China Plays the Long Game in AI While US Chases Superintelligence: Brookings

In brief Brookings says China’s AI industry is advancing through efficiency, global adoption, and integration into physical machines. While U.S. firms chase artificial general intelligence, the report said, Chinese firms aim to spread AI across devices, manufacturing, and global markets. Future of Life Institute’s Hamza Chaudhry suggests that a greater focus should be put on distillation attacks. The global AI race may not be unfolding the way policymakers in Washington had hoped. A new Brookings Institution report , published Monday, says the U.S. has framed the AI race as one driven toward artificial general intelligence, while Chinese companies are prioritizing efficiency, global adoption, and embedding the technology into real-world systems. “The U.S. is obsessed with the race to AGI or artificial general intelligence,” the report said. “American tech companies are pouring hundreds of billions of dollars into new data centers in the hopes of creating AI systems that can match or exceed human-level performance across most cognitive tasks.”  Hamza Chaudhry, AI and National Security Lead at the Future of Life Institute, said the difference reflects two competing views of how technological advantage will develop. “Readers should understand that first, AI development is not a story about two nations racing towards AGI,” Chaudhry told Decrypt . “Rather, it's a story of a handful of companies in Silicon Valley having an obsession with AGI, while companies in China are much more focused on getting this product in the hands of as many users as possible and embodying it across their economy.” The Brookings report also notes how Chinese developers are advancing along several tracks simultaneously, including improving model efficiency, expanding global adoption through open-source models, and integrating AI into consumer and industrial products. “While U.S. tech firms have been building out massive compute clusters with hundreds of thousands of chips, Chinese AI labs have been hyperfocused on squeezing greater performance out of limited compute and memory resources,” the report reads. Chaudhry said that emphasis on distribution and deployment reflects a broader adoption strategy. “China's whole game has just been to get this stack into the hands of as many people as possible in as many physical devices as possible,” he said. Brookings pointed to China’s rapid integration of AI into physical products such as vehicles, smartphones, wearables, and robotics. Companies are also expanding the use of autonomous systems, including robotaxis, delivery drones, and humanoid robots, rather than waiting for breakthroughs in superintelligence. Open-source training The report also said that Chinese AI developers are leveraging open-source AI models, many of which are publicly available online. Chaudhry said that the approach raises security concerns because governments and militaries can access open models. “There’s already public reporting that open-source models have been used by the Chinese military,” he said. “That’s a reality we’re already dealing with. In my view, what needs to change is our broader AI strategy in how we interact with the global community.” He said the Brookings report leaves open questions about the role of model distillation, a technique where AI systems learn from the outputs of more advanced models. “The most surprising thing was the relative lack of analysis about how much model distillation favors Chinese AI development,” Chaudhry said. “There’s a section on efficiency where the author argues this is primarily due to Chinese AI innovations, rather than the distillation attacks reported by Anthropic from DeepSeek, or the distillation attacks reported by OpenAI and DeepMind from unspecified companies.” Distillation attacks involve querying an AI model to collect its responses and using those outputs to train a competing model, effectively extracting the original system’s capabilities. In February, Anthropic claimed that several Chinese AI labs, including DeepSeek , Moonshot, and MiniMax, generated millions of responses from its Claude AI using thousands of fraudulent accounts to train their own models. Chaudhry said the differing priorities in AI development between the United States and China could create room for new arms-control-style agreements on advanced AI systems. “It opens up a unique space for a potential agreement on what we should not build in the future, in other words, red lines established by the United States and China on certain kinds of AI development,” he said. Daily Debrief Newsletter Start every day with the top news stories right now, plus original features, a podcast, videos and more. Your Email Get it! Get it!