Microsoft hypes PCs with NPUs, still can't offer a good reason to buy one (www.theregister.com)

Microsoft is aggressively positioning neural processing units (NPUs) as a core part of the Windows "Copilot+" vision—NPUs, typically packaged with CPUs, accelerate local inference at low power and can run small language models on-device. Microsoft points to features that can leverage NPUs (on-device agents in Settings, Semantic Windows Search, Studio Effects, Notepad/Photos enhancements, and the opt‑in Recall activity log) and argues NPUs make sophisticated AI experiences cheaper and more widely accessible than cloud-only large models. But the practical payoff today is limited: there’s no clear killer app, many features offer marginal productivity gains, and Microsoft hasn’t put NPUs on Windows’ hardware requirements list—yet. The distinction matters because NPUs enable offline, lower-latency, privacy‑friendly workloads and concurrent AI apps, but widespread adoption so far is driven more by marketing and “future-proofing” than real user demand. Industry analysts warn that a future compatibility mandate could force another broad hardware refresh similar to the Windows 11 transition; already, AI-enabled notebooks made up ~40.5% of European distribution in early September. For AI/ML practitioners, NPUs open opportunities for compact on-device models and efficient inference pipelines, but the ecosystem still needs compelling, productivity‑transforming use cases and developer tooling to make NPUs indispensable.