NVIDIA PAIR pools idle PCs for local AI

NVIDIA launched PAIR, its free open-source Personal AI Router alongside a reported 1.9x local-inference throughput gain for optimized llama.cpp on an RTX 5090, according to NVIDIA’s IFA announcement. PAIR discovers compatible Windows, Linux, macOS, DGX Spark, and Apple M4+ machines, then routes requests through Ollama or LM Studio so idle PCs can share work without special cables or a rack. For builders with spare local hardware, this is a private burst pool—not shared memory or a guarantee of 1.9x end-to-end agent speed; the archive’s long-context llama.cpp result shows where local gains actually compound.