worth of GPU Equipment for Free on New Home Exteriors
California startup SPAN has partnered with NVIDIA and homebuilder PulteGroup to launch XFRA distributed computing nodes, installing approximately $250,000 worth of GPU equipment for free on new home exteriors.
Each node includes 16 NVIDIA RTX PRO 6000 Blackwell GPUs, 4 AMD EPYC CPUs, and 3TB of memory, featuring a liquid cooling design, with noise levels around 60 decibels, and a size similar to an air conditioning unit.
The equipment utilizes the idle power capacity of homes (with an average 200 amp residence having about 40% unused) to run AI inference workloads, with SPAN owning and operating all hardware.
Homeowners do not need to pay installation fees, as SPAN covers electricity and internet costs, charging only a fixed fee of about $150/month, and provides backup batteries and smart panels.
The pilot program is set to launch in Q3 2026 in approximately 100 new homes in the Southwest, aiming to deploy tens of thousands of nodes by 2027, creating gigawatt-level distributed computing power.
Market mechanisms indicate event-driven expansion of distributed computing power: NVIDIA and SPAN benefit from low-cost rapid deployment, avoiding the approval and power bottlenecks of large data centers; traditional centralized data centers are under pressure, while homeowners reduce energy costs through subsidies.
Supplementary data shows this model claims to deploy equivalent computing power six times faster and at one-fifth the cost of traditional data center deployments.
Source: Public Information
ABAB AI Insight
SPAN previously focused on smart home panels, now extending its power management capabilities to distributed computing, with NVIDIA providing Blackwell GPU support for edge inference needs; both parties have previously explored collaboration in the intersection of energy and AI.
In terms of capital pathways, resources are shifting from large data center construction to utilizing idle household power, motivated by the need to quickly fill the AI inference power gap while reducing deployment resistance through homeowner subsidies.
Similar cases can be seen in early edge computing and home storage projects, as well as some cloud vendors attempting distributed nodes; the current phase is exploring the transition of AI infrastructure from centralized to hybrid distributed models.
Structural judgment indicates a technological substitution: utilizing existing power grids and residential spaces to replace new data center construction, with the mechanism being the combination of liquid-cooled GPUs and smart panels, allowing high-density computing to be embedded in living environments without disruption to daily life.
ABAB News · Cognitive Law
Idle power is the new type of mine
Computing power is descending to the doorstep
Free hardware in exchange for long-term power leases