More than 20 automakers are following the trend of building robots, but industry insiders dismiss it: so-called factory integration is just making promotional videos.

Why are car companies collectively betting on the new humanoid robot track?

Text | Lin Keying

Editor | Yang Buding

Earlier this year, Jiang Bei (pseudonym) left the automotive parts company where she had worked for eight years and joined a robotics company. But she found she didn’t feel any sense of switching careers.

“All the people around me are the same ones I knew before; it’s just that now we’re all working together in the embodied intelligence industry.”

Robots and automobiles inherently share some genes. The industry generally believes that automobiles will be one of the earliest carriers to realize embodied intelligence, in a sense, can be viewed as a wheeled robot. Li Xiang, Chairman of Li Auto, once said that when Li was founded in 2015, the team believed that the ultimate form of cars is a robot.

But over the past year, robots—especially humanoid robots—and automakers have become intertwined in an unprecedentedly close manner.

At the recently concluded two sessions, proposals from Xpeng Motors Chairman He Xiaopeng and Xiaomi Chairman Lei Jun involved humanoid robots. Additionally, several automakers like Li Auto and GAC announced new related deployments at the start of the year.

According to incomplete statistics from Tencent Auto’s “Long Bright Light,” more than 20 automakers worldwide have proposed humanoid robot plans, including self-developed, cooperative, and investment approaches.

Many industry insiders believe that the long-term technological and supply chain accumulation of automakers can be replicated in robot scenarios, giving them advantages in data collection, engineering capabilities, and more. “Automakers expanding into humanoid robots is a strategic choice driven by a second growth curve, technology reuse, and internal cost reduction,” said Zhu Wei, Chairman of Longqiu Capital.

The connection is also reflected in talent flow and application implementation. Many “Jiang Bei” types are entering the embodied intelligence track, including veteran auto industry figures and senior autonomous driving executives. Stories of robots entering automotive factories are also emerging.

Several automotive industry insiders believe that boosting stock prices and attracting financing are important considerations for automakers joining the robot track. Meanwhile, doubts about when humanoid robots will truly land and whether they will divert existing resources of automakers have never subsided.

Many industry insiders told Tencent Auto’s “Long Bright Light” that it may still be too early to talk about the commercialization of humanoid robots. The so-called “factory entry” is more for publicity than actual efficiency gains, and true embodied intelligence is still a long way off.

Entering the robot track, are automakers crossing the river like Tesla?

In the narrative of capital markets, humanoid robots are not only an industry direction but also a valuation logic.

Traditional automakers are heavy asset manufacturing industries with limited growth space and profit margins; even top companies’ valuations are not very high. For example, the three largest automakers by global sales—Toyota, Volkswagen, Hyundai—have price-to-sales ratios of about 0.87, 0.16, and 0.39 respectively. Among Chinese automakers, XPeng and Li Auto have higher ratios of approximately 2.47 and 1.17 (Note: based on the latest financial reports).

Tesla’s P/S ratio is about 15.82. This is thanks to its long-standing emphasis on being an AI and robotics company rather than just an electric vehicle manufacturer.

In September 2025, Elon Musk stated on social media that 80% of Tesla’s future value would come from its Optimus robot. Before that, he claimed Optimus would transform Tesla into a $2.5 trillion valuation robotics company—equivalent to six Nvidia companies.

Some domestic automakers have begun using similar positioning statements. In January, Li Auto CEO Li Xiang announced internally the creation of a humanoid robot team, aiming to evolve into an embodied intelligence enterprise.

In February, Xpeng CEO He Xiaopeng wrote in an internal letter that Xpeng aims to be the first embodied intelligence company to fully convert technological exploration potential into scalable mass production.

Compared to automobile manufacturing, robots are obviously a more imaginative story.

According to data from the China Association of Automotive Engineers, China’s auto sales will reach 34.44 million units in 2025, up 9.4% year-on-year, setting a new record. But in 2026, due to policy adjustments and other factors, the market has shown signs of fluctuation. In January–February this year, China’s auto sales were 4.152 million units, down 8.8% year-on-year.

