DeepSeek intensifies U.S.-China race for AI supremacy
While U.S. AI vendors still have the upper hand, China has made big gains in AI technology even with U.S. export controls on AI chips and new tariffs on Chinese goods.
The rapid rise of Chinese AI startup DeepSeek jolted U.S. tech companies from an AI slumber and highlighted the escalating AI race between the U.S. and China.
The emergence of DeepSeek in late January with its low-cost, powerful large language model, DeepSeek-R1, stunned U.S. technology investors worried that the big generative AI vendors in the West have been overspending and overrelying on expensive data centers and compute.
The longstanding geopolitical tension and economic competition between China and the U.S. spread to AI about a decade ago, but has markedly intensified with the rapid ascent of DeepSeek and other Chinese generative AI vendors. The recent success of Chinese vendors also raised questions about the effectiveness of measures such as export controls in limiting foreign adversaries' technological innovation.
The U.S. applied export controls under former President Joe Biden in 2023, limiting U.S. companies' ability to sell advanced AI and semiconductor technology to China. This made it difficult for DeepSeek and other Chinese vendors such as Huawei, Alibaba, Baidu and Tencent to acquire the hardware they needed to compete in the AI race. Biden followed up by signing an executive order restricting U.S. companies' investments in Chinese technologies.
Trade war tactics
President Donald Trump has taken a different approach to China, shifting away from export controls toward tariffs and other trade restrictions. In late February, Trump issued an America First Investment Policy memorandum to restrict Chinese investment in U.S. technology. Foreign investment in U.S. technology will also depend on those countries' distance from the "predatory investment and technology-acquisition practices of the PRC [People's Republic of China] and other foreign adversaries or threat actors."
Despite the constraints, the Chinese tech vendors continued to make headway in the AI race.
For instance, Alibaba -- already the world's fourth-ranked cloud provider -- has remained a contender against U.S. AI leaders such as OpenAI with January's release of the Qwen family of foundation models and image generator Tongyi Wanxiang in 2023. Baidu, another Chinese tech company, also competes in the generative AI market with its Ernie LLM.
Even with its recent surge, China has long held a strong position in the AI market. The country's tech leaders -- with significant government support -- continuously worked on developing new AI technologies alongside U.S. AI vendors such as Google, Meta, AWS and OpenAI.
China's been doing AI for a very long time ... and is probably just as good as the U.S. or anybody else. It's as simple as that.
Alan Pelz-SharpeFounder, Deep Analysis
"China's been doing AI for a very long time ... and is probably just as good as the U.S. or anybody else. It's as simple as that," said Alan Pelz-Sharpe, founder of AI analyst firm Deep Analysis.
Meanwhile, in the U.S. Congress, lawmakers have expressed concerns about Chinese nationals in the U.S. During a hearing in January assessing China's influence, Sen. James Risch (R-Idaho) voiced fears about collaboration with China on science and technology projects.
"China siphons American technology and intellectual property from companies it partners with, repurposing it for their own economic benefit and riding on the back of American ingenuity," he said. "China pumps money into American universities and think tanks, attempting to secure access to sensitive military technology."
The U.S. is transitioning from a close research partnership with China to a military rivalry that will reduce or end cooperation and collaboration, said Jennifer Lind, an associate professor of government at Dartmouth College. Though relations with China started to become strained during former President Barack Obama's administration as the Chinese government became more assertive, Lind said she expects the relationship to become even rockier under Trump as the countries go head to head on technological innovation.
"Trump will definitely continue to lead the U.S. in the direction of an intensifying military and geopolitical rivalry with China," she said.
China's AI technology
The shockwave that DeepSeek set off, which triggered a massive sell-off in technology stocks led by investors dumping nearly $600 billion in Nvidia shares, came because many U.S. AI vendors have led the larger tech market to believe that sums on the order of hundreds of millions of dollars are needed for AI to be successful.
But DeepSeek said it spent less than $6 million to train its model -- though some observers have been skeptical, arguing that DeepSeek was not entirely forthcoming about its costs. In contrast, OpenAI CEO Sam Altman has said the vendor spent more than $100 million to train its GPT-4 model.
"It's still innovative," said JB Baker, vice president of marketing and product management at ScaleFlux, an AI vendor that develops system-on-chip software, referring to DeepSeek's LLM. "I don't think it's as earth-shattering as initially proposed or positioned."
Despite the questions about what it spent to train R1, DeepSeek helped debunk a belief in the inevitability of U.S. dominance of the AI market.
"It's all about money," Pelz-Sharpe said.
In any case, DeepSeek's entry into the market narrows the technological gap between the U.S. and China, according to Forrester Research analyst Charlie Dai.
DeepSeek's LLMs signify not only China's progress and innovation in AI technology, but also the Asian superpower's ability to produce competitive AI systems at a much lower cost and computing power than typically required, he said.
