The term “AI laptop” appears in many product listings today. It is worth knowing what it really means. An AI laptop is simply a computer whose processor includes a dedicated unit for artificial intelligence tasks - an NPU. Not everyone needs one, though. Most popular AI tools, such as ChatGPT, Claude, Gemini and Microsoft Copilot, run in the cloud. In that case, the computing is done by the provider’s servers, not by your laptop.
- what an NPU is and what TOPS figures mean,
- what an NPU is actually useful for – and what it is not,
- which laptop to choose for the way you work with AI – including ex-lease devices.
What is an NPU (Neural Processing Unit) and how does it work?
An NPU (Neural Processing Unit) is a specialised chip for neural network computations. Neural networks perform millions of simple multiplications and additions on large sets of numbers. An NPU has many units that carry out these operations in parallel. It usually works with lower-precision numbers (e.g. 8-bit), which saves energy.
That is why an NPU is efficient at AI tasks but is not suited to regular programs. It does not replace the CPU or GPU – it takes over selected, repetitive AI tasks from them.
CPU, GPU and NPU – division of roles in a modern computer
| Chip | What it is for | Example AI task |
|---|---|---|
| CPU (processor) | Operating system, programs, logic | Simple AI computations in applications |
| GPU (graphics) | Graphics and massively parallel computing | Local language models, image generation |
| NPU | Continuous, energy-efficient AI tasks | Camera background blur, noise reduction |
The heaviest AI tasks are still handled by the GPU. The NPU works best where AI runs in the background for long periods.
What does TOPS mean and why is 40 TOPS the threshold for Copilot+ PCs?
TOPS (Trillions of Operations Per Second) is an approximate measure of NPU performance. Figures quoted by different manufacturers are not always fully comparable.
Microsoft requires an NPU with at least 40 TOPS for a Windows 11 computer to carry the Copilot+ PC label. Copilot+ PCs are a new class of Windows computers designed specifically to run AI-based features locally. Only these devices support some of the new system features that run on the device.
- Meet the threshold (examples): Qualcomm Snapdragon X, Intel Core Ultra 200V, AMD Ryzen AI 300.
- Have an NPU, but below the threshold: Intel Core Ultra Series 1 (e.g. Core Ultra 7 165H), AMD Ryzen 7040 and 8040.
The 40 TOPS threshold applies only to Copilot+ PC certification in Windows. It is not a general line that defines a “real” AI laptop.
NPU vs Intel DL Boost and GNA – what is the difference?
Older Intel processors include features that accelerate AI, but they are not NPUs:
- Intel DL Boost – a set of processor instructions (e.g. in 11th Gen Intel Core). It speeds up AI computations on the CPU.
- Intel GNA – a small accelerator, e.g. for microphone noise reduction.
So a laptop with DL Boost is not a laptop with an NPU. It will not support Copilot+ PC features.
Neural Engine in MacBooks – Apple’s NPU
Every MacBook with an Apple M-series chip (M1, M2 and newer) has an NPU. Apple calls it the Neural Engine. According to Apple, it delivers about 11 TOPS in the M1 and about 15.8 TOPS in the M2. The 40 TOPS threshold does not apply here, as it concerns Windows only.
What is an NPU in a laptop used for? Real-world uses
An NPU is mainly useful for AI tasks that run locally and in the background. Below, we describe what it does – and what it does not.
1. Video conferencing effects
Background blur, automatic framing and eye contact correction run throughout the whole call. In Windows 11, these effects are provided by Windows Studio Effects, which requires a compatible NPU. The NPU handles them without loading the processor, so the computer stays responsive.
2. Windows 11 features for Copilot+ PCs
On Copilot+ PCs, Microsoft offers features that run on the device. Examples include live captions with translation (Live Captions), Recall and Click to Do. The list of features and their availability in individual markets change with updates.
3. Lower energy use for AI tasks
NPUs are designed for low power consumption. For long-running AI tasks, an NPU usually uses less energy than a CPU or GPU. The actual effect on battery life depends on the model and how the device is used.
4. What an NPU does not do
Many features associated with an “AI laptop” run in the cloud, not on the NPU:
- Microsoft 365 Copilot – document summaries, Excel formulas, email drafts in Outlook,
- ChatGPT, Claude, Gemini, Perplexity – used in a browser or an app.
For these tools, a reliable laptop, enough RAM and a stable internet connection are all you need. An NPU will not make them run faster.
