China's Psibot reaches $1.48 billion investment in 'world models.'

      Viktor Wang has a particular interest in a historical moment. In 2019, OpenAI's GPT-2 demonstrated that a machine trained solely on internet text could write in a human-like manner. While it was basic, it was a significant step forward. Wang believes that robots are on the verge of a similar breakthrough and is seeking investment to be present at that moment.

      His startup, Psibot, also known as Lingchu Intelligence, is nearing the completion of a fundraising round that could raise nearly $100 million at a valuation of $1.48 billion, according to Bloomberg. The round is led by Chery, the automotive manufacturer, along with Lens Technology, a sensor supplier for Apple and Tesla. Founded in 2024, Psibot has thus far secured around $300 million in funding.

      The company focuses on world models, which are AI systems that enable robots and autonomous vehicles to perceive and interact with the physical environment, rather than merely answering questions. “World models aim to achieve something more significant than large language models and chatbots,” Wang noted. It’s a competitive landscape and a crucial battleground in the US-China AI rivalry.

      The team behind the company is noteworthy. Psibot was established by three individuals: a dean from Peking University, a veteran from Alibaba and Tencent in robotics, and a scholar from Stanford. The scholar was mentored by Fei-Fei Li, who played a vital role in the development of modern computer vision. This is not a gamble on the outskirts of the industry.

      Wang’s belief is firmly based on data, which he considers both an advantage for China and a limitation. Psibot collects its data through custom gloves and humanoid robots, testing its systems with a prominent logistics company and a leading fibre-optic cable manufacturer in China. The challenge lies in achieving both volume and quality. “Even the top AI labs in Silicon Valley, like OpenAI and Meta, struggle to obtain good data,” he remarked. His target this year is to collect one million hours of data.

      He is not alone in his pursuit. Chinese companies are among the most proactive developers of world models. They benefit from government backing, access to industrial data, and a thriving open-source community. The company's own narrative showcases a rapid ascent from initial angel funding to unicorn status in just under two years. Recently, another Chinese startup focusing on world models, GigaAI, applied for a listing in Hong Kong.

      Wang has set a timeline for this advancement. “We should witness the GPT-2 moment in embodied AI within two years,” he stated. He refers to the point when machines begin to make generalizations about the physical world, similar to how chatbots did with language. He could be premature or mistaken.

      However, a valuation of $1.48 billion in a sector that hardly existed two years ago speaks volumes. Serious investors are prepared to synchronize their timelines with his. Previous funding rounds, as reported by Sahm Capital, have also attracted state-backed and industrial investments.