China’s AI ‘Six Dragons’ Mapped Out Ambitions and Caution in Wuzhen
rare joint appearance by the leading Chinese AI startups
Alright, the six dragons met in Wuzhen. I don’t think this has ever happened, as many of them are quite shy. So, I compiled some information from credible Chinese media sources, including Yicai, Caijing, Xinhua, and WeChat blogs, to piece together what the leading six AI startups in China shared.
In the misty canals of Wuzhen, a county near Hangzhou, the six dragons all came out of their caves for this one big rah-rah. Probably due to a combination of encouragement from local authorities and pressure to share more ahead of IPO speculation and interest, the new generation of Chinese AI leaders laid out their vision for the next decade.
Dubbed the “Six Dragons,” these companies have been at the forefront of LLMs, robotics, and spatial AI. The founders and executives shared how they are navigating the transition from the internet era to the age of artificial intelligence, leveraging China’s vast data and manufacturing advantages while grappling with the profound risks and technical bottlenecks that lie ahead.
1. DeepSeek: The Open-Weight Accelerator Sounds an Alarm
“Since its establishment, DeepSeek has followed an open-source approach with the goal of achieving AGI,” said Chen Deli, senior researcher at DeepSeek. “Through exchanges with technical developers in the community, we have gained valuable feedback, which in turn has driven DeepSeek’s own development. Therefore, we believe that open-sourcing technology fosters better collaboration and sharing.”
DeepSeek, the LLM company that broke the internet, is staking its future on an open-source strategy in the pursuit of AGI. Senior researcher Chen Deli said this collaborative approach has been central to its progress, with community feedback sharpening the company’s technology.
He also struck a sober note on AI’s long-term trajectory, arguing that displacement could begin within five to ten years and that, over a 10- to 20-year horizon, AI could replace the vast majority of knowledge worker jobs.
Unlike past industrial revolutions, he suggested, AI may become the primary intelligence, potentially leaving humans without new roles to fill. While confident the technical problems will be solved, he said tech companies must also act as guardians of social order to manage a transition that could see humans “completely liberated from work.”
2. Unitree Robotics (宇树科技): Betting on an Imminent Breakthrough in Embodied AI
Unitree Robotics, a leader in legged robots (humanoids & quadrupeds), is betting that embodied intelligence is on the cusp of moving from sci-fi to daily reality. His thinking is: open algorithms + manufacturing depth = faster physical AI.
Founder and CEO Wang Xingxing voiced bullish optimism, arguing that robotics is closer to a breakthrough than more distant frontiers like nuclear fusion or Mars colonization, and that meaningful progress should arrive within two years. (Seems much more bullish than other industry players so far)
But he was candid about growing pains. Compared with large language models that benefit from abundant internet data, robotics lacks standardized data at scale. There is also no consensus on basic hardware principles, including where cameras should be placed on a robot.
Despite these early-stage hurdles, which he likened to the nascent days of AI models, Wang’s conviction is clear: cracking the general model for robotics is tantamount to achieving AGI. He credited China’s manufacturing ecosystem and the global open-source community for speeding the field’s pace.
3. Manycore Tech(群核科技): Open-source Spatial Data
Co-founder and chairman Huang Xiaohuang recalled 2021 as a pivotal year. With user growth and data compounding rapidly, Manycore open-sourced its spatial dataset, earning industry recognition and opening new opportunities in spatial intelligence.
Huang emphasized that more than a decade of China’s internet expansion created massive user bases and data, providing a solid foundation for training large AI models, and that spatial intelligence is becoming an essential direction for large-model development.
As the backbone technology for many robotics companies we’re seeing, Manycore may be pushed to the forefront now, especially as Feifei Li has just coined that spatial intelligence will be where we see the next breakthrough.
4. DEEP Robotics (云深处科技): Humanoid and Quadruped Robotics
Founder and CEO Zhu Qiuguo described the path from R&D to deployment. He shared a vivid glimpse into that evolution, noting that in 2015, their first robot dog had to be suspended by a steel cable because it couldn’t stand or walk on its own. The milestone of a freely standing dog in 2018 was followed by years of iterative development to address practical issues such as wear and tear and overheating for commercial applications. That persistence, he joked, has finally allowed them to evolve “from making dogs to making humans,” a reference to their recently launched humanoid robot designed to replace people in hazardous jobs.
