I started learning about robots in January 2025, and there is so much to learn! I plan to write a little about what I learn about robots, papers I come across, or companies I know of that are doing cool things in the robot space. For this post, I’ll simply outline the 10,000 foot view of what’s happening in robotics.
A lay of the land
AI robot companies generally fall into a few buckets - note that this is a gross oversimplification, and companies can of course function in multiple buckets.
- Research - Focusing on building foundation models, trying to hit scaling laws and performance for robots as we have seen in LLMs. The leaders here are large labs in the US, including Physical Intelligence, Gemini Robotics, etc.
- Hardware - Building actuators, motors, sensors, cameras, and other equipment required for robots. Many of these companies are in China, and the hardware expertise is unparalleled. There are many strong companies in other parts of the world too, of course.
- Data - Largely an operations issue. Aware that data is the bottleneck, these companies sell data to large labs. I’ve received a few propositions from companies in this area asking me to join them. Operations generally in a country or territory with lower cost of labour. Plenty of such companies sprouting up over the past few months.
- Specific software - Instead of focusing research, hardware, or operations, these companies focus on a slightly more specific part of the robotics stack. This can be sensors, vision, teleoperation, etc.
- Deployment - Companies that focus on getting robots deployed and into the real world. My assessment is that most of these companies are not initially from the general purpose / AI robotics field, but industrial robots with AI on top. Nothing wrong or bad with that, and in fact I think it’s the most sensible and revenue making approach.
- Simulation - This is not the area I function in, but there are many well-capitalised companies that build world models or simulation with the promise that this will be a huge data unlock for robots. I haven’t seen significant results here (though have seen some very interesting papers mixing simulation and real world data). It’s pretty exciting if that happens.
A few common threads in these companies? Most are in the capital burning game. I’m not particularly fond of that, and very much prefer a situation of decent revenue growth (or, even better, profit growth).
DNMZ