Why China Is Winning the Robot Race (And Nobody's Talking About It)

Amazing Reality

Humanoid robots are taking over—well, almost! Dive into the wild race between China and the US as we explore factories pumping out robots that look and move just like us. Can China really claim the crown? You’ll want to see these crazy advancements for yourself! Subscribe for more future-shocking content and drop a comment telling us what blew your mind most! #robots #technology #AI #china #innovation  👉 This channel was created in collaboration with https://www.youtube.com/@realidadeimpressionante   🔊🎧Music Info: Provided by the channel "MokkaMusic" and https://inaudio.org  0:00 - Introduction to China's Humanoid Robot Surge 0:52 - XPeng and China's Industrial Robotics Edge 3:03 - Manufacturing Scale and Strategic Advantage 5:07 - US Competition and XPeng’s Expansion 6:48 - XPeng’s Journey and Humanoid Development Philosophy 10:57 - Industry Growth, Training, and Applications 16:34 - Biomimetic Design and Humanlike Features 22:05 - Supply Chain and The Road Ahead

Transcript

00:00A humanoid robot just walked out on its own right after being built and headed to meet its creator.

00:07But does this footage confirm a statement that seemed exaggerated not long ago?

00:12Did China win the race for humanoid robots? This question arises because it's already

00:17producing these machines at an industrial scale that the United States is still trying to achieve.

00:23And now some Chinese companies are moving in another direction. Robots with faces,

00:29expressions and movements increasingly resembling those of human beings.

00:34But a country starts manufacturing thousands of these robots and making them more and more human

00:40with the promise of sending them all over the world. That's what we're going to unpack now.

00:51As seen in the intro, a humanoid robot has just walked off the production line on its own where

00:55it was manufactured over in China. That one right there is iron from XPeng and it did that with no

01:01human assistance or teleoperation whatsoever. And this striking image could mark a very important

01:07shift in the race between China and the United States for control of humanoid robots.

01:14Because while US companies like Tesla and FIGURE, for example, develop some of the most advanced

01:19machines in the sector, Chinese company XPeng has just taken a major step, turning humanoid robots

01:25into products that can be manufactured at scale. The company officially put into operation in Guangzhou,

01:32a line fully dedicated to the production of the iron, with over 80 of the main processes being automated

01:38in that factory. And it's precisely at this point that China begins to open up a bit of an advantage.

01:44And XPeng is doing this in isolation, okay? Data from the research firm Smart Analytics Global shows

01:51that in the first half of 2026, nearly 20,000 humanoid robots were shipped worldwide and more

01:57than 97 of them came from Chinese manufacturers. And that means China's advantage in manufacturing

02:02these humanoid robots has already begun to show up, not just in demonstrations or new factories.

02:08It shows up mainly in the numbers, and other estimates point in that same direction too.

02:13You know, Grand Force, for example, projects that Chinese production of humanoid robots will grow

02:1894% as early as 2026. And the same consultancy estimates that the Chinese humanoid market will

02:25exceed $2 billion this year and could grow by at least another 60% as early as next year in

02:302027.

02:32China, therefore, doesn't just have several companies developing humanoid robots,

02:37it's starting to build an industrial supply chain that is capable of producing them in large quantities.

02:43The race for humanoids is no longer just about artificial intelligence, impressive movements,

02:47or demonstrations in laboratories or even in public places or factories.

02:52It also depends on the ability to manufacture these machines. Repeatedly, control costs, maintain quality,

02:58and, of course, build a supply chain capable of supporting thousands of these robots.

03:02And you should know that China already has a good part of this infrastructure, you know?

03:06In Xpeng's case, the advantage starts right within the automotive industry itself.

03:11Before manufacturing humanoids, the company spent several years developing and producing their electric

03:15cars, even so-called flying cars, alright? And now it is transferring part of this structure directly.

03:21To the robots, according to their humanoid lead, approximately 85% of Iron's hardware supply chain

03:26overlaps with Xpeng's electric vehicle supply chain. And that includes motors, metal structures,

03:32electrical cables, batteries, and different integration processes. And the company is also

03:38using quality control systems and manufacturing techniques in the humanoids, originally developed

03:42for their automobiles. Even Iron's hands, one of the robot's most complex mechanical parts,

03:48have dedicated automation equipment during their assembly. The result is a factory designed not only to

03:53build a functional humanoid, but to repeat that process thousands of times. And the robot that's

03:59beginning to come off this line shows the level of complexity involved. Inside it are three Turing

04:04chips developed by Xpeng itself. Together, they provide up to 2250 tops of computing power to run the

04:11company's own physical artificial intelligence systems. That's directly inside the robot. According to

04:16Xpeng, the goal is to enable Iron to perform complex tasks autonomously, meaning without anyone controlling it,

04:22and keep improving day by day through its interactions with our real world.

