Physis Raises $100M+ for Foundational World Model

In a significant development for the burgeoning field of world models, AI startup Physis (逆矩阵) has successfully closed a Seed++ funding round exceeding $100 million. This latest infusion of capital, which follows a prior multi-million dollar seed round just two months ago, underscores intense investor confidence in the company's vision for a unified, physics-aware AI foundation. The round saw participation from prominent venture capital firms including Matrix Partners China, 5Y Capital, and Huihe Capital, alongside a strategic investment from Ant Group. Existing backers such as Gaoling Venture Capital and Yanyuan Ventures also continued their support.

The funding propels Physis's mission to develop a foundational model capable of understanding and predicting physical interactions across diverse scenarios. The company recently unveiled its Physis-v0.1, a general-purpose world model designed for broad applicability, from embodied AI and industrial simulation to gaming physics and scientific forecasting. This model emphasizes physical accuracy, long-range consistency, causal reasoning, and broad generalization, aiming to serve as a singular "One For All" solution for physical world applications.

Chen Boyuan, co-founder of Physis, highlighted the rapidly narrowing window for developing such foundational models, estimating it to be approximately 18 months. This compressed timeline reflects the accelerating pace of AI innovation, drawing parallels to the swift evolution of large language models. The company plans to release its flagship model by the end of 2026, accompanied by open-source components and technical documentation. The newly acquired funds will be primarily allocated to the research and development of this general-purpose world model, alongside the construction of a scalable training infrastructure.

The team behind Physis is a unique blend of academic rigor and industry expertise, comprising young scholars from Peking University and seasoned engineers from leading technology firms. This AI-native structure eschews traditional hierarchical reporting and quarterly targets, prioritizing technical judgment and long-term vision. Chen Boyuan emphasized that the core differentiator for a true foundational model lies in its objective of physical prediction, a paradigm shift from previous AI advancements focused on next-word prediction or next-frame generation. The company's internal experiments suggest that their approach exhibits exponential scaling potential, akin to large language models, without hitting saturation points observed in more specialized models.

The market for world models is heating up, with investors increasingly concentrating capital on leading players. This trend is driven by the belief that AI's next major paradigm shift involves moving from virtual to physical understanding, focusing on predicting the evolution of physical states. Physis's strategy of building a unified model grounded in physical laws, adaptable to various applications, aligns with this emerging industry consensus. The company's focus on developing a robust, general-purpose foundation before deep dives into specific commercial applications reflects a commitment to long-term value creation, positioning them to capture a significant share of the future AI-driven physical world.

The implications of advanced world models extend far beyond gaming or simulation. In industrial settings, they promise to revolutionize predictive maintenance, optimize complex manufacturing processes, and accelerate the design of new materials. For robotics and embodied AI, these models are crucial for enabling agents to interact safely and effectively with the real world. The ability to accurately simulate and predict physical phenomena at scale could unlock unprecedented advancements across scientific research, engineering, and everyday applications, making Physis's pursuit a critical endeavor in the AI landscape.