Wang Xingxing emphasizes belief in AI at Unitree's 10th anniversary

A tenth anniversary framed as an AI test

Unitree’s 10th anniversary moment was not presented as a nostalgic celebration of robot dogs, humanoid demonstrations or stock-market milestones. It was framed as a shift in conviction. At the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference in Hangzhou, Wang Xingxing, Unitree’s founder, chairman, general manager and chief technology officer, used the stage to argue that the decisive challenge for embodied intelligence is now the alignment between AI models and physical robots [1].

That message fits the subject of Unitree’s anniversary: Wang is not simply saying that AI is useful for robotics. He is saying that belief in AI has become a strategic requirement for robotics development. Recent Chinese coverage has emphasized his own self-critique: Wang said his biggest mistake over the past decade-plus was not believing enough in AI, after Unitree tested GPT-6 on multiple robot tasks and saw encouraging results [5]. In other words, the anniversary story is not only about Unitree turning ten. It is about the company’s founder revising the technology timetable for the entire robotics sector.

What Wang said in Hangzhou

Wang’s latest Hangzhou remarks sharpened the idea that “physical AI” is moving from concept to implementation. He said the core challenge in crossing embodied intelligence’s “ChatGPT moment” is aligning the inputs and outputs of AI models with real robots [1]. That formulation matters because it identifies the bottleneck between digital intelligence and physical usefulness: a model may reason, write code or understand instructions, but a robot must still perceive, move, manipulate, recover from error and complete tasks in messy environments.

Wang also said that leading large-model companies have, in some sense, already achieved a form of model self-evolution: models can write programs and collect data, while humans increasingly guide and intervene rather than manually perform every step [1]. The point is not that people disappear from the loop. It is that the loop changes. For robotics companies, the target becomes a faster cycle in which AI helps design, simulate, test and refine robot behavior.

Wang added a caution that keeps the anniversary message from becoming pure hype. Breakthrough technologies often take five to ten years to diffuse into ordinary use, he said, which means the public may not immediately feel the impact even if the technical trajectory is improving [1]. That distinction is central to understanding Unitree’s position: confidence in AI does not mean overnight household robots. It means Unitree believes the slope of progress is steepening.

From “not believing enough” to an earlier timetable

The phrase that has followed Wang through recent coverage is unusually candid for a hardware founder: he described under-belief in AI as his biggest mistake over the past decade-plus [5]. That is a revealing admission because Unitree built its reputation on hardware execution: affordable quadrupeds, increasingly agile humanoids, and a visible presence in research, education, entertainment and developer communities. Yet Wang’s anniversary message suggests that hardware maturity is no longer enough.

According to recent reporting, Wang’s view of the embodied-intelligence inflection point has moved forward. Earlier in 2026 he had described a possible two-to-three-year path in optimistic conditions, and a five-to-ten-year path in slower conditions; by October, after GPT-6 robot tests, he was discussing the possibility that the threshold could arrive in one to two years in a faster scenario [5]. The important point is not the exact prediction. Robotics timelines are notoriously difficult. The important point is that Wang is publicly recalibrating toward AI acceleration.

That recalibration changes what Unitree’s 10th anniversary symbolizes. The company’s first decade was about proving that legged robots could be built, shipped and priced in ways that widened access. The next decade, in Wang’s framing, is about whether AI can convert those machines into adaptive workers, tools and platforms.

The G1 as a platform proof point

Unitree’s current message is stronger because it is tied to production claims, not only research ambition. On October 22, KuCoin’s news feed, citing Chinese reporting, said Wang disclosed at the Hangzhou conference that cumulative production of the Unitree G1 humanoid robot had reached roughly 10,000 to 20,000 units since its May 2024 launch, describing it as the world’s best-selling humanoid robot [2]. The same report noted that the G1 starts at 99,000 yuan, includes 23 to 43 joint motors depending on version, uses 3D LiDAR and a depth camera, and offers about two hours of runtime [2].

Those figures should be read carefully. A production range is not the same as independently verified active deployment, customer retention or paid task completion. Humanoids Daily likewise treated the 10,000-to-20,000 figure as an attributed manufacturing claim rather than a verified sales or operating-fleet count [3]. Still, the claim is strategically significant. If a humanoid platform is being produced at that scale, even with uncertainty around exact use cases, it becomes more than a prototype. It becomes infrastructure for developers, integrators, universities, startups and application teams.

