I Put a Qwen3–32B AI Server in Shenzhen — and Learned How to Operate It from Anywhere

Mon, 01 Jan 0001 00:00:00 +0000

Running Ollama on AMD Ryzen AI Max, using Alibaba Cloud Session Manager without exposing the server to the public internetMany people discuss “Chinese AI” from the outside.

They compare benchmark scores, read company announcements, and debate whether Chinese models are catching up with or surpassing American models.

My approach is simpler:

Before forming an opinion, I want to use the technology myself.

I recently built a local AI server in my Shenzhen office using a Windows PC equipped with:AMD Ryzen AI Max+ 395128 GB of unified memoryRadeon 8060SQwen3–32B Q4_K_MOllama on WindowsQwen3–32B runs with 100% GPU offload and generates roughly 10 tokens per second.

Artificial-Intelligence on ニコ技深セン / Nico-Tech Shenzhen website

I Put a Qwen3–32B AI Server in Shenzhen — and Learned How to Operate It from Anywhere