In 2026, the World Physical AI Conference hosted by Unitree Technologies took place in Hangzhou, where founder Wang Xingxing emphasized the necessity of firmly believing in AI, stating that current model capabilities are already sufficient to break through the information world, and reflected on his own insufficient conviction.

Author and source: GeekPark

On October 22, the Hangzhou International Expo Center was bustling—not only with people, but also with many robots.

This is the scene at the 2026 World Physical AI Conference hosted by Unitree Technologies, where you’ll find everything from quadruped robots to humanoid robots—you could even say this is the place with the highest robot density in the world.

In 2026, it will be the tenth anniversary of Unitree Robotics' founding. Interestingly, on today’s stage, Unitree’s founder, Wang Xingxing, spoke more about today’s hottest topic—AI—than about robots.

He not only encouraged everyone to "trust AI," but also reflected that he himself "didn't trust AI enough."

Why would the leader of this most powerful company in China—and indeed the global robotics industry—say this? Is the turning point for physical AI truly upon us?

You must trust AI

In Wang Xingxing’s 36-minute speech, Unitree’s own products occupied only the beginning and the end. The longest segment in between was devoted to one thing—convincing people to believe in AI.

From AI solving a batch of top-tier mathematical problems, piercing the pinnacle of human intellectual achievement; to AI reverse engineering eliminating the barriers of traditional software companies; to AI-generated video characters displaying expressions and emotions far superior to those of real actors; to leading large model companies internally achieving model self-evolution.

This is no joke. He concluded that, with today’s model capabilities, even without further advancement, the information world has already been “completely breached.”

Then he said something that didn’t sound like something a CEO of a public company would say: “I don’t yet feel fully convinced or steadfast.”

This is not off the cuff. Ten days ago at the Shanghai International Industrial Expo, he had already publicly admitted his mistake: “The biggest mistake I’ve made over the past decade is not having believed enough in AI.” He also offered a multiplier: “Whatever level of liking or belief you had in AI five or ten years ago, multiplying it by ten today would still be an understatement.”

Wang Xingxing discussed the future of physical AI at the event | Image source: Unitree Technologies

More tangible changes are reflected on the timeline. At the World Robot Conference in August this year, Wang Xingxing anticipated that the "ChatGPT moment" for embodied intelligence would arrive in as little as 2 to 3 years, or at most 5 to 10 years; two months later, he narrowed this window to "1 to 2 years, or 2 to 3 years at the slowest." As for simply bringing a Unitree robot into an unfamiliar home and instructing it with a single command to tidy up the room, "it could be achievable as early as next year," he said. At the China International Industrial Fair, he put it even more bluntly: "When GPT-7 launches next year, it might completely solve the embodied intelligence problem."

An engineer with a hardware background became the most fervent AI evangelist in the room within two months. This may seem like a sudden cognitive shift, but when you examine Unitree’s situation over the past six months, his belief was almost an inevitable answer.

However, Unitree’s strength—the robot hardware itself—remains a source of pride for Wang Xingxing.

Wang Xingxing's proudest achievement on stage is the G1. This humanoid robot, released in May 2024, has had a cumulative production volume of approximately 10,000 to 20,000 units, which he calls the humanoid robot with the highest single-model shipment volume worldwide.

On October 21, Unitree released a video of the H2-X “Kung Fu Legend Guardian,” demonstrating a kick that dented a car door, lifting a car with bare hands, and a 4-meter high platform backflip landing, while extending battery life from approximately 3 hours on the H2 to about 8 hours. The video shows that Unitree’s team continues to rapidly iterate on its products.

Beyond its own platform, Wang Xingxing, who has been actively promoting AI, is not exclusively promoting Unitree’s own AI.

Subsequent panel discussion featuring Wang Xingxing, Yao Ming, and other guests explored the future of physical AI | Image source: Unitree Technologies

He was candid on stage: “After all, we’re a new company”—it’s unrealistic to build exactly the same model as the world’s leading multimodal large model companies. Unitree will develop models in-house and also collaborate with major large model companies both domestically and internationally. The most impressive demo he showed was connecting GPT-6 to a robot, enabling it to autonomously clean brushes, change paints, and place stones precisely on a Go board without any training. He believes problems like planning and navigation have, in a sense, already been solved; what remains is aligning the input and output of multimodal large models with real-world robots—the errors from sensors and actuators. “I think this is also a problem that can be resolved quickly.”

