Tesla says it has started producing its Optimus humanoid robot on a high-volume line at its Fremont, California factory, in space once used for cars. The announcement came in the Q3 2026 shareholder letter on October 21, 2026, and TechCrunch quoted it: "The bots coming off this line are the first production-quality bots that we have manufactured."
The start of production is a real milestone. The numbers around it are not yet evidence of anything: Tesla gave no output, no price and no customer list, and Elon Musk spent much of the earnings call lowering expectations. This post lays out what was said, how the claims line up with the transcript, and what would count as proof. For earlier robot context, see our coverage of Tesla's Digital Optimus Diablo demo and the Roadster event recap.
TL;DR: Optimus, claimed vs verified
| Question | Answer | Status |
|---|---|---|
| Has production started? | Early production on a Fremont high-volume line | Stated by Tesla in its shareholder letter and call |
| How many robots so far? | Not disclosed | No figure published |
| Line design capacity | One million units a year; most equipment installed | Tesla statement |
| Rate target | 20,000 units a week | Musk: at least a year, maybe longer |
| When can outside companies get one? | Around the end of 2027 | Musk expectation, conditional on engineering |
| Price? | Not disclosed | Unknown |
| AI chip | AI5 integration around April 2027 | Plan, subject to execution |
What Tesla actually announced
According to TechCrunch's report, Tesla began the work in the Fremont space previously used for electric vehicles. The same letter reported revenue of $28.04 billion, flat with a year earlier, and net income of $787 million, down 43% from $1.37 billion, with AI projects and Optimus spending cited as pressures on profit. TechCrunch also reported that the unsupervised version of Full Self-Driving, which Musk had said in April would "probably" arrive in Q4 2026, has slipped again.
On the call, as captured in a published transcript, Musk said: "we've started like what I call early production on our high-volume line in Fremont." He added that the beginning of an S-curve is shallow and "somewhat bumpy because it is such a radically new product." His framing: there are "10,000 things that have to go right," and the line moves as fast as the slowest of them.
The 20,000 per week number, put in context
Twenty thousand a week works out to about one million a year, which matches the line's design capacity. Reports differ slightly on how it was worded. TechCrunch relayed that Musk claimed Tesla will reach 20,000 Optimus robots per week by the end of 2027. The transcript wording is softer: it takes "probably at least a year, maybe a little longer to get to that annualized rate." Read together, that is a target for late 2027 or after, not a commitment, and it is a rate, not cumulative output.
For scale: the leading humanoid makers are still in pilots. Our coverage of Figure's F.02 decommissioning and Helix 2.5 home generalization shows a field measured in dozens to hundreds of units, not thousands per week.
The timeline so far: how the schedule moved
The start date has slid before, which is part of why the caution on this call matters. Summaries of Tesla's earlier quarterly materials describe the sequence this way: on the Q1 call Musk pointed to limited Optimus production starting around late July or August; the Q2 letter in July said Tesla had decommissioned the Model S and X lines at Fremont, was installing first-generation Optimus lines, and expected production "soon," with initial builds going to an internal training effort; the wording then shifted to "later this year." Those summaries come from secondary outlets, so check Tesla's own letters for exact language. What is consistent is the pattern: each milestone has arrived later and in a smaller form than the headline target.
The conversion itself is notable. Tesla chose to tear out an existing car line to make room for a robot, which signals how much of its future plan rests on autonomy and robotics rather than on the Model S and X. It also means Fremont's Optimus line started life with no spare capacity cushion: if the robots take longer, there is no old line to fall back on.
What people are asking
Is 20,000 a week realistic? Nobody has made that many humanoids. The comparison point Tesla itself offers is its Cybercab line, listed above 125,000 units a year, which is a car with 50% fewer parts than prior Tesla vehicles. A humanoid with dozens of actuators and a dexterous hand is a much harder bill of materials, which is why Musk talks about building a global supply chain first.
Will it do useful work in 2026? Not for outside companies. The first units feed Tesla's own training and testing, and external delivery is guided to the end of 2027.
Is Tesla ahead? On announced factory scale, yes. On demonstrated capability, there is no shared benchmark to say. Claims should be checked against uptime and task data when it appears.
Why Musk is managing expectations
The call was unusually cautious on Optimus. Musk said parts of the design are still changing, the robot must be "safe and reliable and has long duration" so it does not "wear out after a month," and there is "still quite a lot of work to do with regard to the AI and the robot for it to be truly useful for open-ended tasks." He expects the AI, electromechanical design and reliability to be solved around the end of next year, at which point Tesla would deliver Optimus to companies outside Tesla.
Other details from the call, per the transcript:
- Early customers: industrial, dangerous and repetitive tasks that benefit from continuous operation.
- Chip: Tesla plans to integrate its AI5 chip into Optimus around April 2027, subject to execution.
- Training data: Tesla is building a system based on human demonstrations, simulation and real-world self-play, and expects to need about 10,000 to 20,000 robots to create the data flywheel.
- Hands: the robotic hand remains the hardest hardware challenge, since it must combine durability, sensors and delicate handling.
- More capacity: additional Optimus manufacturing is being built in Austin.
- Chips at scale: Musk tied the Austin Terafab to the memory, packaging and logic bottlenecks Optimus will hit, with a test chip targeted for May or June 2027 and volume around Q3 2028.
One more note on numbers. Different summaries of the same quarter report different vehicle delivery counts, so we have deliberately not repeated any auto figures here; check Tesla's own investor relations page for the update deck.
What would count as proof
Production "started" is a statement about a line, not about robots shipped or working. Here is what independent readers can watch for:
- A unit count. Tesla has so far not published how many Optimus units exist. A number in a quarterly report would be the first hard data point.
- Outside deployments. The first named non-Tesla customer, with a described task and hours of operation.
- Reliability data. Uptime, mean time between failures and the hand's service life, the exact issues Musk listed.
- Autonomy evidence. Tesla's demo record includes video shown with human help; see our fact-check of the Diablo demo for how to read such clips.
- Price. None has been published, so any retail figure circulating online is speculation.
How this fits the humanoid race
OpenAI has also been moving toward robot hardware, as we covered in OpenAI's humanoid robots and hardware push. Unitree and Chinese makers are shipping combat and sports-demo units, as in the Unitree UniFoLM-X2 post. Tesla's differentiator is vertical integration: its own chips, its own factory, and a car business funding the effort. Its weakness, visible in this very quarter, is that spending on AI and Optimus is already weighing on profit while the unsupervised driving product slips.
Musk's claim that no one is close to Tesla on a general-purpose humanoid is exactly that, a claim. There is no shared benchmark for humanoid generality, and the field's best evidence remains company-run demos. That is why the transition from "demo" to "production line" is useful news, but it only becomes decisive when deployment data appears.
What builders and readers should take from it
For people working with robot foundation models, the practical signal is on the demand side: a company with a million-unit line wants training data, simulation and teleoperation at scale. The 10,000 to 20,000 robots Tesla says it needs for data collection is itself a useful sizing figure for anyone estimating how much real-world data embodied models need. For people tracking the market, treat 2027 as the year to judge Tesla's humanoid claims, not 2026.
Related reading
Figures are from Tesla's October 21, 2026 Q3 materials as reported by TechCrunch and a call transcript, accurate as of publication. Targets can change.