Copilot+ PCs marked a genuine hardware inflection point for Windows: the first mainstream laptops built around a dedicated AI accelerator rather than treating AI workloads as an afterthought for the CPU or GPU. Here’s what actually changed under the hood.
The 40+ TOPS Requirement
To carry the Copilot+ branding, a PC needs an NPU (Neural Processing Unit) capable of at least 40 trillion operations per second (TOPS). That threshold enables a specific set of on-device AI features — Windows Studio Effects, Live Captions with real-time translation, and semantic search over your file system — to run locally without a network round-trip.
Why On-Device Matters
- Latency — background removal in a video call has to happen in milliseconds; a cloud round-trip isn’t fast enough.
- Privacy — features like Recall’s screenshot indexing process everything locally by design, since sending a rolling screen history to the cloud would be a non-starter for most users.
- Battery efficiency — an NPU performs matrix operations far more efficiently per watt than running the same workload on a CPU or GPU.
Checking Your Own Machine’s NPU
# List AI-capable devices recognized by Windows
Get-CimInstance -Namespace root\cimv2 -ClassName Win32_PnPEntity |
Where-Object { $_.Name -match "Neural|NPU" }
# Task Manager's Performance tab also shows an NPU utilization graph
# on Copilot+ certified hardware, next to CPU/GPU/Memory
What This Means Going Forward
Expect the NPU to become as standard a spec line as GPU or RAM within a couple of hardware generations, with more Windows features quietly shifting from cloud-dependent to local-first as the installed base of capable NPUs grows.