That is already visible outside China. Humanoids Daily reported that several U.S. robot startups are building products or prototypes around the Unitree G1 body, using it as a hardware base while differentiating through software, sensing, character design or application focus [3]. The pattern reinforces Wang’s point: the body matters, but the decisive layer is increasingly the intelligence and application stack built on top.

An ecosystem anniversary, not a single-company celebration

The structure of the Hangzhou event also supports the idea that Unitree is trying to move from product company to ecosystem organizer. The 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference was described as a large gathering with more than 3,000 expected guests, a main forum and a partner ecosystem focus under a ten-year theme [4]. KuCoin’s report described more than 3,000 guests, a main forum area of more than 4,000 square meters and over 40 exhibition zones [2].

That scale matters because Wang’s belief-in-AI message is not aimed only at Unitree’s internal engineers. It is aimed at suppliers, developers, researchers, commercial partners and investors. A humanoid robot industry cannot scale through a single company doing everything alone. It requires actuators, batteries, sensors, simulation, data pipelines, safety testing, model integration, maintenance networks and domain-specific software.

Unitree also used the conference to launch the World Physical AI Competition, with a prize pool in the tens of millions of yuan and tracks including locomotion, track and field, combat, dexterous manipulation, production, sports events, multimodal interaction, and health and elderly care [1]. The competition format turns Wang’s argument into an operating model: invite outside teams to stress-test capabilities and create benchmarks around physical tasks.

Why “believe in AI” is a strategic statement

In robotics, “belief” can sound vague. In Wang’s usage, it is more operational. Believing in AI means reorganizing development around AI-native loops. It means assuming that model capabilities will keep improving quickly enough that hardware road maps, data systems and developer tools must be built to absorb that progress. It also means accepting that the robotics company that waits for perfect proof may miss the learning curve.

This is why Wang’s remarks are strategically important. If the hardest problem is aligning AI model outputs with physical robot behavior, then the winners may be the companies with the most useful real-robot data, the fastest sim-to-real testing loops and the broadest developer ecosystems. Unitree’s installed base, if the G1 production range is directionally accurate, gives it a potential advantage: every shipped platform can become a site for experimentation, feedback and application development [2][3].

But the risks are just as clear. Physical AI must deal with safety, reliability, maintenance, cost, liability and user trust. A chatbot failure is usually reversible; a robot failure can break objects, injure people or halt operations. Wang’s own warning that broad diffusion may take five to ten years is therefore not a footnote. It is a reminder that progress in model capability must be matched by progress in deployment discipline [1].

The anniversary takeaway

The most important line from Unitree’s 10th anniversary is not that robots are becoming more impressive. It is that Wang Xingxing now sees AI as the force that can compress the timetable for embodied intelligence. His self-critique — that he did not believe enough in AI — turns a founder’s reflection into a corporate doctrine [5].

For Unitree, that doctrine now appears to have three parts: scale the robot bodies, build and attract the AI layer, and organize an ecosystem that can turn demonstrations into repeatable applications. The Hangzhou conference, the G1 production claim, the partner ecosystem push and the World Physical AI Competition all point in the same direction [1][2][4]. At ten years old, Unitree is presenting itself not merely as a robot maker, but as a platform for physical AI.

The conclusion is optimistic but not naïve. Wang is urging confidence in AI because he believes the robotics bottleneck is narrowing. He is also warning that the path from breakthrough to mass use remains long. That balance — conviction without pretending that commercialization is solved — is what gives the anniversary message its importance.

Sources from the last 72 hours

1. [1]Unitree G1 Humanoid Robot production reaches 10,000 to 20,000 units.Oct 22, 2026, 7:35 AM
2. [2]The Unitree G1 Is Becoming a Platform for American Robot StartupsOct 22, 2026, 2:00 AM
3. [3]Daily Report | The Frontier of Embodied AIOct 20, 2026, 2:00 AM
4. [4]说好"苦活"最后干，DeepSeek却抢着入场：梁文锋的具身转向_搜狐网Oct 20, 2026, 11:14 AM
5. [5]宇树科技王兴兴：模型与真实机器人对齐是核心挑战Oct 22, 2026, 5:26 AM

AI-generated article based on recent web research, then preserved as a dated editorial snapshot.