Before Ford invented the assembly line in 1913, cars were all handcrafted. Assembling a single car took a massive amount of time, making it a veritable luxury item. After the emergence of the assembly line, assembly time was reduced to 93 minutes, and costs dropped significantly, allowing cars to truly enter the lives of ordinary people.
Today's humanoid robots are very much like cars before 1913. The technology is already stunning enough, but building one humanoid robot and stably manufacturing ten thousand of them are two completely different things.
Over the past few years, humanoid robot companies worldwide have been performing dances, somersaults, and serving tea, with various demos dazzling the audience. However, when it comes to the mass production stage, there are no precedents for production lines, the supply chain is not mature, and there is still debate over "what a successful commercial-grade humanoid robot should look like."
This is the biggest deadlock in the humanoid robot industry today: technology demonstrations are spectacular enough, but scaled manufacturing has not yet been truly figured out.
Fortunately, recent signs indicate that this deadlock is beginning to loosen.
On September 8, XPeng Motors' humanoid robot production line was officially launched. The world's first advanced general-purpose humanoid robot completed automated final assembly and autonomously rolled off the production line, with the automation rate of core processes exceeding 80%. This is the first time someone has truly put "robot manufacturing" into an automated production line.
Interestingly, the one taking the lead in this step is not a robotics company, but an automaker.
Scaled Mass Production: The Deadlock of Humanoid Robots
To understand the weight of this matter, one must first clarify a question: why is mass production so difficult?
Looking globally, the progress of various players in mass production is not particularly fast.
According to public data, Tesla's Optimus originally planned to produce 5,000 to 10,000 units in 2025, but actual deliveries were only in the hundreds. The dedicated production line at its Fremont factory will not be fully upgraded until mid-2026. Figure AI has delivered about 350 units and secured an order for 5,000 units from BMW, but the delivery schedule remains unclear.
Data from Omdia shows that throughout 2025, global sales of humanoid robots totaled only about 13,000 units. Meanwhile, the Ministry of Industry and Information Technology (MIIT) expects China's annual production to exceed 100,000 units in 2026. At the current pace, this target will be hard to achieve.
It is evident that over the past few years, the entire industry has actually been solving the problem of going from 0 to 1, meaning building it first and figuring out the rest later. A prototype can be polished regardless of cost, with parts handmade, disassembled, and reassembled if not fitted well. As long as it can eventually take a few steps or perform a few movements, it is considered a success.
Under this model, technological progress is rapid, with new breakthroughs seen every few months. However, while prototypes are everywhere, mass production remains an insurmountable challenge.
The reason lies in the fact that building a single prototype and being able to manufacture it at scale are based on two completely different sets of logic, let alone for an emerging category like humanoid robots. Producing 10,000 commercial units requires every single one to be identical and free of issues. This also involves production line design, supply chain management, quality control, cost accounting, after-sales service, and more. Every single link presents a brand-new challenge.
Moreover, most of the current so-called "mass production" is actually still dominated by manual assembly. Very few companies truly use automated production lines for large-scale manufacturing because there is no reference framework for humanoid robot mass production at this stage. In contrast, the automotive industry has been around for over a century. Who to outsource parts to, how to arrange the production line, how to inspect quality, and how to repair after sales—all these have ready-made answers.
As of now, the humanoid robot industry lacks all of these. Every step must be explored independently. Coupled with the extremely fast pace of technological iteration, a solution finalized today might be overturned next month. The supply chain and production lines simply cannot keep up with this pace of change.
This is where the deadlock lies. Everyone knows humanoid robots are the future, but no one knows how to stably manufacture them. The entire industry is stuck here, unable to move forward.
The Path to Breaking the Deadlock: Why XPeng Took the Lead
Since mass production is the deadlock for the entire industry, why did XPeng, an automaker, take the lead? The answer is actually not complicated: what the robotics industry lacks most right now is not necessarily another company capable of building robot prototypes, but a set of capabilities to stably transform complex technologies into industrial products. And this is precisely what the automotive industry excels at.
In June this year, He Xiaopeng officially announced that he would personally serve as the CEO of the robotics business, comparing this stage to the eve of the launch of XPeng's first mass-produced car, the G3, eight years ago. This move sends a very clear signal: internally at XPeng, robotics has transitioned from a cutting-edge technology project into a true mass production sprint.
Furthermore, cars and robots share deep commonalities. Driving a car and making a humanoid robot work both require AI (Artificial Intelligence) to understand the surrounding world and then execute reliable actions. Having worked on autonomous driving for so many years, XPeng knows exactly how cameras read the road, how AI makes decisions, and how data is collected and trained. These experiences can be directly transferred to robotics.
Therefore, XPeng's approach to building robots is to transplant the system that the automotive industry has already perfected onto humanoid robots.
Of course, this system is not just about software; it also includes mature manufacturing capabilities, a stable supply chain, strict quality standards, and comprehensive experience in turning complex technologies into mature products. These are precisely what the current humanoid robot industry lacks the most.
