Is your "AI PC" actually good at AI?

Short answer: Not necessarily. "AI PC" is a label for a laptop that has a chip made for AI work, called an NPU. The label does not tell you how well the laptop does the jobs a small business needs, such as summarising contracts or answering customer emails. The only way to know is to measure those jobs. That is what our scorecards do.

What the label means#

Different companies define the term in different ways. Here are two official definitions.

• Intel says an AI PC has three parts built to power AI apps: a central processor (CPU), a graphics processor (GPU) and a neural processing unit (NPU). Intel describes the NPU as the part for sustained AI tasks at lower power. See Intel's explanation.
• Microsoft uses the name "Copilot+ PC". Its minimum specification lists an NPU able to do 40+ TOPS, 16 GB of memory and 256 GB of storage. It names the chip families that qualify: AMD Ryzen AI 300 and 400 series, Intel Core Ultra 200V and 300V series, and Snapdragon X series. See Microsoft's specifications.

What TOPS means#

TOPS means trillions of operations per second. Microsoft describes a Copilot+ NPU as one that can do more than 40 trillion operations per second. See Microsoft's developer guide.

TOPS is a rating of the chip. It does not say how long your contract summary takes. That depends on the whole laptop and on the software you use. This is our view, and it is the reason we time real jobs instead of quoting chip ratings.

The software has to use the NPU#

Microsoft says an NPU is a hardware resource like any other, and that software has to be written to take advantage of it. See Microsoft's developer guide. Its own Windows features are built that way, such as real-time translation and image generation.

Ollama, the runner we test with, works differently. Its hardware page lists graphics processors from NVIDIA and AMD, Apple's Metal graphics, and Vulkan graphics on Windows and Linux. It does not mention NPUs. We read that page on 2 October 2026. See Ollama's hardware page. On a Mac with an Apple chip, Ollama uses the graphics chip through Metal.

So for the Ollama route, the NPU label may not help your jobs. For other software, it may. We have not tested NPU paths. If Ollama adds NPU support, we plan to test again and say so. See how we test.

What an NPU is good for#

The NPU is not useless. Microsoft says it uses less power than a CPU or GPU for AI tasks. See Microsoft's NPU page. That can help battery life for the built-in features that use it. It is a real benefit for those features. It is not a promise about the AI jobs in your business.

Apple's chips also have a Neural Engine. Apple says the Neural Engine in its M5 chip works together with Neural Accelerators in the CPU and GPU for AI work. See Apple's note on the M5 chip.

What decides how a laptop does#

For private AI on a laptop, check these before the label:

1. Memory. The model must fit. This usually matters most. Read how much memory you need and check the tiers.
2. Graphics memory or unified memory. This is where most local apps run the model.
3. Heat. A thin laptop may slow down on long jobs. We run each job three times and report the slowdown.
4. Battery. We report the battery cost of two jobs.
5. The job you do. A laptop can be great at invoices and painful at long contracts.

How to check a real laptop#

1. Open the laptop list and find the model. If it is not there, we have not tested it yet.
2. Read the verdict for each job, not the overall score alone. See contract summaries and customer replies for examples of the job rankings.
3. Check its tier and memory.
4. Compare price per point with similar laptops.

Cloud AI is still stronger than anything a laptop runs, and it is easier to use. Local AI wins on privacy, cost per use and working offline. The scorecards show whether a given laptop is good enough for local work.