NVIDIA与Span计划利用家庭未充分利用的电力容量,将住宅转变为分布式AI计算节点。其核心是配备16块NVIDIA RTX Pro 6000 Blackwell GPU和4颗AMD EPYC CPU的XFRA节点,并内置电池以确保AI负载稳定运行。该模式旨在规避传统数据中心在土地、审批和电网升级方面的扩张瓶颈。Span计划于2026年第三季度启动100个节点的概念验证,目标是从2027年起实现年容量超过1吉瓦的分布式算力网络,同时为参与家庭降低电费成本。
Nvidia and Span is turning homes into small AI data centers, using spare household electrical capacity to run compute nodes while giving hosts cheaper power and internet.
The idea starts with a mismatch: many homes are wired for 200 amps, but they often use far less than that, so Span wants its smart panel to measure unused capacity and feed an XFRA Node without overloading the house.
Each node is serious hardware, with 16 Nvidia RTX Pro 6000 Blackwell GPUs, 4 AMD EPYC CPUs, and 3TB RAM, which means this is not a hobby server but a rented slice of AI infrastructure.
The battery matters because AI workloads hate interruptions, so it buffers demand spikes, supports demand-response events, and shifts work away if a home loses power.
The business claim is that distributed nodes can avoid the slowest part of data center growth: getting land, permits, utility approval, grid upgrades, and power access.
Span plans a 100-node proof of concept in Q3-26, then aims for more than 1GW of annual capacity from 2027.
---
pv-magazine-usa .com/2026/04/15/span-and-nvidia-to-develop-ai-data-centers-in-your-backyard-lowering-electric-bills/
来源:Rohan Paul · x.com