04:26The robot also features autonomous navigation, dynamic obstacle avoidance, and the ability to

04:31locate its charging station. But the first units won't be sent directly to people's homes or factories to

04:36work. Xpeng is starting with much more controlled environments. The first irons to be produced will be

04:43used in the company's own stores and corporate facilities performing activities like interacting with

04:48customers, guiding visitors, and even patrolling. These environments will also allow them to collect

04:54data before a much larger commercial expansion. Deliveries to business customers are scheduled for

04:592027, initially in sectors such as retail, hospitality, and customer service, both in China and international

05:06markets. Okay, and this is where the comparison with the United States becomes really important.

05:11Tesla also intends to manufacture Optimus on a large scale. Elon Musk has stated several times that his

05:17humanoid could become one of the company's most important products. And Tesla has many of the

05:22same industrial advantages that Xpeng has. Automotive factories, batteries, electronics, artificial

05:27intelligence, and a global supply chain. American-based FIGURE is also advancing manufacturing in the United

05:33States and is already producing FIGURE-03 units at its facility in Bot-Q. So American companies aren't

05:40standing still, okay? But Xpeng has just put its current generation of humanoid robots onto a dedicated line,

05:46with a high level of automation, while Tesla is still working to expand its manufacturing capacity

05:51for the next generation of the Optimus robot. And this difference is happening within a much bigger picture,

05:57you know? China already has a broad network of companies producing motors, reducers, sensors,

06:02batteries, actuators, and other components needed for humanoid robots. And this allows Chinese manufacturers to

06:09develop robots using a domestic supply chain that already includes sectors such as electric vehicles,

06:13electronics, and industrial automation. And Xpeng is also rapidly increasing the resources available

06:20for Automobile R and D. Its robotics division recently raised over $900 million in its first funding round,

06:26led by IDG Capital, and also with the participation of investors such as Alibaba and Tencent. There are

06:33giants over there. The unit was valued at over $6.3 billion. The funds will be used to develop hardware

06:40and

06:40software as well as to train artificial intelligence models and expand the infrastructure needed to

06:45produce more and more humanoid robots. Does this mean China has already won the robot race? Well, we are

06:52going to analyze a few more facts and you will be able to decide for yourself and answer that question.

06:58And here's a little bit more of iron but from another perspective, okay? Shortly after putting its new

07:03humanoid robot production line into operation and showing the first iron coming off it and walking

07:08autonomously, Xpeng decided to show another part of this story. It's not the factory but rather the

07:15six years of mistakes, falls, and engineering decisions that brought this robot to this moment. In a video

07:20lasting about six minutes, CEO He Xiaopeng presented a very candid look back at the development of the iron.

07:28Instead of focusing the presentation solely on current results, the good things that happened,

07:33Yao Pang recalled some of the moments when the project was met with skepticism and even ridiculed by the

07:38public. And when I say the public, it wasn't just international, even the folks back in China. You know,

07:44when the first versions of iron appeared walking with movements very similar to a person's accusations

07:49surfaced that there could be a human being hidden inside that machine just like Elon Musk had done a

07:54while back with Tesla. Xpeng's reaction to this was unusual. During a live presentation, the company

08:01literally opened up part of the robot's leg right in front of the whole audience to show the mechanical

08:05components that were beneath the outer structure. Even so, some kept saying it was a human being

08:11with a prosthesis. Another episode happened when a version of iron fell during a demonstration in a

08:16mall. The images quickly spread across Chinese social media and the robot became a laughingstock.

08:22According to He Xiaopeng, developing robots means learning and living with this kind of failure all the

08:28time. And these experiences also helped shape a company decision. Xpeng doesn't intend to focus its

08:34efforts on demonstrations developed primarily to produce very impressive videos. Xiaopeng acknowledges

08:39the advances of humanoids capable of running, competing, or even performing choreographed

08:44presentations. And he states that this isn't the company's main goal. The goal is to develop a general

08:49purpose humanoid robot capable of working autonomously. And it's in this context that he uses a phrase to

08:56explain Xpeng's strategy, the hardest path is never crowded. And part of that difficulty lies in the

09:02intelligence that controls the robot. Xpeng is leveraging technologies and infrastructure

09:07originally developed for its automotive systems. An autonomous vehicle needs to observe a constantly

09:12changing environment, identify objects, predict risks, and make decisions in real time. A humanoid robot

09:18faces a very similar problem, but it needs to turn those decisions into full-body movements. That's why

09:24Xpeng is directing iron toward models that combine vision, language, and action. Instead of relying solely on

09:31pre-programmed sequences or a human operator controlling the robot remotely, the company aims

09:36to enable their system to interpret the environment and determine its own actions and decisions.

