Car companies collectively “create people”

Introduction: Car companies and car companies meet again on another track.

The automotive industry is undergoing a somewhat familiar transformation.

Previously, real estate companies transformed computing power, data centers, and AI. Now it is the turn of automobile companies.

Xiaopeng is making robots and raised more than US$900 million in financing in August; SAIC has sent humanoid robots directly into the battery production line; GAC has established an independent robot company; Changan has scheduled the mass production of robots until 2028; Chery has delivered more than 2,000 robots.

According to incomplete statistics, as of September 2026, 14 mainstream domestic car companies have clearly deployed humanoid robots, including Xiaopeng, Ideal, and NIO, to Changan, SAIC, Chery, and then to GAC, Geely, BAIC, and FAW.

Car companies and car companies met again on another track.

The question is, why do automobile companies suddenly want to “create people”?

The answer is actually not complicated, because it is indeed becoming more and more difficult to make money on cars.

According to data from the National Bureau of Statistics, from January to July 2026, the operating revenue of the automobile manufacturing industry above designated size was 6.08 trillion yuan, a year-on-year increase of 2.7%, but the total profit was only 216.24 billion yuan, a year-on-year decrease of 20.4%, and the operating income profit margin was approximately 3.6%. During the same period, profits of industrial enterprises above designated size across the country increased by 17.6%.

According to data from the Ministry of Industry and Information Technology, national automobile sales from January to July were 17.602 million units, a year-on-year decrease of 3.7%; although new energy vehicles continue to grow, the traditional automobile industry as a whole has entered stock competition.

The People’s Daily also directly pointed out this change in September this year: in the financial reports of automobile-related companies in the first half of the year, some vehicle companies reported “increased revenue but not profit.” At the same time, the profit performance of upstream links such as batteries, chips, and smart driving solutions is significantly different.

Cars are becoming more and more like a typical asset-heavy, low-profit business—sales are still there, but profits are gone.

As a result, various car companies have unanimously turned their attention to robots.

01 Robot myths among car companies

If you only look at the surface, car companies building robots is a bit like real estate companies suddenly building computing power.

One sells houses, the other builds cars, and suddenly they all talk about AI and robots. They all seem to be looking for the “second growth curve.”

But robots and cars are actually much closer than real estate and computing power.

What smart cars do, to put it bluntly, is: see the world, understand the world, make decisions, and perform actions.

The same goes for humanoid robots.

Cars use cameras, lidar, and millimeter-wave radar to sense the environment, use chips and models to determine what is ahead, and then decide to accelerate, brake, and turn.

The robot changed its body.

It uses cameras and sensors to see things, uses AI to understand the environment, and then decides to raise its hands, grab, walk, and avoid obstacles.

Therefore, in March 2026, when People’s Daily reported that car companies were getting together to enter humanoid robots, they directly summarized this relationship as follows: When car companies deploy humanoid robots, there is essentially a lot of technical synergy.

This is why car companies dare to rush.

They already have some of the most difficult things for robotics companies to make up for: supply chain, manufacturing capabilities, intelligent driving algorithms, computing power, real scenarios, and even data, such as Xiaopeng.

Unlike many companies, Xpeng does not just use car data to feed robots, but directly puts the car AI system and robots together.

The company has previously made it clear that there is a synergy between robots and intelligent driving research and development, and plans to have robots use the same physical AI base as cars. In August this year, Xpeng’s robotics business completed a first-round financing of more than US$900 million, with a post-investment valuation of more than US$6.3 billion; in September, it officially launched its robot production line.

This is no longer “car companies studying robots by the way”, but treating robots as a real new business.

But there is one thing that is easily overlooked by the market – car data is not equal to robot data.

This is the biggest advantage and biggest misunderstanding for car companies rushing towards robots.

Driving data mainly addresses road perception, driving decision-making and vehicle control.

What the robot really lacks is another type of data: I see a cup, how do I reach out? How much force should you use after your hand touches the cup? After grabbing it, how do you move it? What should I do if I encounter obstacles while moving? What should I do if something slips?

These are “operational data”.

People’s Daily quoted relevant research reports in April as pointing out that by 2025, there will be more than 140 complete humanoid robot companies in China, with shipments of 14,400 units, accounting for 84.7% of the world; however, the industry is still in the stage of moving from technology verification to industrialization.

Therefore, the car companies with real advantages are not companies with “a lot of cars”, but companies that can connect car data, robot data, models and real scenarios. Among them, SAIC is a very interesting example.

In March 2026, “Nengzai No. 1” officially entered the Buick Zhijing E7 battery mass production line, undertaking tasks such as cell grabbing and loading. SAIC said that this is the first application case of humanoid robots in the Chinese automobile industry that has truly entered the mass production line.

It is completely different for a robotics company to buy dozens of robots for training and for a car company to throw the robots directly into a car factory that is producing every day.

The former is doing testing, and the latter is doing data closed loop.

02 The bifurcation of car companies

It is the real business scenarios brought by the factory production line that complete the key leap from prototype testing to data closed loop.

But even if they have a unique landing ground, it does not mean that all car companies entering the game can successfully commercialize humanoid robots.

