Walk into any electronics store and you will see laptops labeled “AI PC”. The term refers to computers with a neural processing unit (NPU), a chip component designed to run machine learning tasks efficiently. But what does that mean for you, and should it change what you buy?
Updated September 2026: we added links to Microsoft’s and Apple’s hardware requirements and to OpenAI’s guidance on running open models locally.
CPU, GPU and NPU
| Processor | Designed for | AI role |
|---|---|---|
| CPU | General computing, a bit of everything | Can run small AI tasks, slowly |
| GPU | Parallel work such as graphics | Fast for AI, but power-hungry |
| NPU | The math behind neural networks | Efficient for sustained, low-power AI tasks |
The NPU’s advantage is efficiency. It can run continuous AI features, such as background blur on video calls or live captions, without draining the battery or heating up the laptop.
What runs on the NPU today
Features vary by operating system and manufacturer, but typical examples include:
- Video call effects: background blur, eye contact correction, noise suppression
- Live captions and translation
- On-device image editing and generation features
- Semantic search of files and settings
- Some writing assistance and summarization
Microsoft’s Copilot+ PC program requires an NPU that can run at more than 40 trillion operations per second (TOPS) for many of its newer Windows AI features; qualifying chips include Qualcomm’s Snapdragon X Elite, AMD’s Ryzen AI 300 series and Intel’s Core Ultra 200V series. On the Apple side, Apple Intelligence needs a Mac or iPad with an M1 chip or later, or an iPhone 15 Pro or newer.
What the NPU does not do
The most capable AI assistants, like the full versions of ChatGPT, Claude and Gemini, still run in the cloud. An NPU does not make them faster or smarter. Most people’s heaviest AI use therefore depends on internet connection and subscriptions, not laptop hardware. We explain the split in our cloud vs on-device comparison.
For running open-weight models locally, as described in our open vs closed models guide, the amount of memory and the GPU often matter more than the NPU. OpenAI says the smaller of its gpt-oss open-weight models can run on devices with 16 GB of memory, while the larger one needs a single 80 GB GPU.
Should it affect what you buy?
It may matter if you:
- Spend many hours on video calls and value battery life
- Want specific on-device features, such as those limited to Copilot+ PCs
- Want to keep sensitive tasks on the device for privacy
- Plan to keep the laptop for many years, as more software will use NPUs
It matters less if you:
- Mainly use cloud AI assistants in a browser
- Need a laptop for other priorities, such as gaming, specific software or budget
Buying advice The usual priorities still apply: enough memory (16GB is a sensible minimum for a new laptop with AI ambitions, more for local models), good battery life, a quality screen and keyboard. Treat the NPU as a bonus, not the headline.
TOPS is not everything
Manufacturers advertise NPU speed in TOPS. Higher is generally better, but real-world performance also depends on software support, memory and how well features are optimized. Compare actual features, not just numbers.
Should you buy an AI PC?
An AI PC is a laptop with an efficient chip for on-device AI features. It is a worthwhile extra, particularly for battery life and privacy-friendly features, but it does not replace cloud AI. Buy for your real needs first.
Sources
- Copilot+ PCs developer guide, Microsoft Learn
- How to get Apple Intelligence, Apple Support
- Introducing gpt-oss, OpenAI, August 2025
