Robotics primer: why robo szn is happening now and crypto's three lanes
Quote367 words
Robotics capital arrived before understanding: Unitree ~$50B, Figure $39B, Skild $14B. Argues the brain is commoditised, actuators are 40 to 60% of humanoid BOM, the moat is teleop data, and crypto's lanes (provenance, capital formation, coordination) total under $1B.
Key claims
Actuators are 40 to 60% of a humanoid's BOM
~1 in 10 humanoids shipped last year does an actual job
Crypto robotics category worth less than $1B
Robotics is the first sector since LLMs where the capital showed up before the understanding did. Unitree opened near $50B. Figure is marked at $39B. Skild raised at $14B and has never shipped a robot.
Still, most people longing the trade couldn’t explain how a robot actually works if you asked them.
That's the gap our newest @a1research__ piece aims to close.
After spending a good part of the last month taking the robotics vertical apart, I came out more bullish than I went in.
But let me explain:
Why is this robo szn happening now and not in 2015?
Because a robot is a loop. It senses, it thinks, it acts. The hardware was solved years ago. The think step was the problem, robots were too dumb to be general, so they stayed bolted to car factories where dumb was fine. Foundation models fixed that in 2023, and the moment the brain stopped being the bottleneck the whole race shifted to the other two steps.
What that race looks like rn:
- The body is a cost problem, not a chips problem. Actuators are 40 to 60% of a humanoid’s BOM and the supply chain runs through China. Beijing’s export lever is magnets, not GPUs.
- The brain is solved and commoditised. Every serious lab runs the same architecture, a slow VLM over a fast policy. Nobody wins on brains anymore.
- The moat moved to data, bc there is no internet of actions. One teleop hour = one data hour, and companies are manufacturing it in halls full of people in VR headsets.
- The open stack is here. A $399 trainable biped did $2.6M in orders in a day.
- Deployments are real but narrow. ~1 in 10 humanoids shipped last year does an actual job. The valuations price general-purpose workers.
Wait, what about crypto?
Crypto has three real lanes in this: data provenance, capital formation and machine coordination. Yet, the entire category is worth less than $1B rn.
The full breakdown, who gets paid in each layer, where the chokepoints sit and the five metrics that tell you whether the thesis is working, is all in the piece below.
Don't fade anon! ↓
QUOTING @a1research__
t.co/a74p0oNwt0
t.co/a74p0oNwt0
1 image on X