NPU or GPU – which laptop to choose for AI?
The choice depends on where the AI you use actually runs. We distinguish three user profiles.
Profile 1: cloud-based AI assistants – no NPU needed
| Model | Processor | RAM |
|---|---|---|
| HP EliteBook 840 G8 | Intel Core i5-1145G7 (11th Gen) | 16 GB |
| Lenovo ThinkPad T14 Gen 2 | Intel Core i5-1145G7 (11th Gen) | 16 GB |
| Dell Latitude 7420 | Intel Core i7-1185G7 (11th Gen) | 16 GB |
| Microsoft Surface Laptop 4 15" | Intel Core i7-1185G7 (11th Gen) | 32 GB |
These laptops have Intel DL Boost, but they do not have an NPU and are not Copilot+ PCs. For working with cloud-based AI tools, they are more than enough.
Profile 2: laptop with an NPU – local AI features
Among ex-lease devices, you will most often find an NPU in MacBooks with Apple M-series chips. Windows laptops with an NPU are still rare on the ex-lease market.
| Model | NPU | Memory |
|---|---|---|
| Apple MacBook Air M1 | Neural Engine, approx. 11 TOPS | 16 GB |
| Apple MacBook Pro 13" M2 | Neural Engine, approx. 15.8 TOPS | 16 GB |
| HP ZBook Firefly 16 G11 | Intel AI Boost (Core Ultra 7 165H), approx. 11 TOPS | 32 GB |
Profile 3: local models and image generation – a workstation
Here, an NPU cannot replace a powerful GPU and plenty of memory. You have two options.
Option A: Windows and an NVIDIA RTX graphics card. RTX cards have Tensor cores and support CUDA, which most AI tools rely on.
| Model | Graphics card | RAM |
|---|---|---|
| Dell Precision 7550 | NVIDIA Quadro RTX 3000, 6 GB VRAM | 64 GB |
| Dell Precision 5560 | NVIDIA RTX A2000, 4 GB VRAM | 64 GB |
This is entry level for local AI. It is enough for smaller language models and basic versions of Stable Diffusion. Larger models need more VRAM.
Option B: a MacBook with large unified memory. In MacBooks, the graphics unit uses a shared memory pool. This lets it run larger models than a card with 6 GB VRAM.
| Model | Chip | Memory |
|---|---|---|
| Apple MacBook Pro 16" M1 Max | Apple M1 Max | 64 GB |
| Apple MacBook Pro 16" M2 Max | Apple M2 Max | 64 GB |
Ex-lease laptop for AI – what to look for?
Ex-lease equipment is a good choice if you match it to your tasks. Before you buy, go through this checklist:
- Where does your AI run? For the cloud, a reliable business laptop is enough. For local use, you need a more powerful configuration.
- Operating system: does the laptop support Windows 11 or a current version of macOS?
- RAM: 16 GB for office work, 32 GB or more for programming and local models.
- Graphics card: for local models, choose an NVIDIA RTX series card. Check the amount of VRAM.
- NPU: look for it in the processor specification. The word “AI” in a listing title is not enough.
- Battery and warranty: check the battery condition and the warranty terms.
Choosing refurbished equipment also extends the life of a device that has already been manufactured.
Summary – is a laptop with an NPU worth it?
NPUs are becoming standard in new processors, but not everyone needs one today.
- Do you use AI in the cloud? A reliable business laptop with 16 GB RAM is enough – an NPU is not necessary.
- Do you want local AI features and long battery life? Choose a laptop with an NPU, e.g. a MacBook with an Apple M-series chip.
- Do you run local models or generate images? The key is an NVIDIA RTX card with enough VRAM, or a MacBook with large unified memory.
Frequently asked questions about NPUs (FAQ)
What is an NPU in a laptop?
It is a unit in the processor for AI tasks, such as camera background blur. It works energy-efficiently and takes load off the CPU and GPU.
Does an NPU speed up ChatGPT or Microsoft 365 Copilot?
No. These tools run in the cloud, so a reliable laptop with 16 GB RAM and a stable internet connection are enough.
How many TOPS does an AI laptop need?
The 40 TOPS threshold applies only to Copilot+ PCs with Windows 11. For cloud-based AI tools, an NPU is not needed.
NPU or GPU for local AI models?
GPU. What matters is an NVIDIA RTX card with enough VRAM, or a MacBook with large unified memory.