Embodied AI race: China can look beyond larger models

The rapid advancements in robotics, showcased by robots capable of walking, dancing, and performing complex movements, have not yet translated into significant commercial value. This highlights a persistent challenge confronting the burgeoning embodied AI sector: achieving widespread adoption requires robots to be general-purpose, reliable, and economically viable simultaneously.

The primary bottleneck remains access to high-quality physical interaction data, the essential fuel for developing the next generation of intelligent robots. Building general-purpose embodied AI models could necessitate tens of millions of hours of real-world interaction data. However, as of early 2026, the global pool of compliant, high-quality physical interaction data is estimated to be only around 500,000 hours, a stark deficit.

Different regions are adopting distinct strategies to address this data challenge. In the United States, major tech companies are pursuing a scale-driven approach, focusing on larger AI models, increased computing power, and extensive AI infrastructure. This strategy inherently favors organizations with substantial capital and computational resources.

Meanwhile, Europe and Japan are prioritizing regulation, safety, and hardware advantages in their approach to AI development. However, they have yet to establish a clear leadership position in AI algorithms and large-scale data ecosystems. This evolving landscape, extending beyond chips and AI into robotics, is increasingly influenced by geopolitical competition, as evidenced by recent US restrictions on Chinese-made humanoid robots, underscoring the growing link between hardware access and international relations.