Industry insiders believe that fierce competition in the auto industry is pushing the supply chain to seek new growth spaces.

“Now the auto industry is very competitive, resources are particularly abundant, and the entire supply chain upstream and downstream is looking for new opportunities,” Jiang Bei told Tencent Auto’s “Long Bright Light.” “Robots and cars are both products of highly integrated hardware and software, and both have capabilities in software development and hardware supply chains that can be reused.”

For example, in software, intelligent driving and embodied intelligence both fundamentally require building foundational models first, then iteratively improving through data collection and training. “The methodology is actually similar; only the task forms have changed,” Jiang Bei said. Just as autonomous driving aims for vehicles to safely travel from point A to point B in road environments, robots might be tasked with moving goods from point A to point B.

Several industry insiders say that compared to most robotics startups, automakers are more mature in technology accumulation, engineering processes, and product deployment. A practitioner who transitioned from autonomous driving to robotics said that the R&D processes and quality systems in the automotive industry are highly industrialized, whereas the robotics industry is still relatively early-stage.

More importantly, over the past decade, the automotive industry has undergone electrification and intelligent transformation, cultivating many teams and talents with deep understanding of AI and robotics.

Tang Ming (pseudonym), who studied artificial intelligence in school, had hardly encountered robots before graduation but still received multiple job offers from robotics companies. After working at one for a year, he left to start his own humanoid robot company.

Tang Ming feels that today’s discussions about robots are very different from the high-precision, fixed-programming industrial robots of the past. The hardware and software have undergone a qualitative change, and having extensive vertical experience is no longer as critical for the robotics industry.

“Now it’s very easy for automakers to make a robot. They already have technical accumulation, and if they just gather some industry insiders and work behind closed doors, they could probably produce a highly complete robot in less than half a year,” Tang Ming said.

On the flip side, talent flow is also accelerating.

In July 2025, former Li Auto autonomous driving R&D head Jia Peng, production head Wang Jiajia, and former CTO Wang Kai co-founded ZhiJian Power, which has raised 2 billion yuan so far; in March this year, Weituo Power announced that former NIO autonomous driving senior director Qin Hailong joined as R&D vice president; recently, there were reports that former Li Auto autonomous driving VP Lang Xianpeng and ex-Alibaba VP Ren Geng are teaming up to create an embodied intelligence company.

Zhu Wei told Tencent Auto’s “Long Bright Light” that overall, teams with autonomous driving backgrounds are more favored in the current market among embodied intelligence startups, as they share commonality in perception, decision-making, control, and data feedback. Moreover, the real-time, safety-critical requirements validated by automakers’ autonomous driving teams are rare capabilities in the robotics industry.

Jiang Bei said that in recent years, increasing competitive pressure in the auto industry has led many automotive professionals to seek new directions, with robotics being one of the main targets.

Profitable? Suppliers are already ahead serving robots

Standing at the forefront of robotics, compared to the large automakers, flexible downstream suppliers have entered the wave earlier.

Take Tesla’s Tier 1 supplier Sanhua Intelligent Controls, for example. It previously mainly supplied thermal management components. In January 2024, Sanhua announced a signing with Hangzhou Qiantang New Area Management Committee for a “Sanhua Intelligent Controls Future Industry Center Project,” planning to invest 3.8 billion yuan to build a robot electromechanical actuator and domain controller R&D and manufacturing base.

The market is also enthusiastic about suppliers’ humanoid robot businesses. On October 15, 2025, news broke that Tesla ordered $685 million worth of Optimus humanoid robot linear actuators from Sanhua, causing Sanhua’s A-shares to hit the daily limit, and Hong Kong stocks closed up 12.98%.

According to Tencent Auto’s “Long Bright Light,” more than 30 automotive parts suppliers—including Topu, Junsheng Electronics, Hesai Technology—have increased their deployment in robotics-related businesses in recent years.

