Nvidia links home PCs into a shared pool for running local AI

Nvidia is introducing PAIR, a Personal AI Router that distributes model inference across compatible computers on the same local network. The tool is meant to combine otherwise idle GPU capacity so people can run larger or multiple AI workloads without sending every request to a cloud service.

The announcement is part of a wider local-AI push at IFA 2026. Nvidia says new optimizations for llama.cpp and vLLM can deliver up to 1.9 times faster local inference and are available directly as well as through LM Studio and Ollama. Simplified Nvidia GPU support is also coming to Hermes Agent, OpenClaw, and Perplexity Portable Computer.

Nvidia and hardware partners Lenovo and Acer plan to ship compact RTX Spark Windows PCs in October. The systems target developers, creators, and enthusiasts who want a dedicated machine for running agents locally. Nvidia also highlighted recent open models suited to its desktop and workstation hardware, including Nemotron 3.5 Lightning, GLM-5.3-Flash, and Qwen3.8 variants.

Local processing can keep data inside a home or office and avoid per-request cloud fees, but the practical limit remains the memory and GPU capacity available across the participating machines. PAIR’s value will depend on hardware compatibility, network overhead, and how smoothly applications can divide inference across several PCs; Nvidia has not turned a mixed collection of consumer devices into unlimited compute.