The best laptop for local AI: what to look for and what to avoid

Buy on memory first. For a Windows or Linux laptop, that means the GPU's VRAM: 8 GB runs 8B models, 12 GB runs 14B, 16 GB runs gpt-oss 20B, and 24 GB runs the 30B class. For a MacBook, it means unified memory, where 32 GB or more opens up much bigger models. Speed comes second.

Laptop GPUs aren't desktop GPUs

A laptop "RTX 4090" is not a desktop RTX 4090. It's a smaller chip with less memory and less bandwidth, and its speed also depends on how much power the laptop maker lets it draw. The names line up badly:

Always check the laptop GPU's memory size, not just its name. Our GPU list has the laptop versions as separate pages.

What each tier runs

Estimates from our calculator, for laptop GPUs:

Here's the RTX 4090 Laptop in full:

These assume the GPU runs at full power. On battery, or in a thin laptop with a low power limit, expect less.

Or a MacBook

A MacBook Pro shares one pool of memory between the processor and the GPU, so a 48 GB or 64 GB model can hold models no laptop GPU can. It's slower per model than a high-end gaming laptop, but silent, cool and runs for hours on battery. The chip matters: Max chips have far more bandwidth than base and Pro chips. See Macs for local LLMs.

What else matters

• Cooling and power. Gaming laptops sustain heavy loads better than thin ones. Run plugged in.
• Fan noise. A laptop GPU at full load is loud. For long jobs, that matters.
• System RAM. 32 GB gives headroom if a model spills over from VRAM, and lets you run small models on the CPU. See system RAM for local AI.
• Storage. Models are big. 1 TB is a sensible minimum.
• No GPU at all? Small models still run on the processor. See running AI on a laptop without a GPU.

Our advice

• On a budget: an 8 GB RTX laptop runs Qwen3 8B well, which covers everyday chat and writing.
• The sweet spot for a gaming laptop: 16 GB of VRAM, for gpt-oss 20B at good speed.
• Biggest models in a laptop: a MacBook Pro with a Max chip and 64 GB or more.
• Don't buy a laptop just for AI if you'll mostly use it at a desk. A used desktop with an RTX 3090 costs less and runs bigger models faster. Our builds page has one from $900 to $1,200.

You can also try a model in your browser first, using your laptop's GPU, in the Studio.