The first batch of mass-produced Tesla Optimus robots rolls off the line; Musk says reaching 20,000 units per week will take another year

After the market closed on October 21, US Eastern Time, Tesla stated for the first time in its business update report that Optimus has begun production on the high-volume production line at the Fremont factory. The robots that came off the line are the first batch of products with mass-production quality, and the existing manufacturing processes are already suitable for commercial production.

Previously, Tesla's disclosures about Optimus stopped at R&D progress and capacity planning. This is the first confirmation that production-grade robots have come off the line. The original Model S and Model X production lines were dismantled and modified in just 46 days.

In the subsequent conference call, Musk mentioned that most of the equipment for the million-unit-class production line in Fremont is already in place, but reaching 20,000 units per week, or an annualized rate of one million units, will take at least another year, perhaps longer.

His explanation was that the product is too new. Optimus's growth is an S-curve, with a very shallow slope at the beginning and fluctuations along the way, because producing one robot requires getting roughly ten thousand things right, and the pace of progress depends on the slowest and unluckiest one of them.

In his own words, the robot cannot break down after just one month.

What remains unsolved on the hardware side is the hand. The actuation for walking and the torso has converged, but the hand is far more complex than the motion actuators. It must simultaneously meet four conditions: wear resistance, the ability to grasp fragile objects, sensor integration, and no damage from repeated operations. Custom gears are still under pressure in terms of energy consumption and reliability.

This is also why Tesla chose to start mass production before the hand design was finalized. Production lines, supply chains, and AI software can only iterate in parallel; waiting until the hand solution is finalized to start would carry a higher time cost. On the software side, the statement is that there is still a lot of work to do before it can truly handle open-ended tasks.

Training data is another hurdle.

Optimus does not have a fleet like cars to collect data, so it can only first build 10,000 to 20,000 robots for self-play, while also setting up Optimus Academy to let them practice various tasks. This is a clumsy method that he himself admits, and it is also the only method available at present.

The specific path is to first use human demonstration data for pre-training, then put the robots into simulation and real production lines for large-scale self-play, turning every round of failure into training samples for the next round. Until this process is proven, the robot's generalization ability will not improve.

The external delivery timing is set for around the end of 2027. Musk expects that issues such as AI, electromechanical design, reliability, and service life will be resolved around the end of next year, and that will be the time to deliver to companies other than Tesla. Initially, only industrial scenarios will be pursued, with high-risk, repetitive, and all-weather operations prioritized, and household use placed later.

The chip timeline was also given. AI5 is planned to be integrated into the robot in April next year, with computing power about an order of magnitude higher than AI4. To adapt to robot scenarios, AI5's memory has been reduced to 72GB, half of the original design, and AI6 will be reduced by another third to 144GB. The company says this will not affect performance.

A demonstration video will also be released within one or two months. He particularly emphasized that there should be no preset movements, and that a real machine should handle situations it has never seen. This remark is aimed at peers in the market who complete demonstrations through remote control and staged shots.

Compared with previous statements, this is a clear calibration.

At the beginning of the year, the statement was to produce 50,000 to 100,000 units in 2026. In January at Davos, he said it would be sold to the public in 2027 at a cost of $20,000 per unit. Now it has changed to delivering first to enterprise customers by the end of 2027.

The production line status has shifted from the construction phase to early ramp-up. The Fremont line is designed with an annual capacity of one million units, the second line in Texas is still under construction, and the long-term designed annual capacity is 10 million units.

The supply chain is another underestimated variable. At a meeting on the 10th of this month, Musk mentioned that China's new rare earth regulations had once set the project back by two to three months, and this remains a real constraint to this day. For the rare earth magnetic materials needed for the robots, import licenses are an unavoidable part.

Tesla's third-quarter revenue was $28.046 billion, basically flat year on year; net profit attributable to shareholders was $787 million, down 43%; operating profit was $718 million, down 56%. The full-year capital expenditure guidance was lowered from the previous more than $25 billion to about $20 billion. After the earnings release, the stock fell more than 4% after hours.

This creates a contradiction. The industrial side is pushing forward, while the financial side is contracting. The market's pricing of robots can only rely on discounting future capacity, while tolerance for current-period profits is declining.

Institutions define 2027 as the year of small-batch delivery and scenario validation. Large-scale volume ramp-up will have to wait until three things are simultaneously in place: the hand design is finalized, the supply chain is straightened out, and AI5 is installed on the robot.

The nodes to be verified next are still those few. Whether AI5 can be installed on schedule in April next year, whether the demonstration video within one or two months is truly without presets, when the custom gears for the hand will be finalized, and whether the external delivery at the end of 2027 will be delayed for the fourth time.

Back to today's A-shares. Daye Co. sealed its fifth consecutive limit-up board at 14.51 yuan, Zhaowei Electromechanical hit the limit-up at 83.39 yuan, while Fulai New Materials, which had risen in the morning session, fell back from 8.39% to 7.11%, and Dema Technology fell back from 4.10% to 1.83%. The Shanghai Composite Index fell 1.09% on the same day, with 1,961 stocks declining and only two sealing limit-up.

Musk's line that "Optimus will be the greatest product on Earth, and no one can come close to Tesla in general-purpose humanoid robots" has been said many times over the past two years.

This article is from the WeChat public account "Gelonghui APP" (ID: hkguruclub), author: Godzilla, published by 36Kr with authorization.