“The industrial synergy between automotive and robotics is very high. The same components, if made smaller by suppliers, can become usable in embodied intelligence,” Jiang Bei said. Over the past two years, many automotive supply chain resources have spilled over into embodied intelligence. Suppliers originally serving automotive needs have rapidly transformed into embodied intelligence suppliers and filed a series of patents.

Zhu Wei said that robotic parts are extensions and reuses of automotive-grade products. For example, solid-state LiDAR, which is mainly used for blind-spot detection in vehicles, can also be used in robots.

By 2025, sales of robot LiDAR from Speedtong Juchuang will exceed 303,000 units, a year-on-year increase of 1141.8%. In the fourth quarter alone, sales for robots and other applications reached 221,200 units, approaching the 238,400 units used in ADAS (Advanced Driver Assistance Systems).

One of the main drivers for supplier transformation is gross profit margin.

Financial reports show that in Q3 2025, Speedtong Juchuang’s LiDAR gross margin in robotics and other businesses was 37.2%, far higher than the 18.1% in ADAS products.

Speedtong Juchuang stated in its financial report that the company’s path is clear: vehicle-mounted business provides scale advantages and a technical foundation, robotics contributes high growth, and Robotaxi opens long-term imagination space. In the future, Speedtong Juchuang aims to become a “world-leading robotics technology platform company.”

Humanoid robots entering factories: efficiency boost or gimmick?

Over the past two years, the most visible robots in the public eye have been on TV—this Spring Festival, companies like Yushu Technology, Galaxy General, Songyan Power, and Magic Atom appeared on the CCTV Spring Festival Gala, with even more humanoid robots participating in local galas. Behind the scenes of martial arts and dance, stories of humanoid robots entering factories are also flourishing.

In the U.S., Boston Dynamics’ fully electric Atlas robot has entered Hyundai’s new Georgia plant for testing. In a public demonstration, the approximately 1.75-meter-tall Atlas completed sorting tasks on an automotive assembly line for roof racks.

Hyundai, which owns about 88% of Boston Dynamics, calls this test “the start of a great journey.” According to plans, Boston Dynamics aims to produce 30,000 humanoid robots annually by 2028.

In China, UBTECH’s humanoid robot Walker S1 has already entered training programs at FAW Audi, Geely, BYD, and other automakers. To date, UBTECH is the only humanoid robot company worldwide that has announced cooperation with multiple automakers, and the Walker S series has become the most widely used humanoid robot for training in factories.

Despite these examples, many industry insiders remain cautious about the actual progress of humanoid robots in factories.

Jiang Bei told Tencent Auto’s “Long Bright Light” that tasks in factories vary greatly in complexity. The simplest are搬运—like moving boxes from point A to point B—which mainly tests grasping and loading/unloading capabilities and is relatively easy to automate. But once fine operations are involved, the difficulty quickly rises.

“For example, Xiaomi’s recent video of aligning a screw with a screw hole and then screwing it in is already very challenging,” she said.

Most automation equipment in automotive factories still consists of traditional industrial robots—pre-programmed mechanical arms. In contrast, embodied intelligence robots need not only general perception and decision-making abilities but also extensive training for specific tasks.

“Industrial scenarios are indeed very promising, but currently, robots can only do very limited things in factories,” Jiang Bei said.

Another robotics professional also pointed out that many “landing scenarios” shown by some companies are just demo videos recorded after debugging. “If it’s just for videos, the underlying technology path isn’t really important—whether AI control, remote manual control, or writing tens of thousands of lines of code, as long as the robot performs those ten minutes of actions.”

From a technical perspective, robots still face multiple bottlenecks before truly entering production lines. For example, dexterous hands are still far from human-level flexibility, and key technologies like electronic skin are not yet mature for mass production.

