The rise in popularity of a high-performing and cheaply built Chinese artificial intelligence (AI) model has shaken the confidence of investors, while raising larger questions about the future of American-made AI and upping the stakes of Washington’s tech rivalry with Beijing.
DeepSeek’s new AI model has taken the internet by storm, sparking a significant sell-off in the tech sector as investors fear the billions of dollars U.S. firms have invested into AI infrastructure may be unnecessary.
“It upends the way that investors have thought about how AI needed to be developed and implemented,” said Steve Sosnick, chief strategist at Interactive Brokers, who suggested the industry might be at an “inflection point.”
After launching its latest AI model, R1, last week, DeepSeek surged to the top of Apple’s App Store over the weekend. The application was the No. 1 free app on the store on Monday, while ChatGPT-maker OpenAI sat at the No. 2 spot.
DeepSeek claims its R1 open-source reasoning model has a “performance on par with” OpenAI, which is regarded as one of the U.S.’s leading AI reasoning models.
The company, founded in May 2023, claims to have spent just $5.6 million to train its latest models, The Wall Street Journal first reported.
The price tag pales compared to that incurred by leading U.S. AI firms like OpenAI, Meta and Google, all of which have spent billions of dollars in recent years on AI infrastructure and development of large language models.
A significant chunk of AI developers’ expenses in the U.S. goes to infrastructure, including the data centers and chips used to power the AI training process.
Now, DeepSeek is disrupting the market and showing how AI can be developed at a fraction of the price, while allegedly not relying on the vast data sets, chips and infrastructure thought to be the holy grail of AI development.
“Let’s say the convention wisdom as of yesterday, when you look at the AI market, is that essentially the best models come directly from those who have the deepest and broadest data sets and those that have the most brute force processing power behind them,” said Kenneth Lamont, a senior researcher at Morningstar, a financial services company.
“There was assumed to be a natural monopoly where if you’re the biggest player, that’s why we saw this massive concentration of these companies and they were just seen to be the clear winners in this space,” Lamont said. “China found a way to not rely on the chips.”
Nvidia, a leading producer of the chips behind the AI boom, saw its stock price plummet nearly 17 percent by market close Monday, shedding almost $600 billion in value. The semiconductor firm Broadcom similarly dipped 17.4 percent.
Taiwan Semiconductor Manufacturing Company’s stock sank 13.3 percent over the course of the day, while Arm was down 10.2 percent and ASML Holding’s share price dipped 5.8 percent.
Tech firms that have heavily invested in AI also took a hit in Monday’s sell-off. Shares in Oracle, one of three companies leading a new Trump administration project to invest in AI infrastructure, tumbled 13.8 percent.
Microsoft also dipped 2.1 percent, while Alphabet, Google’s parent company, was down 4.2 percent.
Callie Cox, chief market strategist at Ritholtz Wealth Management, noted that Monday’s market reaction was likely in part driven by the fact that the industry was “ripe for a sell-off.”
“Tech has done incredibly well and that means that the bar for excellence is incredibly high, and any one skeptical headline can really knock the sector off its axis,” Cox told The Hill.
With a disrupted narrative, Lamont warned that U.S. AI companies will now be facing increased pressure to justify their high expenditures.
“It really puts a huge pressure on them to justify their fees, whether that be they make better products — I’m sure that’s what they’re trying to do anyway — but also almost certainly to lower their fees to compete with this,” Lamont said.
OpenAI offers models at different price points, including a free version and “plus” and “pro” plans, which cost $20 to $200 a month, respectively. Google’s Gemini AI model can cost $0 a month with a Google account, but its premium plan costs $20 a month.
The surge in the China-based app comes nearly a week after President Trump was sworn back into office. On his second day, the president announced a joint investment of up to $500 billion to build the infrastructure needed to power AI over the next four years.
OpenAI is an initial investor in the venture, along with Oracle and SoftBank.
However, DeepSeek’s development could indicate that AI does not require this level of hardware investment, Sosnick noted.
“All of a sudden, there’s this upstart that threatens to do it more cheaply and more efficiently than what we thought,” he said.
DeepSeek reportedly claimed last year it had limited access to chips and used just 2,000 second-tier Nvidia chips to train its models v3 and R1.
Some AI business leaders have cast doubt about the company’s claims.
Scale AI CEO Alexandr Wang on Monday claimed DeepSeek has about 50,000 H100 graphic processing units (GPU), which can be used for AI development, but cannot discuss them because of the U.S. chip export controls in place.
When asked about DeepSeek, an Nvidia spokesperson told The Hill the Chinese company is “an excellent AI advancement and a perfect example of Test time scaling.”
“DeepSeek’s work illustrates how new models can be created using that technique, leveraging widely-available models and compute that is fully export control compliant,” the spokesperson said. “Inference requires significant numbers of NVIDIA GPUs and high-performance networking. We now have three scaling laws: pre-training and post-training, which continue, and new test-time scaling.”
The Hill reached out to DeepSeek for further comment.
Up until its final days, the Biden administration aggressively tried to curtail China’s advancements in chipmaking and AI through export controls on some semiconductor chips and equipment.
When asked about DeepSeek’s surge Monday, the Trump White House emphasized President Trump’s commitment to leading on AI and laid the recent advancements by China at the feet of the previous administration.
“By stifling innovation at home and failing to cut off China’s access to American technology, President Biden created an opportunity for our foreign adversaries to make gains in AI development,” a White House Office of Science and Technology Policy spokesperson said in a statement.
Trump repealed former President Biden’s AI executive order last week, stating it “established unnecessarily burdensome requirements” for AI developers.
He later signed a new order to ensure AI development is “free from ideological bias,” and called for an AI development action plan within 180 days.
“DeepSeek R1 shows that the AI race will be very competitive and that President Trump was right to rescind the Biden EO, which hamstrung American AI companies without asking whether China would do the same. (Obviously not.) I’m confident in the U.S. but we can’t be complacent,” David Sacks, White House AI and crypto czar, said in a post on the social platform X.
DeepSeek’s rise comes as the Trump administration weighs the future of TikTok amid a looming government ban.
After a brief shutdown earlier this month, TikTok came back online following assurances from Trump that he would not enforce a law that required the app’s China-based parent company ByteDance to divest by Jan. 19 or face a ban amid national security concerns.
The Chinese government has tried to distance itself from ByteDance’s business decisions, though some observers still question how close the two are intertwined.
“In China, the distance between the state and the private world is much more blurred than it is elsewhere,” Lamont remarked.
DeepSeek’s arrival in the AI scene “really puts the U.S. leadership in doubt in such a surprising way,” said David Bader, director of the Institute for Data Science at the New Jersey Institute of Technology.
“We’re in this incredible period of competition between the U.S. and China,” he said. “Here, we see a great example of China leapfrogging the U.S. based on their nimble, their agile development of this DeepSeek AI model, and I think this is just the start of what’s to come.”
