Seemingly out of nowhere, a small Chinese startup called DeepSeek released one of the world’s most advanced AI models, rivaling those of OpenAI, Google, and other American tech leaders. Chinese engineers accomplished this feat at a fraction of the cost, using none of the expensive, cutting-edge and prohibited US chips.  

Wall Street reacted with a massive sell off of US AI companies. Star chipmaker Nvidia slid more than 17%, wiping out more than $600 billion of market value. Perhaps even more importantly, the Chinese achievement undercut many European and American assumptions about the future of the revolutionary AI technology. 

The US believed that it was well ahead in the AI arms race. It imposed draconian restrictions on the sale of AI chips to China. DeepSeek shows that this policy may be a failure — or has even boomeranged, hurting US chip companies with lost sales more than it slowed down China. It forced Chinese engineers to become creative and design clever workarounds. “Because they had to figure out work-arounds, they actually ended up building something a lot more efficient,” said Perplexity CEO Aravind Srinivas. 

Europe feared that it had already lost the AI race. President Donald Trump’s $500billion artificial intelligence hardware plan announced as soon as he reentered the White House raised the question of whether the continent can compete. DeepSeek’s success shows that huge sums may not be needed to produce world-class AI products. It calls into question the massive spending Western companies and governments are undertaking into building AI models and splurging on data centers.  

Admittedly, much remains unknown about DeepSeek’s claims, including what sorts of chips the company had access to despite sanctions. Several chip analysts disputed the notion that DeepSeek built something on par with advanced US AI models at such a low cost. Others questioned whether aggressive AI spending on chips and data center will actually cool.  

Even so, DeepSeek’s success seems to show that second movers have a chance to catch up in the AI race. While pioneers such as OpenAI invest enormous resources to develop cutting-edge models, second movers such as DeepSeek can replicate and optimize these advancements at a fraction of the cost — between 10 to 100 times cheaper. By refining existing technologies and improving efficiency, they can economically outpace the original innovators.  

DeepSeek’s breakthrough marks the first time a Chinese AI model has been recognized by US industry leaders as a potential competitor to leading Western AI technologies. Founded in 2023 in Hangzhou, DeepSeek-R1 rivals OpenAI’s frontier reasoning model, GPT-4o, in math, coding, and reasoning tasks – at a fraction of the cost. It delivers similar performance while being 90-95% more affordable, making it an attractive alternative. 

The Chinese startup reportedly trained its model using Nvidia’s H800 chips costing under $6m. It introduced numerous innovations — both in AI architecture and data processing — many of which have been replicated by others. Andrej Karpathy, who previously helped co-found OpenAI, commented that it was a “highly impressive display of research and engineering under resource constraints.” 

Should Americans scrutinize Trump’s $500 billion Stargate project? Is it truly a national initiative, or merely a bailout for the large corporations sponsoring it? Ironically, while China, under sanctions, leverages market-driven innovation, US companies could burn through massive government-backed investments with little incentive to become more efficient. Ro Khanna, a US House Rep, commented on the matter, posting on X: “How about requiring Stargate to license some of the $500 billion of compute to startup firms so we can have new DeepSeeks with small talented teams emerge here instead of in China? Seems like an appropriate role for the federal government.” 

DeepSeek’s success serves as a stark reminder that true innovation thrives in diversity – not in a market dominated by a select few. It succeeded by drawing on open source code, a process boosted when Meta freely shared an AI system called Llama. The Chinese company has “open sourced” its latest AI model under the MIT license, allowing free commercial and academic use. Developers around the world can modify the code and build their own applications with it. 

If US regulators impose excessive restrictions on open source advancements, China could gain a decisive edge. If the most advanced open source AI models originate from China, US developers may inevitably build their systems on Chinese technology – potentially positioning China at the forefront of global AI research and development in the long run. 

DeepSeek’s success also calls for a rethink of assumptions on both sides of the Atlantic. China is a serious competitor and it succeeds not just with top-down government subsidies. DeepSeek is a nimble startup.  A European startup can prove just as innovative, particularly if it is not hamstrung with expensive, complicated regulations. The US government may rethink its reliance on export sanctions and its surprising turn to expensive industrial policy to counter Chinese advances.   

Another casualty for policymakers may be a rethink around AI safety. Both European and US policymakers have focused on reducing biases, errors – and “serious risk” that could harm national security or public health. Both have imposed some transparency requirements on closed proprietary AI models. DeepSeek, coming from China, reportedly refuses to answer questions about sensitive subjects such as the 1989 Tiananmen Square tragedy. Open source models allow anyone to review and test programmers’ assumptions and output. While raising alarm bells, DeepSeek may provide all of us with a much-needed wakeup call.  

David Kirichenko is a freelance journalist and an Associate Research Fellow at the Henry Jackson Society. He can be found on X/Twitter @DVKirichenko. 

Bandwidth is CEPA’s online journal dedicated to advancing transatlantic cooperation on tech policy. All opinions expressed on Bandwidth are those of the author alone and may not represent those of the institutions they represent or the Center for European Policy Analysis. CEPA maintains a strict intellectual independence policy across all its projects and publications.