How the PC layer decides which open-weight model becomes the default on-device runtime The OS vendor and OEM act as gatekeeper: they decide which model the fleet runs. Neutral selection disperses the fleet across open families; a single integrated pick concentrates the whole fleet on one runtime. device fleet generative AI moves on-device Cloud inference capacity, cost, latency, energy Qwen permissive license Llama open-weight Gemma open-weight PC distribution layer OEMs Win10 EOS refresh OS vendor Copilot+ · Apple select NPU silicon Copilot+ TOPS bar Intel Qualcomm AMD Apple dispersion permissive license + full size spectrum wins developer mindshare concentration one integrated runtime saturates the fleet, regardless of benchmark lead