DeepSeek's innovations will likely spur Chinese companies and tech vendors to accelerate efforts to develop domestic semiconductor capabilities and advance the surrounding technology ecosystem, Dai said. It will also drive global AI investment in chipsets as cost reductions and efficiency improvements in model training create a paradigm shift in training approaches, he added.
"The AI workloads will accelerate shifting from training to inferencing, and as a result, large-scale AI data center investment will increasingly shift toward more distributed architecture," Dai said.
Dartmouth's Lind said while China still lags behind the U.S. on most innovation metrics, the country's recent rapid ascent is "stunning." DeepSeek shows that independently operated, innovative companies can still rise in China despite the country's state-centric innovation model.
"DeepSeek shows what China's talented, well-trained workforce can achieve," Lind said.
The view of the U.S. as a top competitor doesn't change for China, and its prerogatives will remain technological innovation and AI developments, said Daniel Castro, vice president at the Information Technology and Innovation Foundation (ITIF).
For the U.S., DeepSeek should serve as a wake-up call, he said.
"They're seeing that China, which some in the West tend to discount, is really making rapid headway in AI just as it's made rapid headway in other advanced sectors it's concentrated on," Castro said.
This idea that effective generative AI models need to cost a lot to train and run stemmed from the theory that the more GPUs a vendor had, the more likely that vendor could be the winner in the AI race. Even the U.S. government supported this idea, highlighted by the Trump administration's support of projects like the Stargate collaboration among OpenAI, Oracle and Softbank, in which investment money will be pumped into AI vendors to build more AI hardware infrastructure in the U.S., particularly big new data centers.
"Against this backdrop, it's easy to see why DeepSeek looked like an inspiring David to the U.S. big tech Goliath," said Kashyap Kompella, CEO of RPA2AI Research. "It's the classic underdog story and sowed doubts that maybe scale and deep pockets don't guarantee [a return]."
The U.S. faces off against China in an AI race.
China's AI known worldwide
Beyond its low-cost claims, DeepSeek's instantaneous success underlined China's AI technology innovation.
"An interesting phenomenon is the amount of interest DeepSeek has generated around the globe," Kompella said. "It elevates their innovation brand."
Like the U.S., China has a highly developed AI ecosystem. It boasts advanced AI models such as Antelope for the manufacturing industry, SenseNova for legal and Baidu Lingyi for life science, he noted.
At the same time, DeepSeek's open source strategy threatens AI vendors in the U.S. If model weights are available, users in the U.S. who might not like Chinese government censorship of Chinese LLMs can customize DeepSeek's models. Enterprises and vendors in other countries can do so as well, Kompella said.
"Many countries across the world will be open to Chinese AI solutions," he said, adding that the worldview of the company that creates an AI model is not as important as the model itself.
Meanwhile, Chinese citizens' trust in AI technology is stronger than that of U.S. citizens, leading to many advancements inside China. Besides generative AI, China has made significant strides in AI payment systems and facial recognition technology.
Chinese companies and individuals have filed more AI patents than U.S. entities. However, some argue that the quality of U.S. patents is higher, citing as evidence their higher number of citations. Also, while Apple Pay, Google Pay and other forms of AI payments have steadily gained traction in the U.S., China's WeChat Pay claims nearly 10 times the number of users as Apple Pay -- more than 900 million.
A fight over facial recognition technology has played a role in the U.S.-China AI race. It has been widely reported that China is using facial recognition technology to monitor citizens of Uyghur ethnicity in the Xinjiang region. In response, the U.S. sanctioned Chinese facial recognition technology vendor SenseTime and several others by restricting U.S. exports to the companies and banning U.S. investment in them.
China and some other Asian countries don't perceive facial recognition and monitoring technology as invasive in public areas.
"It is seen as contributing to the safety of public spaces," Kompella said.
While some U.S. states have banned facial recognition technology, China's top facial recognition vendors have access to the Chinese government's database of photos of its citizens.
Trump's approach to China
Trump has already slapped an additional 10% tariff on Chinese goods and explored limiting Chinese investment in U.S. technology, companies and land. During his first administration, Trump imposed tariffs on Chinese goods, which the Biden administration kept in place.
Export controls on advanced technologies consider issues such as national security and economic competitiveness. Dartmouth's Lind said such restrictions are considered reasonable policy against military rivals.
However, export controls also incentivize targeted countries to find ways to overcome scarcity -- in this case, scarcity of chips, she said.
"If the target is highly motivated with high levels of human capital, as China is, the process of overcoming scarcity can lead to innovation," Lind added. "That's what DeepSeek's success shows us."
Indeed, while export controls might protect a country's technological edge, they aren't the sole determinants of leadership in AI, Forrester's Dai said.
"DeepSeek demonstrates that innovation in algorithms and engineering practices will fill in the gaps," Dai said. "Collaboration, open research and the free exchange of ideas have always been and will always be key drivers of technological progress."
Rather than punitive measures, reciprocity and expanding market access in tech should be a key focus of the Trump administration in the U.S.-China AI competition, ITIF's Castro said.
