One hand has grasped Tesla's trillion-dollar market value.
The record - high delivery of new energy vehicles failed to drive up Tesla's stock price. One of the reasons is that there is still no certainty about the mass - production of the much - anticipated humanoid robot. However, do humans really need humanoid robots?
Tesla is trapped by "a single hand".
On October 23, Tesla released its financial report for the third quarter of 2025. Its revenue reached $28.1 billion, turning from a year - on - year decline of 11.78% in the previous quarter to a 12% increase this quarter. With a record - high quarterly delivery of 497,000 vehicles, the growth rate reached 7.39%, turning positive from negative.
After the release of the financial report, Tesla's stock price dropped by more than 4% before the opening of the market the next day. As of the close on October 23 (U.S. stock time), Tesla's stock price closed slightly up 2.28% at $448.98 per share. However, it has fallen by nearly 5% from the high at the beginning of the month.
The record - high delivery of new energy vehicles failed to drive up the stock price. One of the reasons is that there is still no certainty about the mass - production of the much - anticipated robot.
Elon Musk himself has said many times that autonomous driving and artificial intelligence are the key points for Tesla's future growth. He said on social media that about 80% of the company's value will ultimately come from the humanoid robot Optimus.
At this earnings conference, Musk said that Tesla will start building a production line for Optimus with a capacity of one million units by the end of 2026. However, what is more certain now are the risks of supply - chain shortages and difficulties in breakthroughs in dexterous hands.
Behind Musk's obsession with humanoid robots, is it that the capital needs a story, or is it a real human demand?
The Halted Optimus
Three years after another three years!
More than a month ago, Tesla announced its "Master Plan 4.0". The robot took the "absolute center stage", and Musk even said that 80% of Tesla's future value will come from robots.
Shortly after this statement, news spread that the mass - production of Optimus had been suspended.
At the earnings conference, Musk talked about the difficulties of mass - producing robots -
No supply chain: The automotive industry has a mature supply chain, and the computer industry has a mature supply chain, but the robot industry does not. To mass - produce robots, Tesla needs to build a supply chain from scratch, and the difficulty is imaginable.
In fact, this is not the first time that the mass - production of Optimus has been postponed.
Since Musk first proposed the concept of "creating a general - purpose robot that can walk and work" at AI Day in 2021, this project has undergone three major rhythm changes.
The first was in 2023. Optimus was originally planned to have its "first batch of shipments" in 2024, but due to the sub - standard basic motion control algorithm, it was postponed by one year.
The second was at the end of 2024. The mass - production target was lowered to a trial production of several thousand units in 2025.
The third was in October this year. Tesla was reported again to have suspended the mass - production of Optimus.
In addition to the "broken promise" of mass - production, there was also news of changes in the robot's senior management. Some time ago, Milan Kovac, the person in charge of Tesla's robot project, left the company. This core figure, who came from Boston Dynamics and led the Optimus system architecture, was regarded as "the person who knows the most about robot motion control" within Tesla. His departure once triggered a team restructuring. According to multiple foreign media reports, Musk then took over the project direction personally, and the R & D reporting line was transferred from the autonomous driving department to the AI chip team.
Nevertheless, at the earnings conference, Musk did not forget his usual grand - narrative style and said that the Optimus humanoid robot is regarded as having the potential to become the greatest product in history. Tesla plans to showcase the V3 prototype in the first quarter of 2026.
"We will be ready to showcase a production - intent prototype in the first quarter of next year, probably in February or March. We will start building a production line for Optimus with a capacity of one million units around the end of next year. However, it will take some time for the production to ramp up. Eventually, the production of Optimus 4 will reach 10 million units, and Optimus 5 may reach 50 million to 100 million units."
However, after weighing the long - term uncertainty and short - term risks, investors chose to leave the market in the short term. Tesla's stock price dropped by more than 4% before the market opened after the release of the financial report, and its market value shrank by about 400 billion RMB.
