Singapore Video Startup Founded By Tencent’s Former AI Head Bets Big On World Models

A Singapore-based video startup, founded by the former head of AI at Tencent, has raised $80 million from AMD and Hyundai Motor to build the next frontier of artificial intelligence: world models that generate real-time interactive 3D environments on the fly. Video Rebirth is now competing neck-and-neck with OpenAI, Google, and Meta on major AI video leaderboards, signaling a seismic shift from static video generation to dynamic, physics-aware simulations.

The company was founded in 2023 by Dr. Li Wei, who previously led Tencent's AI research division and oversaw the development of its video foundation models. Video Rebirth's core technology—a transformer-based world model—doesn't just create clips; it learns the rules of motion, light, and object interaction to produce 3D scenes that users can explore and change in real time. The $80 million Series B round, co-led by AMD's venture arm and Hyundai's mobility fund, values the startup at over $600 million, sources say.

Why now? The AI video space has exploded since the launch of OpenAI's Sora in early 2024, with Chinese startups like Kuaishou and Zhipu AI also racing to market. But world models represent a more ambitious leap: they aim to build an internal representation of the physical world that can predict consequences—a crucial capability for robotics, autonomous driving, and gaming. Video Rebirth's approach is to train its model on massive datasets of video and 3D scans, then distill that knowledge into a real-time engine.

Key details: The startup's leaderboard, run by a consortium of AI labs, currently ranks Video Rebirth's 'Phoenix-1' model as the second-best for video quality and consistency, just behind Sora but ahead of Meta's Emu Video. The company has already signed pilot agreements with two major game studios and an autonomous vehicle company. Its team has grown to 120 people across Singapore, Beijing, and Shenzhen.

Analysis: World models are a hotly contested concept in AI research. LeCun has promoted them as a path to common sense; Bengio calls them essential for safe AI. If Video Rebirth can deliver a production-grade world model that works in real time, it could unlock services from virtual prototyping to immersive education. The involvement of AMD suggests the chip giant sees a need for its GPUs in training and inference beyond just language models.

Outlook: Video Rebirth plans to release a public API for its world model later this year and is exploring a Singapore IPO in 2027. Key milestones to watch: whether Phoenix-1 can top the leaderboard, and whether its 3D environments meet the latency requirements for live interactive applications. If it succeeds, the startup could redefine how machines perceive and simulate reality.