This year, the company released a humanoid robot designed to replace or assist humans in hazardous environments. Zhu also pointed to industry-wide constraints: despite bold visions, like a robot soccer team beating humans by 2050, progress is checked by challenges in data collection and standards. (echoing what Unitree founder Wang Xingxing was saying) And added that, without external support, startups struggle with autonomous mobility, data capture, and the large compute required.
5. BrainCo (浙江强脑科技): BCI Moves from Lab to Clinic
Founder and CEO Han Bicheng highlighted a staggering statistic—approximately 36% of medical spending is linked to brain-related conditions like Alzheimer’s and autism, which currently have no pharmacological cures. He said the company is working to introduce a non-invasive solution to clinics to help solve this issue. Their neural-controlled prosthetics already allow people with limb disabilities to use “thought” to accomplish fine-motor tasks such as writing and playing the piano, and to restore a natural gait.
As brain-computer interface (BCI) moves from lab to practical use, the industry is also exploring ways to help blind people “see.” Han framed BCI as part of human-centered digitalization—not only making machines smarter but improving care for people.
Challenges remain significant: complex neural-signal parsing and high stability requirements, especially under complex signals. Even so, BCI may offer therapeutic potential for Alzheimer’s and other brain disorders.
6. Game Direct (游戏科学): Creative Gains and Warnings of Concentration of Power
Game Science, the studio behind Black Myth: Wukong, sees its success as proof of China’s vast domestic market and rising appetite for top-tier local content.
Founder and CEO Feng Ji highlighted a dual reality. On one hand, AI empowers individuals; some creators already use AI to complete lyrics, music composition, and video production that once required a full team. On the other hand, he warned of the risk of excessive concentration of technological power.
With caution, he echoed the broader call to embrace AI rationally while guarding against risks.
Common Thread
Across the six, the message is consistent: open collaboration, internet-era data foundations, and manufacturing strength are speeding progress, while embodied-AI data and standards remain real bottlenecks.
Their consensus view is that once key thresholds are crossed, AI will advance in leaps, and building trustworthy AI will require technology and regulation working in tandem, so society can effectively embrace the present and AI.
What’s comforting to see is that all of them acknowledged the societal risks AI could bring, despite being the builders of the technology. And whether they’re in model or physical AI, the shared belief is that open source/ weight technology is actually the better model and pushes proliferation faster.
If you take a step back, the Chinese AI companies embracing open-source are essentially subsidizing the common goal of boosting the Chinese economy, as no one has really figured out a clear monetization strategy yet. The ultimate winners of this are Chinese consumers and the economy as a whole, through productivity gains and unlocked business opportunities.
Side note: I was listening to a podcast and editing this post while sipping on coffee at FinePrint today, to my left sat two millennial women dressed in all pink debating why women should even marry and isn’t traveling much more fun, and to my right were two older men talking about creating a demo and pitchbook for a robot protype that will clean and act as a security guard for office buildings at night. I turned off my podcast and eavesdropped for a solid ten minutes, and it was like listening to the two versions of myself talk past each other.
Btw, Geopolitechs provides the full transcript of the event here:
And some relevant reads FYI:
Rise of China’s Robotics Industry: from Manufacturing Arms to Embodied AI
Introducing Unitree, China’s leading AI-embodied Robotics Company
The Jevons Paradox in AI Infrastructure: DeepSeek Efficiency Breakthroughs to Drive Energy Demand
Everybody Losing Sleep Over DeepSeek: Industry Implications to LLMs and AI Infrastructure
DeepSeek V3 puts China AI on the global map: consumer use and capital expenditure implications
Thank you for linking my substack, Grace.
This article comes at the perfect time, giving such great insite into these companies. What are your thoughts on how this open-source push from DeepSeek might influence the global AI landscape?