09:41A video shows small examples of this approach. In one of them, the system receives an image of an

09:46environment and is asked about the presence of a green couch there. The object simply doesn't exist in

09:51the image, and iron correctly identifies its absence. A simple test meant to check if the system can

09:57interpret what it is actually observing instead of just responding based on expected associations or

10:04orders from humans with, shall we say, ill-conceived intentions. In another demonstration, the robot

10:10interacts with a person and ends the conversation with a handshake. Xpeng also aims to develop persistent

10:17memory, allowing the robot's system to recognize different people and, of course, even better, retain

10:23information from previous interactions. The idea is to combine visual perception, language, and physical

10:29control without relying on joysticks or behind-the-scenes teleoperation, as often happens in some

10:34presentations, you know? But Xiaopeng also acknowledges that putting iron into production doesn't mean

10:39development is over, okay? According to him, the first commercial version might not be perfect, but his

10:46robots, the ones coming out of these factories, would be able to do, right here in the real world, what

10:51they

10:51were designed to do, work autonomously, and take away a lot of jobs.

10:57And Chinese manufacturers already account for more than 97% of humanoid robots shipped worldwide.

11:03This is a widely reported fact by several global sources and not just China, okay?

11:08This estimate is from research form Smart Analytics Global and covers the first six months of 2026.

11:14According to the survey, approximately 19,100 humanoids were shipped globally during the period,

11:19compared to about 5,100 in the first half of 2025. According to this survey, approximately

11:25anti-maulid 1,100 units of humanoid robots were shipped globally during that period. In contrast,

11:31we only have 5,100 units in the first half of 2025. And that's an absurd growth of 272%

11:38in just a year.

11:39And the top five companies appearing in the ranking are all Chinese.

11:45Hibot took the lead with approximately 8,400 units and 44% right behind his unit 3 with about 5

11:52,900 units

11:53and 31% of the market. Combined, the two companies account for roughly three quarters of all humanoid

12:00robots shipped worldwide. But the Chinese industry already has a second tier of manufacturers ramping

12:06up this production, you see? Golbot, for example, appears in third place with approximately 900 units.

12:13Meanwhile, Obitec, which was accused of using CGI in its robots, holds fourth place with about 700 units

12:19shipped. And the least famous one, Leju, rounds out the top five with approximately 600 humanoids shipped.

12:26Another survey, this one conducted by Counterpoint Research, uses a different methodology and estimates

12:32over 22,000 humanoid robots shipped globally during that same period, you know?

12:36The individual numbers are also different, but the ranking result is exactly the same.

12:41Hibot, Unitree, Galbot, Obitec and Leju occupy those top five positions.

12:47Production, however, is just one part of the problem. A Reuters investigation showed that China

12:53is still struggling to turn humanoid robots capable of impressive demonstrations into machines that are fast

13:00and reliable enough to work in factories. And to solve this, the country is creating large centers

13:05dedicated specifically to training these machines. For example, in a facility visited by Reuters itself

13:12in southern China, more than 100 humanoid robots stand arranged in rows, while human trainers use

13:17headsets and controllers to teach them tasks. The operators see the environment from the robot's

13:23perspective and demonstrate to them the movements that are later used as training data.

13:28The humanoids take part in activities like sorting boxes, packing food, and making coffee.

13:34It's practically like a school for robots. Ubitec won an approximately $18 million contract

13:40to supply humanoid robots and other equipment for that center. But this process still shows that the

13:46technology needs to evolve a lot. According to Chinese employees interviewed, for example, a novice

13:52trainer might need to make approximately 300 attempts to produce a single usable movement.

13:58An experienced operator can reduce that number to around one useful attempt every 50 times.

14:04In certain simpler tasks, the humanoids observed by the report worked at only 20% of the speed or

14:10productivity of a human being. But despite these limitations, some commercial applications are already

14:16starting to emerge. Gelbot, for instance, already uses its robots in pharmacies across more than two

14:21dozen Chinese cities. When it receives a digital order, the robot navigates the store, identifies the

14:27medicine, and takes the product off the shelf. The system uses cameras, including one mounted on the

14:32wrist, along with a gripper and a suction cup to handle different packaging. According to Gelbot,

14:37each product can be located and prepared in about one minute, with a success rate exceeding 95%.

