San Francisco startup SPAN is piloting a distributed data center solution that would install mini data centers in residential homes, offering homeowners subsidized electricity and internet in exchange. The system uses liquid-cooled Nvidia GPUs and aims to provide a more efficient alternative to traditional data centers while avoiding community opposition. SPAN plans to deploy 80,000 units by 2027, targeting inference workloads rather than AI model training.
Background
As AI workloads surge, companies are exploring innovative ways to expand computing capacity beyond traditional data centers. This comes amid growing concerns about the environmental impact and community resistance to large-scale data center projects.
- Source
- Ars Technica
- Published
- May 13, 2026 at 05:59 AM
- Score
- 7.0 / 10