The newest AI boom pitch: Host a mini data center at your home

SPAN, a San Francisco startup, is piloting a program to install mini data centers, called XFRA nodes, in homes. This "distributed data center solution" aims to significantly expand compute capacity for AI workloads by utilizing excess power capacity in US households. The company plans a 100-home trial this year, with a vision to scale to 80,000 nodes across the US by 2027.

Homeowners participating in the program would receive subsidized electricity and Internet access, along with backup batteries. SPAN would cover utility bills, potentially offering a flat fee or even no charge. This approach could avoid the substantial land and water usage associated with traditional, large-scale data centers, and mitigate community opposition.

The XFRA nodes, equipped with liquid-cooled Nvidia RTX Pro 6000 Blackwell Server Edition GPUs and AMD EPYC Server CPUs, are designed to operate quietly and discreetly. Each node would typically consume 80 amps, well within the 200-amp service capacity of most modern US homes. A 16 kilowatt-hour battery and SPAN's PowerUp software would manage energy consumption and provide backup during power outages.

This distributed network is not intended to replace hyperscaler data centers used for intensive AI model training. Instead, it would support applications like cloud gaming, content streaming, and AI inference. SPAN emphasizes that this model makes electricity more affordable for hosts and communities by increasing sales over existing grid infrastructure, thus reducing the need for costly upgrades.