DeepSeek: What We Know And What We Don’t Know
Don’t take all of the claims about China’s new AI model at face value.
Following the release of DeepSeek (China’s new AI model), Wall Street experienced an unprecedented sell-off of AI-related tech stocks on Monday. This included the single largest loss by one company in one day, the nearly $600 billion lost by Nvidia, maker of the advanced chips that have spurred the race to AI supremacy of late.
The reason for the sell-offs and losses? The reigning narrative seems to be based on two things. First, DeepSeek is reported to have achieved top-ten AI model status by only spending a self-reported $6 million, much less than what is being spent by the world’s leading firms. Second, DeepSeek did it with limited access to state-of-the-art chips that its competitors have been relying on to build their own AI models.
Yet, how do we know that it took only $6 million to build DeepSeek? Chinese companies and the CCP are known for misrepresenting such numbers and for lack of transparency. Are we sure that is all they spent? Companies like Nvidia are losing value because of the statement that DeepSeak only spent $6 million to develop its model, implying that Nvidia and other leading AI development companies have been foolish, wrong-headed, or possibly even profiteering. Yet, we don’t really know how much it cost to build DeepSeek’s model.
How do we know DeepSeek didn’t get any help from advanced Western chips? An article from The Wall Street Journal highlights that DeepSeek did get access to such chips before the export controls kicked in in 2022. Moreover, do we know they didn’t get access to even more advanced technology from outside of China after the controls kicked in? We do not. So, we don’t really know the extent of DeepSeek’s self-reliance when it comes to chips.
In China, whether a company is “private” or state-run, it is subject to intrusive laws like the 2017 National Security Law, which requires companies to cooperate with the Chinese party-state. Given the stakes for the CCP and for China, it is likely that the party-state was closely involved with the leading AI research firm. Does DeepSeek’s $6 million figure account for state subsidies and support? If it did get help from the Chinese state, how do we know that it was also not given stolen intellectual property from advanced companies outside of China? Again, we don’t know for sure. Yet, this would be par for the course as far as how the Chinese state and Chinese companies have operated in the past. We don’t know whether DeepSeek got such help, or if so, how much, but we should seek to find out more about this before major sell-offs or changes to AI model development schemes.
Similarly, how do we know that the DeepSeek app is safe and not another Chinese data-gathering operation? This an important question given all of the recent hoopla over the question of a TikTok sale or ban.
The short answer is we do not know if it is safe. Some might respond that DeepSeek is open source, so how dangerous could it be? That is true, but it is currently the single most downloaded free app in U.S. app stores. There are already concerns that the app is vulnerable to malware. Given all the hard work of the U.S. government to try to protect its citizens from Chinese spying, are we now simply downloading another app with even greater potential surveillance capabilities?
Given the lack of information we have about DeepSeek and what it says about itself, we should be cautious about assuming DeepSeek is simply a Cinderella story of a scrappy team of recent graduates beating Silicon Valley’s best and brightest. If it turns out that this story is true, that is great, exciting, and commendable. But let’s be sure.
Despite the enormous Wall Street sell-off on January 27, cooler heads should prevail. Stockholders, companies, and engineers should not assume that they’ve been on the wrong track. They should learn from the successes of DeepSeek, but they should not take at face value some of the statements that have been made about DeepSeek. As Deng Xiaoping said, somewhat commonsensically, “Seek truth from facts.” We can learn from China and Deng in this case. Let’s get the facts before we react to the latest developments in this crucial technology.
Gregory J. Moore is a Professor of International Relations in the Department of Government at Patrick Henry College and a Senior Fellow at the National Institute for Deterrence Studies. Previously, he was a Stanton Fellow at the U.S. Air Force Academy, a professor at Colorado Christian University, the University of Nottingham (Ningbo, China), Zhejiang University (Hangzhou, China), and a fellow at the Foreign Policy Institute of Johns Hopkins University’s School of Advanced International Studies. His research interests include international relations, security, Chinese foreign policy, U.S. foreign policy, Sino-American Relations, East Asian IR/security issues, emerging technologies, and politics.
Image: Mamun_Sheikh / Shutterstock.com.