The latest episode of The President’s Inbox is live! This week, Jim sat down with Adam Segal, the Ira A. Lipman chair in emerging technologies and national security at the Council, to discuss how Chinese company Deepseek’s new AI model has upended assumptions about the state of U.S.-China technology competition.

Here are three highlights from their conversation:

1.) DeepSeek’s new AI matches as U.S. models at a fraction of the cost. Developing AI models was thought to require access to chips, a vast energy supply, and deep pockets. DeepSeek’s new AI chatbot throws those assumptions into question. It rivals the large language models developed by U.S. firms but was built with fewer advanced chips and at a lower price tag. As Adam put it, “The assumption was that U.S. AI was far ahead… but [DeepSeek] was able to do the same with less.” Although DeepSeek may have invested more on its model than it originally let on, its success shows that advanced AI models can be built easier and faster than previously assumed. This is good for innovation broadly, but not necessarily for U.S. AI firms.

2.) U.S. export controls are encouraging China to innovate. The Biden administration sought to stifle China’s technology development by banning the sale of high-end semiconductors to Chinese firms. Unable to match the advantages of firms in the United States, DeepSeek built better software so that it could use a limited number of chips to train its AI model faster and with less computing power. While a lack of chips has made it harder to build AI models in the traditional manner, more Chinese firms are now competing and developing alternative techniques. As Adam said, “Export controls are catalyzing changes in the Chinese system that are going to make it more innovative.” As is commonly said, necessity is the mother of innovation, and Chinese firms will continue to work with what they have and work around what they don’t.  

3.) Whoever is leading in AI today may not be leading tomorrow. DeepSeek’s success threw into question the assumption of U.S. dominance of AI. Traditional advantages in access to chips, energy, and finance may not keep firms at the cutting edge. However, no single breakthrough will “win” the competition. Other AI innovations are certain to come, making the AI race much more of a marathon than a sprint. As Adam noted, “Unlike Sputnik with the moon, AI doesn’t have an endpoint.” AI firms, whether they be in the United States, China, or elsewhere, are redoubling their efforts to build the next breakthrough and grab the lead.  

If you’re looking to read more of Adam’s analysis on the future of AI policy, check out the piece he co-wrote for the CFR blog Net Politics titled “Artificial Intelligence Priorities for the Next Administration."