14:45Unitree's strategy is slightly different. Its UG-1 robot, a smaller and more affordable humanoid,

14:51has already reached approximately 11,000 units produced in total by mid-2026. It's true that a good

14:58portion of these robots continues to be used in research, education, and development, as well as

15:02presentations. And Ledger Robotics is also part of this ecosystem, all right? With its Quavo series,

15:08used even in centers dedicated to data generation and humanoid training. And there is right now,

15:13at this exact moment, a parallel race to improve the artificial intelligence that controls these

15:18machines. As you got introduced this month, the GE ACT 2.0, a new model developed specifically for

15:25embodied artificial intelligence, meaning artificial intelligence that is installed inside the robot.

15:30The company trained versions of its systems using datasets ranging from 300 to 30,000 hours. And as the

15:36quantity and quality of data increased, the model started performing tasks that the smaller versions

15:41couldn't complete, including folding towels, stacking paper cups, opening, and arranging flowers.

15:47In the tests, the same model was evaluated without additional specific training, in environments,

15:54objects, and lighting conditions that were not present in the previously used data.

15:59The evaluation involved 100 tasks, 20 skill categories, and two different types of robots.

16:05The results were released by Agabot itself, and still do not mean that the generalization problem

16:10is truly solved. However, they clearly show where part of the Chinese industry is focusing its efforts.

16:17The hardware is already being produced in thousands of units. Now, companies like GBOT,

16:22Agibot, Unitree, Galbot, Ubitech, and Leiju itself need to turn this industrial capability into robots capable of working with speed,

16:30autonomy, and reliability here in our real world. Especially reliability, right?

16:35China isn't just developing humanoid robots capable of working. Some companies there are trying to make these machines

16:42also look more and more like us, human beings. And yes, I am a human being, okay? I'm not an

16:49AI creation,

16:50in other words, flesh and blood. If you pinch me, it'll hurt, alright? And creating robots that are more like

16:55us,

16:55involves much more than just putting a silicone face on a mechanical frame.

17:00Chinese companies are separately developing biomimetic skin, micro-expressions, moving eyes, emotion recognition,

17:06more realistic hands, and even new ways to make robots walk exactly like a person.

17:11One of the most advanced experiments comes from Aheadform, a Shanghai company. A company specializing in biomimetic robots for human

17:19interaction.

17:19Its flagship model is the Origin F1 which was specifically developed to reproduce facial movements.

17:25The system uses dozens of actuators to control regions like the eyes, eyebrows, mouth, and other parts of the face.

17:32The structure is covered by a flexible artificial silicone skin that deforms as these mechanisms move.

17:38The goal is to reproduce details that don't usually exist on a robotic face. Blinking, moving the eyes,

17:44changing the position of the eyebrows, and even blowing a kiss if needed, you see?

17:49Synchronizing the lips with speech is something they do too, and producing slight changes in expression

17:54while the robot is interacting with a person is second nature for these guys.

17:58During WRC Walk 2026 in Shanghai Aheadform presented versions of its systems capable of controlling

18:04hundreds of points related to facial micro-expressions. Artificial intelligence is also part of this

18:10process, alright? Cameras installed around the eyes allow it to identify and track the person speaking to

18:16the robot. Microphones pick up the conversation. The system then brings all of this together, combines the

18:22visual information, voice, and context to coordinate the gaze, speech, and facial movements.

18:27But another Chinese company is trying to take this kind of technology to a slightly different

18:33stage, you know? Huan Intelligent is commercially producing biomimetic robotic heads. According to

18:39information released by the company itself, nearly 400 units had already been shipped by September of

18:44last year. Meanwhile, backlogged orders surpassed a thousand units. Wuhan also claims that its new

18:51emotion recognition system achieves an accuracy of over 95%. This result, however, was released

18:57by the manufacturer itself and is not the result of an independent evaluation. The breakthrough is

19:01significant, mainly because it shows that realistic robotic heads are starting to move beyond being

19:06just individually built prototypes. And there is also a market emerging for robots developed

19:12specifically for social interaction. Over in China, Ubitec introduced a line of complete companion robots

19:18powered by artificial intelligence and featuring a nearly human appearance. The models were developed in

19:23male and female versions, and these use emotional recognition and conversation systems to interact

19:28with their users. That's bizarre. The most sophisticated versions reach approximately

19:34990,000 yuan, around 145,000 dollars. And the concept is different from the humanoids used in factories.

