Language models are trained on language. The next generation of AI will need to be trained in the physical world, where errors are more consequential.
Soumitra Dutta, former dean of Oxford's Saïd Business School and co-creator of the Global Innovation Index and the Network Readiness Index, says that next-gen AI will be about systems that understand causality, adapt to uncertainty and correct mistakes. Multiple research paths are leading to those capabilities. Turing Award winner Yann LeCun and other researchers have long advocated building representations of the physical world into AI systems. World Labs, a startup founded by Fei-Fei Li, is focused on spatial intelligence. Google DeepMind creates 3-D simulations to help develop embodied AI.
Where the US has the edge
This type of AI has unique characteristics compared with chatbots. Robotics and other physical AI projects require long lead times and cannot easily be tested and improved over short periods; results are unpredictable. These projects require long-term investment. The United States has an edge here. Its investors are accustomed to funding multiple companies in an industry, knowing that most will fail but some may help define it.
The research is hard. It requires collaboration across robotics, physics, materials science, and biology. Such expertise is not easily contained within one firm and does not develop rapidly. According to Soumitra Dutta, who is an AI scholar and AI entrepreneur, US advantages include the ease with which people and ideas move among firms, universities like Carnegie Mellon, MIT, and Stanford, and startups. New inventions move quickly to market and then feed further research.
China, a strong rival
China's advantages are different. Chinese companies sold more than 90% of the world's humanoid robots in 2025. Thousands of robotics-related patents were filed by Chinese applicants over the past few years. A significant competitive advantage for China is access to data for physical AI. Sources of data in China include its cities, factories, and electric vehicles.
The competition in the next phase of AI involves two sets of strengths. The United States has a strong research and development system and deep pools of capital and investors willing to back many risky, long-term bets at once. China’s advantages include robots deployed at scale and vast amounts of real-world data. Neither country has a clear advantage in all areas. “The outcome of the next phase of AI will depend on how these different strengths evolve-and how effectively each country connects them into a coherent strategy," says Soumitra Dutta.