Putting these statements together reveals the business logic behind this “belief.” If the robot’s brain is ultimately provided by companies like OpenAI and Google, and becomes increasingly powerful and cheaper, as he says, then the key to winning the race in embodied intelligence will return to who can most cheaply and at the largest scale produce reliable bodies—something Unitree is best at.

Among the most important things Wang Xingxing listed for the future, the top priority is not the model, but large-scale manufacturing capability for hardware. Unitree currently produces tens of thousands of robots and hundreds of thousands of joint motors per year, but he considers this “completely insufficient”—true mass adoption will require at least “tens of millions per year.” He even discussed compute clusters, chip production capacity, and mining. As for Unitree’s own model capabilities, he is betting on a “self-evolving” cycle: AI autonomously retrieves papers, writes control code, trains in simulation, tests on real robots, and feeds back performance scores. The foundation of this approach is that the base model grows stronger every month; in his words, “the more the model’s capability improves, the stronger this cycle becomes.”

World Physics AI Competition Launches | Image Source: Unitree Technologies

In other words, for a robotics company, "believing in AI" is not a superficial statement—it is a fundamental prerequisite for the entire strategy to succeed. The faster AI advances, the sooner Unitree’s hardware scale advantages will be realized.

At the end of his speech, Wang Xingxing shared a personal "intersecting timeline" with AI. In 2009, as a freshman, his first robot was a bipedal humanoid. For a computer science class assignment requiring a short paper, he chose the topic: “As neural network scale increases, capabilities become stronger,” reasoning that since the human brain is larger than those of chickens, dogs, or pigs, humans are smarter. Then came AlphaGo in 2016, and ChatGPT at the end of 2022, which directly inspired Unitree’s official entry into the humanoid robotics field in 2023.

This recollection is heartfelt and selective, turning a hardware engineer’s decades-long journey into a straight line pointing始终 toward AI. Perhaps he truly believed it early on, just not enough; or perhaps this era forces every robotics company to tell the same story.

Beyond the bustling booth

At the World Physics and AI Conference hosted by Unitree, the exhibition area was a classic scene where robots outnumbered people. In addition to traditional showcases like robot duels and table tennis matches, you could even see robots juggling diabolo and playing kendama—yes, the developer of the kendama-playing robot spoke Japanese.

Robot playing Kendama | Image source: GeekPark

Robotic dogs in various forms | Image source: GeekPark

Throughout the exhibition, we can see that, for both quadruped and humanoid robots, the largest application scenario remains guiding and entertainment. For example, you can see the G1 robot wearing a cowboy hat, as well as quadruped robot dogs dressed as pandas, dinosaurs, and lion dancers. A company called Yijing Technology has equipped its robots with facial expressions via "human-like faces." The team lead told GeekPark that in settings such as elderly care and guided tours, robots with a "face" are perceived as more approachable than ordinary robots, despite some presence of the "uncanny valley" effect.

In addition, applications such as security and patrolling have truly been implemented. A representative from a company that has modified robotic dogs for patrolling told GeekPark that quadrupedal robotic dogs equipped with LiDAR and related devices are now capable of replacing human personnel for nighttime patrols in residential communities and campuses. In emergency situations, human operators can remotely control the robots. Additionally, if fitted with fire suppression equipment, these quadrupedal robots can also perform firefighting tasks.

Staff are remotely operating | Image source: GeekPark

One of the most popular booths at the exhibition is dedicated to robotics education and curriculum. A classmate of Unitree told GeekPark that the team offers a mature solution, providing not only hardware but also a comprehensive package encompassing hardware, software, and trained personnel to equip schools with full robotics education labs, enabling more students to hands-on experience the appeal of robotics programming.

Robot Learning to Shoot Baskets | Image Source: GeekPark

On stage at the conference, Wang Xingxing and other guests launched the World Physics AI Competition, with Unitree offering tens of millions of RMB in prizes to encourage global developers to advance robotics innovation—marking one of the few major competitions in China since DJI’s “RoboMaster” to genuinely cultivate robotic developers.

Just as the recent wave of AI over the past few years has cultivated countless AI entrepreneurs and super individuals, when physical AI reaches its tipping point, this industry will similarly require a vast number of robotics talents—and the World Physical AI Competition is undoubtedly an excellent starting point.