Manifold AI, a Beijing-based artificial intelligence startup, has completed a significant new funding round, bringing its total capital raised to nearly 1 billion yuan ($148 million) across six rounds since its founding just over a year ago. The fresh funding round, led by state-backed CRHC Fund and Temasek-backed InnoVen Capital, reflects growing investor confidence in “world models,” an emerging category of AI technology aimed at enabling machines to develop a sophisticated understanding of and interaction with the physical world.
The startup’s rapid fundraising trajectory, which has accumulated nearly $150 million in capital in just over a year, underscores the intense investor interest in world models technology. The funding round also included participation from BAIC Capital and KHK Fund, with existing shareholders increasing their stakes, indicating strong confidence from both new and established investors.
What Are World Models?
World models represent a frontier area of AI research focused on developing systems that can build and maintain internal representations of the physical world. Rather than simply recognizing objects or patterns in images, world models aim to understand causality, physics, and the consequences of actions in three-dimensional space. This capability is critical for applications ranging from robotics to autonomous vehicles to industrial automation.
The concept gained significant traction following OpenAI’s release of Sora, a video generation model that demonstrates sophisticated understanding of physical dynamics and object interactions. Sora’s ability to generate realistic videos of physical scenarios suggested that world models could be trained at scale using modern deep learning techniques. However, the technology is still in its early stages, with significant challenges around accuracy, generalization, and computational efficiency.
Manifold AI’s Product Strategy
Manifold AI has developed two primary products addressing different aspects of world modeling. WorldScape is a real-time world model that processes visual input and generates dynamic representations of physical environments. WorldScape Policy, by contrast, is a world action model focused on understanding and predicting the consequences of actions within physical environments.
This two-product approach reflects recognition that world modeling involves multiple distinct challenges. Understanding the current state of a physical environment (WorldScape) is fundamentally different from predicting how that environment will evolve in response to actions (WorldScape Policy). By developing specialized products for each challenge, Manifold AI is positioning itself to address different customer needs and use cases.
Competitive Landscape and Global Momentum
Caixin Global reported that Manifold AI’s funding round occurs within a broader context of accelerating investment in world models technology globally. US-based Odyssey announced a $310 million Series B funding round at a $1.45 billion valuation, indicating that world models companies are attracting venture capital at unprecedented scales. Additionally, Chinese startup GigaAI raised 1 billion yuan in June, suggesting that world models technology is attracting investment from multiple Chinese venture capital firms.
This global momentum reflects investor conviction that world models represent a critical frontier in AI development. As large language models have matured and competition has intensified, investors are increasingly looking toward adjacent AI technologies that could generate substantial value. World models, with their potential applications in robotics, autonomous systems, and industrial automation, represent one such frontier.
The Post-Sora Inflection Point
The surge in world models investment follows directly from OpenAI’s Sora release. Prior to Sora’s demonstration of sophisticated video generation based on physical understanding, world models were largely theoretical. Sora’s success suggested that world models could be trained at scale using modern techniques, transforming the technology from an academic research area into a practical, investable opportunity.
However, the technology remains at an early stage. While Sora can generate impressive videos, it still makes errors in physics simulation and struggles with complex, multi-step scenarios. Manifold AI and other world models companies are working to improve accuracy, generalization, and computational efficiency. The next 12-24 months will likely determine whether world models can achieve sufficient maturity to support commercial applications.
Applications and Market Potential
The potential applications for world models are substantial. In robotics, world models could enable robots to predict the consequences of their actions before executing them, improving safety and efficiency. In autonomous vehicles, world models could enhance prediction of how other vehicles and pedestrians will behave. In industrial automation, world models could enable systems to optimize complex manufacturing processes by simulating different scenarios before implementation.
These applications suggest a large potential market for world models technology. However, realizing this potential requires solving significant technical challenges around accuracy, generalization, and computational efficiency. Companies like Manifold AI are betting that these challenges can be overcome within the next few years, creating a multi-billion-dollar market for world models technology.
Implications for Chinese AI Development
Manifold AI’s funding round also carries strategic significance for Chinese AI development. World models represent a frontier technology where Chinese companies are competing on relatively equal footing with US companies. Unlike large language models, where US companies have established early leads, world models are still in early development stages globally. Chinese companies like Manifold AI have opportunities to establish leadership positions in this emerging category.
Additionally, the participation of state-backed investors like CRHC Fund suggests that Chinese government entities view world models as strategically important. This government support could accelerate Manifold AI’s development and deployment, potentially enabling the company to establish market leadership in China and Asia.
Investor Confidence and Future Prospects
The strong participation of both new and existing investors in Manifold AI’s funding round suggests confidence in the company’s technology and leadership. The fact that existing shareholders increased their stakes indicates that early investors believe the company remains well positioned to capture value as the world’s markets evolve.
However, the sustainability of investor enthusiasm for world models will depend on demonstrable progress toward commercial applications. If world models companies fail to achieve significant technical breakthroughs or demonstrate clear paths to profitability, investor interest could cool rapidly. Manifold AI and other world models companies will need to deliver tangible progress in the coming months to justify the high valuations and substantial capital they have attracted.
(Related: DeepSeek’s $7.4B Fundraise Comes With a Strict No-Poaching Clause)