Back

Nvidia and PulteGroup Partner With Span on In-Home AI Mini Data Centers

The companies say each residential unit includes 16 Nvidia Blackwell GPUs, four AMD EPYC CPUs and 3 terabytes of RAM for AI inference.

Summary

Nvidia and homebuilder PulteGroup are partnering with Span to deploy in-home mini data centers designed for AI inference (running trained AI models). According to the provided announcement, each unit will contain 16 Nvidia Blackwell GPUs, four AMD EPYC CPUs and 3 terabytes of RAM, and will use unused household electricity as its power source. The plan points to a distributed computing model, where compute capacity is placed inside homes rather than centralized in traditional data center facilities.

Terms & Concepts
  • AI inference: The process of using a trained artificial intelligence model to generate outputs or predictions from new data.
  • GPU: Graphics processing unit, a chip widely used to accelerate artificial intelligence workloads because it can process many calculations in parallel.
  • Distributed computing: A model that spreads computing workloads across multiple physical locations or devices instead of relying on one central facility.