Running a 72GB AI model on a laptop via external GPU — setup walkthrough
A user connected an RTX Pro 5000 external GPU (72GB memory) to a ThinkPad laptop to run large AI models locally. They also set up Tailscale so the server is reachable from anywhere. It's a niche setup that requires expensive hardware.
Running large language models (LLMs) locally usually needs a GPU with a lot of memory — more than most laptops have built in. This person solved that by attaching an RTX Pro 5000 Blackwell graphics card with 72GB of VRAM to a ThinkPad P52 via an eGPU enclosure — an external box that connects a desktop-grade graphics card to a laptop, typically over Thunderbolt.
Tailscale ties it together by letting the owner securely access the home server from anywhere on the internet. For anyone wanting to self-host a heavyweight model like Hermes locally rather than using a cloud service, this shows it's possible — but the cost of the hardware puts it out of reach for most people.
Key points
- The RTX Pro 5000 has 72GB of VRAM, enough to load most large AI models entirely into memory
- An eGPU enclosure lets a laptop use a powerful desktop graphics card via a cable (usually Thunderbolt)
- Tailscale allows secure remote access to a home AI server from outside the house
- The hardware costs involved make this impractical for most everyday users
Quick term guide
- AI models
- The core brain or underlying program that powers an artificial intelligence tool.
- Tailscale
- A tool that lets you securely access your home devices over the internet, as if they were on the same local network
- hardware
- The physical parts of a computer that you can touch.
- large language model
- The type of AI behind ChatGPT or Claude — trained on huge amounts of text to read, write, and code.
- graphics card
- A component inside a computer that handles displaying images and running game visuals.
- home server
- A personal computer setup at home used to run services or store files instead of regular daily use.
- self-host
- To run a website, app, or service on your own server instead of using a hosted provider.
- remote access
- Connecting to and controlling a computer from another place.