Hardware guide
What makes an “AI PC”? NPUs, TOPS and memory, explained
Updated October 28, 2026 · By the SuperiNow team
"AI PC" is printed on a lot of laptop boxes now. Some of those laptops will run AI features smoothly for years. Others just have a sticker. Three specs tell them apart.
1. The NPU: a chip made for AI math
An NPU (neural processing unit) is a part of the processor built for the kind of math AI models do. It runs things like live captions, background blur, noise removal and on-device assistants using far less power than the main processor or graphics chip would. That matters most on battery.
- Windows laptops sold as Copilot+ PCs must meet Microsoft's minimums, which include an NPU rated at 40 or more TOPS and at least 16 GB of memory.
- Apple's M-series and recent A-series chips include their own AI block, which Apple calls the Neural Engine (see the MacBook Pro and MacBook Air spec pages).
2. TOPS: useful, but not the whole story
TOPS means trillions of operations per second. It's the headline number for an NPU, and higher is generally better. Two catches:
- Companies don't always measure it the same way. TOPS figures are usually quoted at low precision, which inflates them compared with other ways of measuring. Compare TOPS between laptops from the same generation, not across years or brands.
- An NPU only helps if the app uses it. Many AI apps still run on the graphics chip or in the cloud. Check that the features you care about actually run on the NPU.
3. Memory: the spec people underestimate
AI models have to fit in memory to run on your machine. More memory means bigger, smarter models, and fewer slowdowns while other apps are open.
| You want to... | Aim for |
|---|---|
| Use built-in AI features (captions, effects, search) | 16 GB |
| Run small open models on your own laptop | 16–32 GB |
| Run larger open models, or image generation | 32 GB or more, or a desktop graphics card with plenty of video memory |
A rough rule for running open models yourself: a model needs about half a gigabyte of memory per billion parameters when it's compressed to 4-bit, plus room for everything else. So a 7-billion-parameter model needs roughly 4–5 GB just for itself.
What to skip
- "AI-enhanced" with no NPU listed. If the spec sheet doesn't name an NPU and its TOPS rating, the AI is probably all in the cloud.
- 8 GB of memory, if you want AI features to stay usable for a few years.
- Paying extra for TOPS you won't use. If the AI you rely on runs in the cloud (most chatbots do), a strong NPU won't make it faster.
The short version: for most people, an AI-ready laptop means a named NPU, 16 GB of memory or more, and good battery life with AI features turned on. Check current specs on the maker's site before you buy. This field changes every few months.