A thousand robots on a production line, and a price target nobody has hit yet.
2026-07-28
Musk gave an Optimus update this week and paired it with a warning that the production ramp will be "long and flat". That is a more interesting sentence than any of the demo footage.
Over a thousand Gen 3 units are working on the production floor at Fremont, on the live line building cars that get shipped to customers. That is the part worth paying attention to: not a stage demo, but a robot doing repetitive work in a building that already exists, where the tolerance for failure is low and someone measures throughput.
The V3 hardware is 173 cm, 57 kg, with 22 degrees of freedom in the hands and 50 actuators. The brain is Tesla's AI5 chip, roughly five times the memory bandwidth of the previous one, with xAI's Grok handling voice. Hands are the giveaway spec. Legs got solved years ago in this field; manipulation is where general purpose either happens or does not.
Production of V3 starts this summer, consumer availability is targeted at end of 2027 at the earliest, and the eventual capacity numbers being discussed are a million V3 units a year and ten million V4. The long term price target is twenty to thirty thousand dollars against a current build cost somewhere between fifty and a hundred thousand plus.
That delta is the whole story. Halving a build cost is normal manufacturing work. Getting a hundred thousand dollar machine to twenty is a different exercise, and it usually means the thing that ships is not quite the thing that was demonstrated.
A humanoid robot with a general purpose brain, network connectivity and physical actuators is, from where I sit, an endpoint with hands. Every assumption in endpoint security is built around devices whose worst case is data loss. The threat model for a compromised device that can pick things up and move around a building has not really been written yet, and I suspect it will be written the same way most of them are: after something happens.