Via radii.co
Chinese startups are raising billions to build AI that understands the physical world, with robotics funding already surpassing all of 2025 by mid-year
Chinese startups are now pouring capital into “world models,” a category of AI designed not just to generate text but to understand how the physical world actually works, from gravity to grip strength to the geometry of a warehouse floor.
The numbers are staggering. Chinese robotics startups alone raised roughly $5.6 billion through 176 deals by mid-May 2026, already exceeding the total raised across all of 2025. The broader AI startup ecosystem pulled in over $16 billion in Q1 2026 alone.
What are world models, and why does the money care? #
A large language model predicts the next word. A world model tries to simulate reality. It learns physics, spatial relationships, and cause-and-effect so that an AI agent, say a humanoid robot or a self-driving car, can navigate the real world without crashing into everything. World models are the bridge between digital intelligence and physical competence.
Shengshu received a funding round of over 290 million yuan (north of $290 million) led by Alibaba Cloud in April. The company’s goal is to build a multimodal “general world model” with applications spanning physical robotics and autonomous driving.
Manifold AI accumulated nearly 1 billion yuan, approximately $148 million, by June through a mix of state-backed investors and Temasek-linked capital. Striding AI announced a nearly $100 million angel round around June 23 focused specifically on world models and humanoid robots.
The unicorn factory is running overtime #
AI² Robotics and X Square Robot have both been valued at over $2.9 billion in their 2026 funding rounds, earning unicorn status. State-backed investors are participating in nearly every major round, blurring the line between venture capital and industrial policy.
What this means for crypto and tech investors #
None of the major world model or robotics startups have issued tokens, integrated blockchain infrastructure, or signaled any interest in doing so. The capital flowing into Chinese AI is overwhelmingly traditional equity investment, much of it sovereign or quasi-sovereign.
US export controls on advanced chips were supposed to slow China’s AI progress. World models require different computational profiles than pure text generation, and China’s chip ecosystem, while constrained at the bleeding edge, may be sufficient for many robotics applications.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our