China’s DeepSeek impresses. But is a ‘fast follow’ good enough in AI?

American stock markets shuddered on Monday, prompted by China’s announcement that it has created a capable, cheap, artificial intelligence machine. It’s the biggest cloud yet to darken the West’s blue-sky enthusiasm over AI, calling into question the efficacy of America’s export controls and the billions of dollars the United States is pouring into the technology’s expensive cutting edge.

China startup DeepSeek says its AI assistant uses less advanced chips than its rivals’ models do, and it costs less to train. Unlike the West’s billions, the Chinese model was developed for just $5.6 million, by one estimate.

“Are we going to spend $500 billion to get to the frontier so that China can find a way to copy our homework for pennies on the dollar?” asked Gregory Allen, director of the Wadhwani AI Center at the Center for Strategic and International Studies, on a podcast Friday.

On Monday on Wall Street, leading AI company stocks fell hard. The price of Nvidia, the leader in AI chips, plunged about 17%, as rattled investors evaluated the prospect that Chinese players could threaten U.S. tech profits. The S&P 500 fell nearly 1.5%, and the tech-heavy NASDAQ fell 3%.

But experts cautioned against panic. It’s important not to overstate the Chinese threat, analysts say. DeepSeek’s new AI model is both genuine achievement and Beijing hype. It shows China’s ability to simplify U.S. models, creating a way for fewer and less powerful computer chips to deliver answers that rival America’s more expensive models. DeepSeek said last week that its latest model can compete with a version of OpenAI’s ChatGPT released four months ago.

In many ways, DeepSeek’s model parallels China’s “knock-off” manufacturing prowess. And in this case, DeepSeek has put its product on the public domain, which means anyone can use it for free. By Monday, it had become the top-rated free app on Apple’s app store.

Testing export controls, propaganda tools

DeepSeek positions China as a fast follower of American technology. But that’s a far cry from challenging the U.S. position at the cutting edge, analysts say.

The “concern is overstated since, based on available data, we believe DeepSeek’s model … relies on larger foundation models to be developed first,” according to a Bank of America/Merrill Lynch research note on Monday. Followers, in other words, can’t be leaders.

The technology also serves as a propaganda tool for Beijing. By showing China’s capability, the regime hopes to prove that U.S. export controls don’t work. But analysts suggest that they merely show the failure of the 2022 export controls under President Joe Biden. Those controls have since been tightened.

It is perhaps no coincidence that DeepSeek announced it had put its AI software in the public domain last week after Mr. Biden issued three new rounds of export restrictions aimed at further curtailing China’s ability to access the most advanced chips.

“As the Land of Liberty, America – not China – must lead the world in the development of AI,” then-President Biden said in his farewell address on Jan. 15.

Analysts say export controls can’t stop the flow of technology. But they can make it harder and more expensive for China to compete and keep up with the technology, especially with President Donald Trump’s new public-private joint venture that could invest up to $500 billion in AI infrastructure.

Public domain and silver linings for U.S. firms

And while putting DeepSeek’s model in the public domain was a propaganda win in the short term, it may backfire. U.S. companies now have the Chinese code to simplify their own search engines. And with better access to powerful chips, the Americans can create their own near-cutting edge AI models at much lower cost and offer them, perhaps, for free.

Which model will the global market adopt? Technology that might give the Chinese government access to their data? Or Western private-sector products where, presumably, geopolitical risks are lower?

The bigger question DeepSeek poses is for investors. If many companies can operate with cheaper, good-enough AI, how much investment should flow into the technology’s cutting edge?

At present, those investments are losing money. If history is any guide, it will take several years to turn a profit in an intense race that may winnow the leaders to one or two. Think Microsoft’s Windows in personal computers. Or Google in Web search.

One thing is sure. The DeepSeek cloud of doubt on investors’ horizon could grow if investors begin to weigh more carefully the promise of AI against the costs to create it.