Even in relatively simple assembly tasks, training costs are high. Tang Ming said that in specific scenarios, teams usually need to first determine the robot’s form—such as using a robot with fixed legs on an assembly line to improve stability—and then conduct time-consuming data collection.

“For vision-guided assembly tasks, conservative algorithms may require thousands of hours of data training,” he told Tencent Auto’s “Long Bright Light.” "Even after hundreds or thousands of hours of data collection, the success rate might only be 70–80%. Honestly, that’s not very impressive; often, it’s still easier to just have humans do it."

Another question is whether humanoid robots are truly the optimal solution for factory automation.

Many industry insiders believe that for搬运 or装配 tasks, wheeled robots or traditional mechanical arms are more efficient and cost-effective, making them a more practical choice.

“To date, humanoid robots are mostly used for entertainment,”

Despite the influx of automakers, capital, and startups, the true commercialization of humanoid robots is still widely regarded as a long-term goal.

The popular industry story for why robots should be humanoid is:

First, society has been transforming humans for many years; humanoid robots with bodies similar to humans can more easily utilize existing infrastructure.

Second, humans have a natural affinity for creatures with similar forms. When mass production of robots becomes feasible, the cost of manufacturing legs versus wheels becomes less significant. When costs are not a major concern, human preferences dominate, and people tend to prefer humanoid robots.

But in Tang Ming’s view, these stories are still far away. “At least for now, the application of humanoid robots is mainly entertainment.”

He believes that in the longer term, humanoid robots could become an important form of future society. “If we look fifty or even a hundred years ahead, humanoid robots are likely to be one of the mainstream forms.” But for startups, such a distant vision doesn’t solve immediate survival issues.

Currently, many robotics companies are still seeking commercial applications through partnerships with manufacturing firms, such as automakers, industrial automation companies, or assembly plants, for testing or deployment.

In these collaborations, both sides usually get what they need: robot companies get orders and sales volume to build financial data and stories; manufacturing firms can showcase their “smart factory” technology.

But from an industry perspective, a truly stable business model has yet to emerge.

“Right now, no robotics company has truly got the business flywheel spinning,” said an industry insider. Most companies are still in data accumulation stages, controlling robots remotely in experimental environments or data centers to continuously collect training data and improve algorithms.

This means robots have not yet entered a stage of continuous evolution driven by real-world applications. The ideal path is that future robots will perform real tasks in homes, factories, or commercial environments, continuously collecting data and improving capabilities, forming a data feedback loop similar to autonomous driving.

From a technical standpoint, creating a “highly complete” robot isn’t very difficult. The real challenge is where the application is headed.

He gave an example: small startups can survive by taking on customized orders, while top companies like Yushu focus on gaining attention through large events and selling products to entertainment or display companies. In contrast, automakers, caught in the middle, face limited application scenarios that are insufficient to support their large industry scale.

Currently, Chery’s use cases for the MoJia robot include 4S dealership guides, traffic police commands, and factory inspections. An AI quality inspector at Chery said that inspection is mainly about “patrol,” similar to patrolling, and unrelated to quality control. “Using humanoid robots for quality inspection is a bit of a waste; in the future, they might use dexterous hands for assembly or similar tasks.”

An insider close to Chery’s robotics project revealed that Chery’s MoJia robot is actually produced by other manufacturers under contract. But he also said this isn’t unusual. “Basically, every automaker is like this; they don’t have their own factories yet. You can see from Chery’s example—they have no real scenarios, just doing it for now.”

In the view of some industry insiders, this is more like a form of technological reserve. If embodied intelligence matures in the future, automakers can leverage their manufacturing capabilities to produce robots, or, as they do today with autonomous driving, bring in external “robot brains” to quickly form a complete product system.

Over the past century, human society has always had imaginations about robots. Whenever new technological breakthroughs occur, these imaginations are reignited. The robot products that truly change the world may not have appeared yet. They might be slowly iterating in some labs, or may only enter mass life in ten or even twenty years.

But before that day arrives, the industry continues to move forward.