Tesla Is Building Hundreds of Optimus Robots a Week. The Hands Keep Breaking
Tesla now builds several hundred Optimus human-shaped robots a week in Fremont, California. A few months ago it built only a few dozen. The goal written on the wall is 20,000 a week.
But there is a catch. Each arm holds more than 100 screws. People still put them together by hand.
That detail comes from a report by The Information. It was cited this week by Electrek and Gizmodo. Their sources are people who know the program well. Tesla did not comment. What they describe is a factory that works, sort of, and a robot that is not ready for you.
Car parts are big. Robot fingers are tiny
The Optimus line sits where Tesla used to build the Model S and Model X. Tesla shut down those two cars earlier this year and moved the workers over. Car bodies are easy to get right. The panels are large, and small gaps of a millimeter or two are fine. Robot fingers are much less forgiving.
The report says the holders at several work stations cannot do their job well. These holders, called jigs, keep parts in place while machines join them. With small, tight-fitting parts, they cannot line things up the same way every time. Hands, joints and testing of the electronics are all named as trouble spots. So robots come off the line needing repairs before anyone can use them.
The hand and forearm together reportedly hold over 100 screws and small parts. Workers fit them by hand. The hands also wear out faster than they should. Some of the touch sensors in the fingertips do not work reliably.
Tesla’s answer is a “sensing glove” you can pull off. The glove carries the sensors. If a sensor fails, you swap the glove instead of the whole hand. Tesla plans it for next year’s model.
That is smart engineering. It is also a quiet admission. The hands break often enough that quick swaps had to be designed in.
Days of training for one simple job
The software is the bigger problem. People who know the system told The Information that each job must be taught to Optimus separately. Simple tasks take several days to learn. And the robot acts unpredictably in situations it has not seen before.
Tesla is building a library of basic movements. It hopes the robot can mix and match them to handle new jobs. The company has reportedly collected over 500,000 hours of training data. Some of it comes from paid helpers wearing camera helmets and movement-tracking suits. Tesla runs training sites in Colorado, Arizona and Florida.
In Tesla’s own factories, the robots work in small, fenced-off areas under close watch. Most of the robots being built are not working at all. They go to testing, training and data gathering. And the V3 robots rolling off the line now are not the version Tesla plans to sell. That one must first prove it can last longer and work more reliably.
You cannot buy one soon, and at first you may not be able to buy one at all
In January, at the Davos conference, Elon Musk said Tesla would sell human-shaped robots to the public by the end of 2027, when he expects “very high reliability, very high safety” and a robot you can ask to do anything. He has said before that the price would eventually be around $30,000.
The report describes a different short-term plan. Tesla would rent Optimus to a small group of business customers. Their factories and warehouses would look much like Tesla’s own. Tesla would then use data from those sites to improve the robot’s software. Ship the machines, collect data from them, and promise the software will catch up. Tesla has used this approach before with its driver-assist system.
Now compare the promises with the record. In January 2025 Musk said Tesla would build about 10,000 Optimus robots that year. He said several thousand would be doing useful work. A year later he admitted none were doing useful work at Tesla. A public reveal of the V3 robot was promised for mid-2026. It still has not happened.
Here is what is real. The jump in production is real: ten times more robots in a few months. And Tesla is not the only one trying. XPeng started an automated line for human-shaped robots in Guangzhou this month, aiming to sell to businesses in 2027.
Here is what is not real yet. No machine walks into a room it has never seen and does something useful. Until a hand can last a full shift, and a robot can learn a job in an afternoon, 20,000 a week is just a factory target. It is not a product.
