Buying a Laptop for Local AI: What to Check Before You Spend
Note: Some future links on this site may be affiliate links. We never claim hands-on testing; guides are researched from the public sources listed in each article.
How we wrote this: A buying guide based on official documentation from Ollama and Microsoft (linked below). We have not tested any laptop. Specs and support lists change, so confirm against the manufacturer's page before you buy.
"AI laptop" is now a marketing label on almost everything. If your goal is to run language models locally, which means on the laptop and not through a cloud service, only a handful of specs really matter. Here they are, in order of importance.
1. Memory decides which models you can run
A model has to fit in memory to run at a usable speed. On a laptop with a dedicated GPU, that means video memory (VRAM). On systems where the CPU and GPU share one pool of memory, it's the total system memory.
How much is enough depends on the model size, the quantization and the context length. Ollama's context-length documentation shows how strongly memory shapes the experience. It sets a 4k-token default context under 24 GiB of VRAM, 32k from 24 to 48 GiB, and 256k at 48 GiB or more.
Before you buy, run the numbers for the models you want with the Private AI Computer calculator. The guide on memory for 70B models explains why the biggest open models are out of reach for most laptops.
Rule of thumb: laptop memory usually can't be upgraded later. Buy the most you can afford.
2. GPU support is about software, not only speed
A fast GPU only helps if your software supports it. Ollama's GPU documentation lists:
- NVIDIA: compute capability 5.0 or newer, with driver version 550 or newer (cards from 5.0 to 6.2 need driver 570+).
- AMD Radeon: a specific list of cards via ROCm (ROCm v7 on Linux), with more AMD cards supported through Vulkan.
On Windows, the Ollama Windows page requires Windows 10 22H2 or newer, with NVIDIA driver 551.61+ for NVIDIA cards. Check the laptop's exact GPU against these lists, not just the brand.
3. NPUs: useful, but check what uses them
Many new Windows laptops include a Neural Processing Unit (NPU). Microsoft's Copilot+ PC developer guide describes Copilot+ PCs as Windows 11 machines with an NPU that can perform more than 40 trillion operations per second (TOPS). Developers reach that NPU through tools such as ONNX Runtime.
What that means for you: an NPU powers built-in Windows AI features and apps written for it. Many popular local-LLM tools run mainly on the GPU and CPU, so don't assume an NPU makes general chat models faster. Check whether the specific app you plan to use supports NPUs.
4. Mac or PC?
Apple Silicon Macs share one memory pool between the CPU and GPU, which can let them load larger models than a PC laptop with a small VRAM budget. PCs with NVIDIA GPUs have the broadest software support. Both paths work. Private AI Computer has a full comparison: Mac or PC for local AI?
5. Storage and the boring stuff
- Storage: models are large files, often several gigabytes each. Leave room for a few of them on top of your normal use. Ollama lets you move the model folder with the
OLLAMA_MODELSsetting (FAQ). - Cooling and power: long AI sessions are sustained loads. Laptops can slow down when hot or on battery, so expect the best speeds when plugged in.
Pre-purchase checklist
- Memory (VRAM or unified) fits the models you want, per a calculator
- GPU model appears in your software's support list
- If buying for an NPU, your apps actually use it
- At least a few hundred GB free for models
- Return window long enough to try your real workload
If your budget is a desktop rather than a laptop, see Private AI Computer's best private AI computer under $1,000 and its related guides.