NPU switches between per-task adapters… AMD Ryzen AI 1.9

AMD released Ryzen AI Software 1.9, adding "dynamic LoRA support" that lets its NPUs switch between task-specific LoRA adapters during language model inference, initially targeting Qwen-family text models including Qwen3-4B and Qwen3.5-4B. LoRA fine-tunes small low-rank adapters on top of a frozen base model rather than retraining full weights, so the feature allows a single base model to serve multiple tasks, with adapters selected per user or workload at inference time. The update matters for AMD's on-device AI strategy because it improves the efficiency and versatility of NPUs in Ryzen processors, strengthening its position against rivals in the client AI PC market. Investors should watch adoption of dynamic LoRA in developer toolchains, expansion of supported models beyond Qwen, and how AMD's NPU software cadence stacks up against competing AI PC platforms.