Many U.S. companies, including OpenAI and Meta, can't make their AI services available in China, while Chinese companies, including DeepSeek, are allowed to operate in the U.S. -- although U.S. government officials are seeking to ban DeepSeek on government devices.
With China gearing up to be a lead adopter of AI, lack of access to the Chinese market could hurt the dynamism of U.S. companies, Castro said.
"This is where it could be good for the U.S. if the Trump administration is able to negotiate access for U.S. companies to the Chinese market, and these companies are able to compete there and thrive," he said.
Fears about Chinese nationals in the U.S.
The China-U.S. tech rivalry could affect Chinese students and workers in the U.S.
Jeffrey Stoff, founder of the Center for Research Security and Integrity, testified during the January Senate Foreign Relations Committee hearing about the Chinese government's unfettered access to AI research through Chinese nationals.
The Hoover Institution, a Stanford University public policy think tank, also highlighted this concern, noting that Chinese students fear the Chinese Communist Party's interference.
Idaho's Risch said the CCP's access to university research and work on designing computer chips, building manufacturing facilities and developing AI raises significant national security concerns.
"The Chinese students here are not studying ancient Greek history," Risch said. "They're here with the STEMs," he said, referring to science and technology-based curricula.
"Each and every one of them, whether they like it or not, is an agent of the Chinese Communist Party," Risch continued. "When they go back, we all know they get debriefed, and any information they've garnered here in the U.S. becomes the property of the CCP."
The concern is that once Chinese nationals return to China, they fall outside the purview of U.S. laws, including intellectual property laws. Still, ITIF's Castro said any measures advanced by Congress and the Trump administration would have to walk a fine line and stay focused on the CCP.
"There are very legitimate concerns," Castro said. "But at the same time, you have to make sure those concerns don't become anti-Chinese sentiment, which is very different than being anti-CCP."
The Optional Practical Training program allows foreign students to work in the U.S. on an F-1 student visa for up to three years should they graduate with a STEM degree. Once the student visa expires, foreign students would need an H-1B visa to continue working in the U.S. Roughly 12% of H-1B visas are awarded to Chinese nationals, according to a U.S. Citizenship and Immigration Services annual report.
China has denied and criticized the U.S. for such accusations. In a January 2024 press conference, Chinese Foreign Ministry spokesperson Wang Wenbin said the U.S. "has long been wantonly suppressing and ill-treating Chinese students traveling to the U.S. with lawful and valid identifications and visas." Wenbin said the U.S. "weaponized academic research" to deport Chinese students and asked that the U.S. "stop bullying and restricting Chinese students in the name of national security."
And U.S. academics would likely resist U.S. attempts to crack down on Chinese nationals' access to university research, Lind said.
"Students and researchers from China are real people who are valued colleagues and friends -- not abstract geopolitical ideas," she said.
An unknown winner
The question of whether China or the U.S. is ahead in AI is not settled. Despite the advances in China, it's clear that the contributions of big U.S. AI vendors like OpenAI and Nvidia have transformed the worldwide AI landscape.
"It's a race, and OpenAI is the clear consumer winner right now, and Nvidia is, of course, a winner from an economic standpoint," said Randall Hunt, CTO of Caylent, a cloud services vendor that works with AWS. "The chips that they're selling are the preferred method for building these models right now."
Nvidia could gain even more control over the worldwide AI chip market it dominates because DeepSeek uses AI hardware GPUs for its R1 LLM. The fact that DeepSeek achieved what it did with a limited number of Nvidia GPUs shows just how valuable AI hardware is to the advancement of AI, Hunt said.
"Their GPUs suddenly became more efficient and even better," he said.
While Nvidia's GPUs are powerful, Chinese vendor Huawei's Ascend 910C chips could be another win for China if they can perform the same job as Nvidia's GPUs. The chips have high computation power, which makes them suitable for AI model training and inferencing. They could be a strong competitor to Nvidia's dominant A100 and H100 GPUs. DeepSeek said it used Ascend 910C GPUs to inference its reasoning model.
"If having the world's best LLM is seen as a matter of national pride, then expect the AI race to heat up," RPA2AI's Kompella said.
However, Huawei faces problems in the U.S. and other countries, which might deter others from using its technology. The Chinese tech giant has been accused of threatening national security and using its 5G telecommunications technology to spy. This has led the U.S., Australia and Japan to ban the vendor from building its 5G network in their countries.
Still, Huawei's 2024 revenue exceeded expectations with all these challenges, showing it can survive despite the circumstances. Its resistance is yet another example of the volatile nature of the U.S.-China AI race, and the winner has yet to be determined.
"The competition is fierce, and investment is extremely heavy," ScaleFlux's Baker said. "The investment levels are tremendous in the U.S. and China."
Makenzie Holland is a senior news writer covering big tech and federal regulation. Prior to joining Informa TechTarget, she was a general assignment reporter for the Wilmington StarNews and a crime and education reporter at the Wabash Plain Dealer.
Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.