The significance of robots to Tesla exists both in the real world and in the story.
In the real world, Tesla needs a second growth curve.
Although Tesla's delivery volume reached a record high this quarter, the long - term stability of its sales growth is being questioned.
On the one hand, the U.S. government provides a federal tax credit of $7,500 for eligible electric vehicles, and this incentive expires on September 30, 2025. So many consumers rushed to make purchases before the end of the third quarter. On the other hand, it is not difficult to see from the quarterly growth rate curve from 2022 to now that Tesla's year - on - year growth rate of delivery volume has declined significantly.
Data source: Tesla. Charting: Baobian
In the story, Tesla's market value includes the market's expectations for robots.
Tesla's current market value is $1.4 trillion. What does this mean? Last year, Tesla sold 1.789 million vehicles, while Toyota sold 10.8 million vehicles, six times that of Tesla; BYD sold 4.27 million vehicles, 2.4 times that of Tesla. However, Toyota's total market value is $263.2 billion, less than 20% of Tesla's; BYD's market value is 946.2 billion RMB (equivalent to $121.2 billion), less than about one - tenth of Tesla's.
Obviously, this market value is not for an automaker, but for a technology company.
Tesla is right. In terms of the robot supply chain and mass - production capabilities, they are already leading the world. If robots are mass - produced and the demand explodes, it will be like a second successful start - up for Tesla. Therefore, neither Tesla nor the market will miss this "dangerous yet charming" gamble.
The Trapped Hand
There is an Optimus walking around the clock at Tesla's headquarters in Palo Alto. "He can even show you the way," Musk said at the earnings conference.
"But the key problem is the hand."
Zhang Bo, who specializes in robot system research and development at Peking University, told "Baobian" that "the small size and high degree of mobility are the core reasons for the difficulty in mass - producing dexterous hands."
Zhang Bo said, "The four wheels of a car are independent of each other. It only needs to move an object from one point to another in a two - dimensional space. However, the movement of a hand is completely three - dimensional, and the connection between each joint of the fingers is related. Therefore, building a hand is much more difficult than building a car."
Making a dexterous hand move requires several core components: motors, reducers, linkages and skeletons, sensors, and thermal management.
The motor is embedded in the palm or forearm and is responsible for providing rotational force. However, the output speed of the motor is too high and the torque is too small. It also needs a reducer to amplify the force. There are multiple sets of harmonic reducers in the arm of Tesla's Optimus, and almost every joint requires one.
After the power is transmitted to the joint, there also needs to be a skeleton and linkage system to allow the "finger bones" to transmit the force. Each human hand has 14 bones, and the robot's hand uses titanium alloy, carbon fiber, or lightweight aluminum parts to mimic these structures. The skeleton not only needs to bear the load but also needs to remain lightweight; otherwise, the arm cannot be balanced.
For these mechanical systems to really "move accurately", they cannot do without the nervous system - sensors and control circuits. There are angle sensors on each joint, force sensors on the fingertips, and some even have tactile arrays that can sense pressure distribution and sliding trends.
Data source: Public information. Charting: Baobian
Due to the extremely small physical space of the finger joints, the precision manufacturing requirements for each corresponding component are very high. Zhang Bo told "Baobian" that most of today's dexterous hands have several problems:
Insufficient gripping force, lack of flexibility, and lack of action accuracy.
The insufficient gripping force is still because of the small space. Only a small motor can be placed in the small space, and the motor volume will affect the energy density. "A robot weighing six or seven dozen kilograms can only pick up a heavy object of about ten kilograms," Zhang Bo said. "This is far from our expectations for robots."
In order to make the dexterous hand both small and powerful, engineers have tried to use "electric muscles" to replace traditional motors. For example, they use shape - memory alloys, electro - active polymers (EAP), or magnetostrictive materials to manufacture retractable actuators, allowing the "muscles" to directly drive the fingers. These "artificial muscles" can theoretically provide greater force and flexibility in a very small volume, but there is still a long way to go before commercialization.
