Robots assembling the servers that train AI. 🔌
Cambrian Robotics built a robot that assembles AI servers, specifically, it connects all those tiny connectors (QSFP and OSFP plugs) that link server components together.
I had a chance to see their demo with Universal Robots arms during @AutomateShow in Chicago week ago.
This is manual work right now. Tedious. Precise. Expensive.
The robot uses Cambrian Vision to locate connectors with 0.1 mm precision, then picks and inserts them.
That precision matters because if a connector isn't fully seated, it causes rework. The system verifies insertion and flags any issues before they become costly problems downstream.
Off-the-shelf RGB cameras, but custom AI trained on synthetic data. They're not using expensive specialized hardware. They're using standard cameras plus smart algorithms to see what matters.
Data centers are under pressure to automate every step of server assembly. Right now each step is a bottleneck. Automate one step, move to the next.
Hyperscalers are pushing for end-to-end automation, robots building entire servers.
This loop is wild: robots build servers → servers train AI models → those models improve the next generation of robots. Closed loop.
Automation scaling faster because the AI powering the robots keeps getting better.
The machines building the infrastructure that powers AI that's building the next machines.
~~
♻️ Join the weekly robotics newsletter, and never miss any news → https://t.co/GoA3ZuwoPB
New Anthropic research: Emotion concepts and their function in a large language model.
All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.