19:42Instead of transporting materials or working on assembly lines, these systems are mainly intended for

19:47interaction and companionship, whatever that might entail, right? Another Shanghai company, Drawdryup, is

19:54developing Moya. This is the robot that generated a lot of engagement here on the channel. It also uses

20:00biomimetic skin, facial movements, and eye contact, but it also tries to reproduce very human characteristics

20:06across the rest of its body, if you know what I mean. The manufacturer claims that its gait has a

20:1192%

20:12similarity to human walking patterns. And that number is also a claim made by the company itself,

20:17not a result obtained through independent evaluation. Just to make that crystal clear here, okay?

20:23An appearance isn't the only way to make a humanoid robot's movements more natural, so to speak.

20:28Engine AI, for instance, developed the SE-01 specifically focusing on an anthropomorphic gait.

20:34The robot has the approximate dimensions of an average height person and uses a structure designed to

20:38produce leg, hip, and torso movements that are much closer to the human pattern throughout an entire

20:43walk. This shows that different problems are being tackled simultaneously. E-headform works mainly with

20:49the face expression and interaction. Juan is trying to manufacture entire biomimetic heads in large

20:56quantities. Optech explores robots aimed at companionship. Droid Dryjump combines appearance and

21:02movement. And Engine AI works to make full body locomotion much more natural.

21:07And these technologies aren't yet brought together into a single machine capable of perfectly

21:12replicating human appearance, movement, and behavior. And the numbers released by the manufacturers

21:18themselves still need to be evaluated independently. But the development of these platforms shows that

21:24the Chinese humanoid robot industry is really also moving in a direction that differs from industrial

21:29production. Besides building robots in large quantities, Chinese companies are developing the

21:35technologies needed to change the way these machines look, walk, and interact with us humans.

21:42For now, these features and capabilities are spread out across different companies and

21:47different robots. But pretty much all the components are being developed at the same time over there in

21:52China, right? Beamimetic faces, synthetic skin, social expressions, AI-driven interaction, and bodies whose

21:59movements are getting closer and closer to our own human movements.

22:04And now let's answer why China is able to manufacture so many humanoid robots.

22:09So far, we've seen that they can already manufacture thousands of these units, right?

22:14But there's an industrial reason behind this scale. Even before a major humanoid robot industry existed in

22:20the country, a large part of the supply chain needed to build them was already being set up behind the

22:24scenes. A robot of this type relies on motors, servomotors, precision reducers, joints, sensors,

22:31cameras, batteries, chips, structural materials, and electronic systems, in addition to artificial

22:37intelligence. And many of these components are already largely produced, all of them within China

22:41itself, mainly for electric vehicles, electronics, drones, industrial machinery, and automation systems.

22:48According to TrendForce, components like servomotors, reducers, and lithium batteries can be

22:53quickly supplied by Chinese manufacturers, cutting costs, and reducing the time needed to go from

22:58development to production at scale. The electric vehicle industry played a very important role in

23:03this process. During its expansion, China developed factories and suppliers specialized in batteries,

23:09electric motors, power electronics, sensors, materials, and control systems. Part of this

23:14infrastructure can now be used by the humanoid robot industry. Therefore, when a company develops a

23:20new generation of robots, it doesn't necessarily have to develop the entire supply chain at the same

23:25time. They can simply turn to a network of suppliers that already produces similar components on a large

23:31scale. And this ecosystem also allows production to scale up quickly. According to TrendForce, Agrobot took

23:37approximately three months to go from the milestone of 1,000 robots to 5,000 units. Robots!

23:45And Unitree has also already built up the industrial capacity to produce up to 75,000 humanoid robots

23:52every year. This doesn't necessarily mean the company will produce or sell all that quantity,

23:58but it shows the scale of capacity that Chinese manufacturers are already beginning to set up.

24:04But that's not all. There is an even more important issue to determine the future of this industry.

24:09It's the financial return, of course. That is why one of the next steps will be China.

24:17It has already shown and proven, according to international bodies, that it has an industrial

24:22supply chain capable of producing humanoids in large quantities and scaling up that production really

24:27fast. The next test will be different, right? These robots will need to work for much longer,

24:34make fewer mistakes, and perform tasks with enough productivity to justify their costs.

24:38In other words, replace humans with robots. Because manufacturing thousands of humanoid robots

24:44is only one part of that equation. To call it and say, China won the race for humanoid robots. They

24:50need to prove they can turn that scale into real work. And that work has to be turned into money

24:56for

24:56these guys. If you liked this episode, go ahead and hit that like button and make sure you're subscribed.

25:01There are two new recommendations on your screen, one right here and another one over here. Just pick

25:05one of them and click and I'll catch you in the next one, alright? Thanks, see you in just a

25:10bit. Bye-bye.

Category

Transcript

Comments

Recommended