The problem of flexibility comes from the contradiction between "degrees of freedom" and control complexity. The human hand has more than 20 degrees of freedom. If a robot wants to mimic this, it means that it needs hundreds of motors, sensors, and control circuits. Filling the small space of the palm with cables and transmission mechanisms while ensuring the lifespan and durability is a huge engineering challenge.
As for action accuracy, it is a battle between "perception and control". Robots can often "grab" an object, but it is difficult for them to "hold it steadily". Due to the delay in tactile feedback and inaccurate estimation of friction, the fingertips cannot make fine - adjustments. The current solution is to integrate AI vision and touch, allowing the hand to judge the shape of an object with a camera before grabbing it and then correct the force when the tactile sensor detects the moment of contact.
Companies like NVIDIA and Flexiv are trying to let robots optimize their actions through imitation learning in a large amount of grabbing data.
Technical bottlenecks are one problem, and high costs are another. In February this year, Morgan Stanley estimated in its report that the cost of one unit of Tesla's Optimus Gen2 is between $50,000 and $60,000. Musk has said that he hopes the overall cost of the robot will be lower than that of manufacturing a car, with the cost of one unit being about $25,000 - $30,000. That is to say, the current cost is twice the target cost.
Must It Be Humanoid?
When the company that most wants and is most likely to mass - produce humanoid robots in the world is trapped by the dexterous hand, we can't help but ask: Why must it be a humanoid robot?
Actually, non - humanoid robots are already very common in our lives.
Most of the robots that deliver water and meals in hotels are cylindrical in shape, with wheel - based navigation at the bottom and a shelf on top. The cleaning robots in shopping malls are similar to enlarged versions of vacuum cleaners, using lidar and ultrasonic waves to avoid obstacles and silently covering hundreds of square meters every day. In the logistics warehouses of JD.com and Amazon, hundreds of automated guided vehicles are running on the ground, accurately moving from the shelves to the sorting stations.
The success of these machines is precisely because they "gave up imitating humans".
They meet single - point needs.
Cleaning, handling, and delivery have single functions, definite paths, and controllable costs. Moreover, these robots already have a fairly mature commercial market. The annual global sales of vacuum cleaners exceed 40 million units, and the market for commercial cleaning robots is growing by more than 40%. In other words, whether a robot can make money never depends on how much it looks like a human, but on whether it has a clear task and cost - return ratio.
So why does Musk still want to build humanoid robots?
What he wants to replicate is not the form, but the versatility. The humanoid robot is a tool, and the humanoid robot is a platform.
From the perspective of industrial history, the evolution logic of machines has always been towards "generalization". Every iteration that can be called a "revolution" is not because of higher efficiency, but because it can adapt to more uncertain scenarios.
In a world designed for humans, the height of doorknobs, the width of shelves, and the slope of stairs are all built around the human body size. The most direct way to let robots enter these environments is to make them look like humans.
However, we must answer a second question: Is versatility really what we need from robots?
Or we can rephrase the question: What exactly do humans need from robots?
We need them to solve problems that we cannot solve, or at least achieve an efficiency that we cannot reach.
For example, in the case of deep cleaning, when humans cannot fully clean the space under the bed because of their large body size, the physical form of the vacuum cleaner solves this problem.
So versatility actually sacrifices a certain degree of specificity. Whether users need a robot that can do everything or a robot that can do the best in a special scenario is the ultimate question when discussing the demand for humanoid robots.
It is still unknown whether the grand narrative needed by the capital market and the real needs pursued by users' minds can overlap in the case of humanoid robots.
(At the request of the interviewee, Zhang Bo in this article is a pseudonym)
This article is from the WeChat official account "Baobian" (ID: baobiannews). Author: Nana, Editor: Xing Yun. Republished by 36Kr with permission.