Tian Keyu's unnamed AI lab reportedly raises nearly $30M for world-model research
The AMW Read
A newly funded lab adds a distinct visual-representation approach to the foundation-model player map, but its limited team, absent product and unvalidated performance claims constrain demonstrated impact.
Tian Keyu's unnamed AI lab reportedly raises nearly $30M for world-model research
Former ByteDance researcher Tian Keyu has founded an AI laboratory that reportedly raised nearly $30 million from investors including 5Y Capital and IDG Capital, reaching a $200 million post-money valuation, according to Jiemian News. The roughly 10-person team has not announced a company name or released a finished product. It is developing world models intended to learn physical rules and predict how objects and environments change, with potential applications in robotics, autonomous driving and interactive video.
The investment backs a foundation-model research effort before product or commercial validation. Tian's proposed approach uses a proprietary vocabulary of 200,000 visual symbols to compress video, represent information and predict future frames. He plans to train on about 100 million hours of video and says the approach could reduce the cost of generating one second of video by at least tenfold. A complete model is expected in 2027, according to his stated plan. These remain research targets: the article provides no model test results establishing the claimed savings or physical-prediction capabilities. Robotics applications alone do not establish a robotics business; the team's current focus is the underlying model.
For investors, the concrete diligence question is whether the team can turn its proposed representation into measurable capability and cost advantages. Evidence would need to distinguish cheaper video generation from reliable prediction of physical outcomes. Collaboration controls also warrant scrutiny: the report recounts Tian's dismissal from ByteDance over interference with colleagues' training tasks and his acknowledgment of serious misconduct. Funding establishes investor backing, while technical performance and commercial demand remain unproven.