The supply chain, manufacturing, intelligent driving algorithms and road data accumulated on the automobile track have only laid the foundation. If you want to truly make a robot, you need to complete another set of capabilities such as body control, dexterous operation, and vertical scene adaptation. Different car companies have different cards, and their development paths have also diverged.

The first category is the “software + AI + robot” type like Xiaopeng.

Xpeng’s advantage is not its scale of automobiles, but that it has pushed itself in the direction of an AI company very early.

In the first quarter of 2026, Xiaopeng has defined itself as a physical AI company covering smart cars, autonomous driving, flying cars, humanoid robots and AI chips; the robot mass production base covers an area of ​​about 110,000 square meters, and plans to achieve mass production by the end of 2026, and enter stores and commercial scenarios in 2027.

In September, the robot production line was officially launched.

In other words, Xiaopeng has completed the first round of closed loop of “model-robot-scene-manufacturing”.

The second category includes traditional automobile giants such as SAIC, GAC, and Changan.

Their biggest advantage is their factory and industrial organization capabilities.

SAIC has already put robots into mass production lines and is making industrial investments around the robot body and core components.

GAC goes one step further. In February 2026, GAC officially established Huilun Technology, turning the robot business from an internal R&D project within the group into an independent industrialization entity; in August, Huilun Technology completed another financing of over 100 million yuan and continued to promote the large-scale commercialization of humanoid robots and core components.

Changan has clearly proposed to achieve mass production of humanoid robots in 2028 and deploy them in factories, stores and homes.

Their advantages are not so sexy, but very practical. They have factories, supply chains, workers, customers, and after-sales systems.

After robots are truly implemented on a large scale, these things will become important.

In addition to the three categories mentioned above, Chery is particularly worth mentioning separately because it is not simply betting on “universal humanoid robots”, but is commercializing robot products.

In April 2026, Chery Mojia Robots completed the order signing of 1,000 units and centralized delivery of 100 units; in July, the cumulative global delivery of Mojia Robots exceeded 2,000 units, with products and services covering more than 60 countries and regions.

These 2,000 units cannot be simply understood as 2,000 humanoid robots, because Mojia’s product matrix itself includes multiple categories such as humanoid robots, smart police robots, medical guidance robots, and robot dogs.

But this matter is still very important, because it proves one thing: robots do not have to wait until the “universal humanoid” matures to make money. It is also a way to start from vertical scenarios such as security, medical guidance, industry, and services.

This actually raises two different questions for car companies: Do they want to be a “robot company”? Or do you just want to be a “robot business”?

These two things are very different.

03 Being able to build a car does not mean being able to build a robot

Some people may ask: Can all car companies make robots?

The answer is obviously not.

One of the most common illusions in the robotics industry is that since we have built cars, robots can also build them.

This is not true.

Automobiles can rely on platformization, parts standardization and mature supply chains to spread out the complexity.

The biggest problem for humanoid robots at present is precisely the “last meter” – hands, feet, joints, dynamic balance, dexterous operation and model generalization.

This is why people in the auto industry themselves are very cautious.

The Dongfeng R&D team publicly pointed out that the brains of humanoid robots are still evolving rapidly, and the technical routes have not yet truly converged; at the same time, the industry lacks massive, high-quality real scene data, and safety, reliability, as well as the overall machine cost and work efficiency still need to be resolved.

This means that there are several types of car companies that are more dangerous.

One is that it only has concepts and no robot research and development system. Publishing a robot, shooting a video, or attending a robot conference does not mean owning a robot business.

The other one has only manufacturing capabilities but no AI capabilities. This kind of company can “pretend” the robot, but it may not be able to make the robot “learn”.

There is also one that only has car data but no operational data. Millions of cars can teach self-driving models how to drive, but they cannot automatically teach robots how to tighten screws.

Therefore, those who are likely to be eliminated in the robot industry in the future are not necessarily companies without automobile manufacturing capabilities. Instead, they may be those companies that equate “automotive supply chain advantages” with “universal robot capabilities.”

Entering the game is only the first step, the real difficulty is the next ten years.

Move the car’s AI capabilities there, move the car’s supply chain there, turn the factory into a training ground, and then sell the robots.

If any link in the middle is missing, the second growth curve will become the second R&D black hole.

Xpeng is proving whether it can become a “robot + car” company; SAIC is proving whether industrial capital, automobile factories and robotics companies can form a closed loop; GAC is proving whether robots can transform from group laboratories into independent companies; Chery is taking a more direct approach, looking for commercialization from products and scenarios first.

As for a large number of car companies such as Changan, Geely, Weilai, Ideal, and Dongfeng, the real test questions have also been put on the table.

In the end, this competition will not reward the car company that “released the robot first”, nor may it even reward the car company that “the robot is the most human-like”. It may ultimately reward another ability: who can make the robot actually go to work.

In the first few decades of the automobile industry, the competition was about engines, then batteries, and then chips, software and autonomous driving. Today, what car companies are beginning to compete for may be a new industry entrance – putting together the body of a robot, the brain of AI, and the manufacturing system of a car.

Therefore, this time, the car companies’ development of robots is really not the same as the transformation of real estate companies’ computing power.

The transformation of real estate into computing power is about leaving one’s old world; the transformation into automobiles into robots is somewhat like continuing to move forward along one’s old world.