Sources
NVIDIA shipped PAIR, an Apache-2.0 router that pools every GPU on your LAN — including Apple M4+ Macs — behind Ollama and LM Studio
Announced at IFA on 2026-09-03, Personal AI Router auto-discovers participating machines on a home network, tracks readiness, schedules independent jobs and streams responses back, with no changes to agent harnesses. Supported hardware spans GeForce RTX 20-series and newer, RTX PRO Turing and newer, DGX Spark, and Apple M4+ silicon. NVIDIA's own five-subagent demo with Qwen 3.6 35B A3B went from 18 minutes on a single RTX Spark laptop to 8 minutes 48 seconds on a three-device cluster, which they explicitly label a configuration-specific demo rather than a benchmark. Source is on GitHub at NVIDIA/Personal-AI-Router (609 stars, Apache-2.0).
↳ Follow the thread