For instance, in terms of manufacturing, XPeng independently designed and developed the world's first automated production line for advanced humanoid robots, with the automation rate of core processes exceeding 80%.
Although this production line is specifically designed for the structural characteristics and assembly logic of humanoid robots, the underlying quality control methods still originate from the automotive industry. Every part undergoes reliability testing, every process has standards, and comprehensive inspections are conducted before rolling off the line, ensuring stable quality for every single product.
On the supply chain front, the overlap between the supply chains of XPeng's robotics and automotive businesses exceeds 85%. This means that a large number of suppliers already validated in the automotive supply chain can directly supply humanoid robots, eliminating the need to find suppliers from scratch and significantly reducing supply chain risks and costs.
These are advantages that pure robotics companies find difficult to possess.
Besides transplanting the mature system of the automotive industry, XPeng also made a crucial move: full-stack in-house development.
It is not that they do not want to seek cooperation, but suppliers cannot bear the costs of collaboration, so they chose to take it on themselves. Robotics technology is evolving too fast. If core components rely on external suppliers, the other party would have to go through a whole set of processes including internal project approval, investment, tooling, and testing. By the time these processes are completed, XPeng might have already completed a new round of technological iteration.
If they then tell the supplier, "Sorry, we have upgraded," all the money the supplier invested upfront would be wasted, which no one can afford.
Therefore, facing the industry challenge of "mass production," XPeng actually used three keys: lacking a reference framework, they used the mature system of the automotive industry as a reference; with an immature supply chain, they transferred the already validated automotive supply chain capabilities; and with technology iterating too fast, they kept the pace of R&D and manufacturing in their own hands through full-stack in-house development. Ultimately, XPeng's advantage is not just being better at building a single robot, but knowing better how to stably transform a complex robot into a product that can be delivered at scale.
Of course, this path also means higher difficulty.
He Xiaopeng has also stated that the technical difficulty of advanced general-purpose humanoid robots is higher than that of smart cars, with a gap of at least 20 times. Moreover, XPeng has been on this path for 8 years. Fortunately, after 8 years, driven by XPeng, humanoid robots have officially entered the mass production sprint stage.
However, mass production is not the ultimate goal; XPeng has even greater ambitions.
From One to Ten Thousand: What Truly Needs to Be Established is Industrialization Standards
A robot rolling off the production line proves that the product is viable, while a production line running smoothly proves that the system is viable.
XPeng's current goal is indeed to promote the mass production of humanoid robots, but more importantly, to build an industrial platform capable of continuously, stably, and cost-effectively producing humanoid robots. This platform will make hardware architectures reusable, software systems iterable, production lines expandable, and quality control traceable, ultimately achieving a data closed loop.
Once this underlying platform is built, subsequent new product R&D, cost control, and capacity ramp-up will become highly manageable.
Among these, the data closed loop is the most critical part of this platform. A highly anthropomorphic hardware platform can reuse human daily behavioral data and more easily adapt to various scenarios designed for humans. As IRON operates in real-world scenarios, it will collect massive amounts of data daily. This data is used to train AI, and as AI becomes smarter, it can enter more scenarios and generate more data, forming a positive cycle.
More importantly, the entire industry will benefit from this. Scaling is not just for self-amusement, but about driving the entire industrial chain forward together. This is also why XPeng's recent move is so worthy of attention.
XPeng's choice of full-stack in-house development is, to some extent, also paving the way and taking the hits for everyone. Once they thoroughly break through the most difficult core components and manufacturing processes, the experience gained from this exploration will gradually permeate the entire industry through supplier cooperation, the establishment of industry standards, or the flow of talent and technology.
When one player first runs through the industrialization path, it can often push the progress bar of the entire industry forward by a large margin. This has already been verified in the mobile phone and automotive industries, and now it is the turn of the humanoid robot industry.
This integration of the industrial chain also perfectly supports XPeng's greater ambitions.
Starting this year, XPeng has repositioned itself from an automaker to a "Physical AI Company," with automobiles, robotics, and globalization as three growth curves. The launch of the humanoid robot production line and the rollout of the first humanoid robot are crucial steps in implementing this strategy. From a set of Physical AI models to different intelligent entities, and then to scaled manufacturing, XPeng is gradually bridging the most critical missing part in the humanoid robot industry.
After becoming accustomed to countless humanoid robots showing off their technologies, XPeng's move is indeed more worthy of attention than just another robot demo.
Conclusion
Transplanting the manufacturing, supply chain, and quality systems accumulated by the automotive industry over decades, using full-stack in-house development to solve the problem of overly fast technological iteration, and building an automated production line from scratch.
This is very much like what Ford did over a century ago: not only making cars better but also completely changing the way cars are made. This is also what XPeng is doing now—redefining the way robots are built.
As for what comes next, it is hard to say and uncertain right now. There are still a bunch of problems to be solved in the robotics industry, which can only be left to time to prove. But with mass production achieved and the platform built, the entire industry can hit 2x speed.
If looking back from the future, September 8, 2026, might just be the starting point where the deadlock of robot mass production begins to loosen.
#XPeng #HumanoidRobots