DeepSeek's big shake-up of AI has bigger policy lessons

Governments are not good at picking winners and big vanity projects for tech leadership can let us down

- RAHUL JACOB

In today's world, the only predictable constant, as the cliché goes, is unpredictability. On 21 January, US President Donald Trump appeared with OpenAI's Sam Altman, Oracle's Larry Ellison and Softbank's Masayoshi Son to announce a $500 billion artificial intelligence (AI) project. "This monumental undertaking is a resounding declaration of confidence in America's potential under a new president," Trump declared. Not to be outdone, Son said this amounted to the dawn of a "golden age."

Less than a week later, the loss of a trillion dollars in market capitalization suffered by tech companies on 27 January alone on US stock markets appears to be some kind of bizarre celestial retribution. Investors grew nervous about the competitive challenge posed by the Chinese startup DeepSeek, which has developed highly efficient large language models that appear to match its American counterparts using 2,000 chips versus 16,000. It then went one better and published a paper, detailing its methods. On 10 January, DeepSeek's first free chatbot app was released and has since quickly become the most popular free app on Apple's app store. It is not perfect. As with China's other tech companies, DeepSeek keeps all criticism of the Chinese Communist Party off limits, but few users will care about such censorship if its AI models work almost as well as their US counterparts at a fraction of the cost or for free.