Microsoft announced Monday that all Copilot+ PCs sold this year will include a neural processing unit (NPU) designed specifically to run large language models without requiring an internet connection. The NPU marks a hardware shift for Windows laptops, enabling the company’s Copilot chatbot to operate offline for the first time since its launch.
Copilot debuted in 2023 as a replacement for the discontinued Cortana, built on OpenAI’s GPT-4 and GPT-5. Until now, every interaction required a live internet connection. The NPU changes that by handling the mathematical computations behind large language models locally, reducing battery drain and freeing the main CPU from that workload.
Key features of Copilot will work without a network connection. Latency drops, and the assistant remains available even when the user is offline.
For developers, the NPU enables local inference. For users, the assistant no longer pauses while waiting for the cloud to respond. Microsoft first signaled its hardware ambitions at Build 2023, when it announced plans to integrate Copilot into Windows 11 via the taskbar.
That was a software update. Monday’s announcement completes the transition to hardware.
The NPU is a specialized processor built for the math behind large language models. The timing follows a series of AI product moves by Microsoft. The company introduced its chatbot as Bing Chat in February 2023.
By the end of that year, it had unified its various AI products under the Copilot brand. The “copilot” analogy—an AI that works alongside the user—was established.
Monday’s announcement closes the loop by placing the AI brain inside the computer itself. The shift from cloud-dependent to local AI changes the relationship between user and machine. A cloud-dependent assistant is a shared resource. It is slow.
It is subject to outages. It cannot see local files or understand offline workflows.
A local assistant is personal. It is fast. It is always on.
That changes what developers can build and what users can expect. For the PC industry, the stakes are existential.
Laptop sales have been flat. The pitch that a new processor or a thinner chassis will change computing has worn thin. An NPU that runs AI locally is a genuinely new argument for an upgrade.
It is also a lock-in argument. Once users depend on local Copilot features, switching to a machine without an NPU means losing that capability.
Microsoft did not name partners or pricing. It did not say which models will ship first. What it said is that these are the first Windows laptops designed from the ground up for local AI.
That is the claim. The proof will come when developers actually write software that uses the NPU and users actually notice the difference.
Before Monday, Copilot was a cloud service. After Monday, it is a hardware feature. The line between a connected PC and a standalone AI device just got thinner.
