Just over 18 months after opening its doors, Shanghai-based embodied intelligence startup Psibot, known in Chinese as Lingchu Intelligence (灵初智能), has entered unicorn territory, closing a roughly $100 million round that values the company at approximately $1.48 billion. According to a report by The Next Web, the round was led by Chinese automaker Chery Automobile, with participation from Lens Technology, the Hunan-based glass specialist that supplies Apple and Tesla. The deal brings Psibot’s total capital raised since its 2024 founding to around $300 million, an extraordinary pace even by the standards of China’s frothy robotics sector.
The financing lands at a moment when Chinese embodied-AI startups are being repriced at breathtaking speed. GigaAI, another world-model specialist, filed for a Hong Kong listing this week at a reported $3 billion valuation, and Unitree Robotics received regulatory clearance earlier in July to raise 4.2 billion yuan ($620 million) on Shanghai’s STAR Market.
Psibot’s leap into the unicorn club fits neatly into this broader shift, which we recently examined in our coverage of China’s AI IPO wave, and reflects investors’ growing appetite for companies pursuing what many now view as the next frontier beyond large language models: physical intelligence.
A Founding Team Assembled Around ‘World Models’
What separates Psibot from the crowded field of Chinese humanoid ventures, a field that, as we reported yesterday, now includes more than 400 registered humanoid robot models, is its explicit bet that the future of embodied AI runs through so-called “world models,” neural systems that learn to simulate and predict the physical dynamics of the real world rather than merely map text tokens to responses.
The company’s leadership signals the seriousness of that thesis. Psibot was founded by Viktor Wang and is anchored by a trio of co-founders drawn from academia and industry: a dean at Peking University, a robotics veteran who previously led work at both Alibaba and Tencent, and a Stanford-trained researcher who studied under Fei-Fei Li, the computer-vision pioneer whose own startup, World Labs, has helped popularize the world-models concept in Silicon Valley. That pedigree has helped Psibot recruit engineering talent and attract strategic investors who see world models as the missing link between today’s LLM-driven chatbots and tomorrow’s general-purpose physical machines.
Chery’s participation as lead investor is particularly telling. The automaker, one of China’s largest exporters of passenger vehicles, has been aggressively expanding into autonomous driving and increasingly views world models as the connective tissue between self-driving cars and humanoid robotics. Both problem domains, in Chery’s view, ultimately reduce to the same challenge: teaching machines to anticipate how objects, agents, and environments will behave. Lens Technology’s involvement, meanwhile, hints at potential manufacturing tie-ups; the company’s precision components already sit inside iPhones and Teslas, and its investment gives Psibot a credible pathway to hardware scale.
The Two-Year Countdown to a ‘GPT-2 Moment’
Psibot’s founders have set a strikingly aggressive public target: they aim to reach what they describe as a “GPT-2 moment” for embodied AI within two years. The reference is deliberate. GPT-2, released by OpenAI in 2019, was not a mass-market product, but it demonstrated for the first time that scaling transformer models could produce qualitatively new capabilities, an insight that, four years later, became ChatGPT. Psibot’s leadership argues that embodied intelligence is approaching an analogous inflection point, at which sufficiently large, well-trained world models will suddenly begin to exhibit generalizable physical reasoning.
To fuel that scaling, Psibot has committed to collecting one million hours of robot training data during 2026 alone. Data acquisition is widely considered the binding constraint in embodied AI: unlike text, which has been indiscriminately scraped from the internet for a decade, physical-interaction data must be painstakingly recorded from teleoperation rigs, simulation environments, and deployed robots. A million-hour dataset, if achieved, would place Psibot among a very small group of companies globally with the raw material to attempt frontier-scale world-model training.
Not everyone in the industry shares the two-year timeline. At the Nikkei Global Digital Summit in Tokyo this week, Irving Chen, Unitree Robotics’ Asia-Pacific director, told attendees that the “GPT moment” for robotics remains “two to five years” away, citing the sheer number of environmental variables physical systems must handle. Chen noted that in the first nine months of last year, 74% of humanoid robot revenue came from research and education customers, with only 9% from industrial buyers, a reminder that commercial pull remains thin. “Right now, most robotics projects are in the pilot stage,” he said.
Psibot’s leadership is effectively wagering that Chen’s estimate is too conservative and that a combination of world-model architectures, aggressive data collection, and Chinese manufacturing scale can compress the timeline.
Why Investors Are Willing to Pay Up
The valuation Psibot commanded, nearly $1.5 billion for a company barely eighteen months old with no meaningful commercial revenue, reflects three converging bets on the part of Chinese capital.
First, world models are increasingly seen as the natural bridge between the LLM boom and its next act. If large language models mastered symbolic reasoning, world models are pitched as the technology that will master physical reasoning, unlocking applications from autonomous vehicles and warehouse robotics to household humanoids.
Second, Beijing’s policy environment is heavily tilted toward embodied intelligence. China’s 15th Five-Year Plan, whose contours we’ve been tracking, explicitly identifies humanoid robots and physical AI as strategic priorities, and Shanghai’s recent WAIC conference featured embodied AI as a headline theme. Local government financing vehicles, the “Hefei model” now spreading across China, are pumping capital into precisely the sectors Psibot occupies.
Third, and perhaps most importantly, capital is beginning to rotate out of pure-play foundation model developers and toward companies that can deploy AI into physical products. With DeepSeek, Moonshot, MiniMax, and Z.ai already commanding multi-billion-dollar valuations at the model layer, embodied AI represents one of the few remaining frontiers where a startup can still credibly claim to be building something structurally new.
Whether Psibot can deliver on its two-year timeline remains to be seen. But at $1.48 billion, its investors have made clear that they are no longer content to wait for the answer.