(Bloomberg) — DeepSeek, a Chinese artificial-intelligence startup that’s just over a yr old, has stirred awe and consternation in Silicon Valley after demonstrating AI models that provide comparable performance to the world’s best chatbots at seemingly a fraction of their development cost.
DeepSeek’s emergence may offer a counterpoint to the widespread belief that the long run of AI would require ever-increasing amounts of computing power and energy.
Global technology stocks tumbled on Jan. 27 as hype around DeepSeek’s innovation snowballed and investors began to digest the implications for its US-based rivals and AI hardware suppliers reminiscent of Nvidia Corp.
DeepSeek was founded in 2023 by Liang Wenfeng, the chief of AI-driven quant hedge fund High-Flyer. The corporate develops AI models which might be open-source, meaning the developer community at large can inspect and improve the software. Its mobile app surged to the highest of the iPhone download charts within the US after its release in early January.
The app distinguishes itself from other chatbots like OpenAI’s ChatGPT by articulating its reasoning before delivering a response to a prompt. The corporate claims its R1 release offers performance on par with the most recent iteration of ChatGPT. It’s offering licenses for people interested by developing chatbots using the technology to construct on it, at a price well below what OpenAI charges for similar access.
Follow The Big Take every day podcast wherever you listen.
DeepSeek says R1’s performance approaches or improves on that of rival models in several leading benchmarks reminiscent of AIME 2024 for mathematical tasks, MMLU for general knowledge and AlpacaEval 2.0 for question-and-answer performance. It also ranks among the many top performers on a UC Berkeley-affiliated leaderboard called Chatbot Arena.
Though not fully detailed by the corporate, the fee of coaching and developing DeepSeek’s models appears to be only a fraction of what’s required for OpenAI or Meta Platforms Inc.’s best products. The greater efficiency of the model puts into query the necessity for vast expenditures of capital to amass the most recent and strongest AI accelerators from the likes of Nvidia. It also focuses attention on US export curbs of such advanced semiconductors to China — which were intended to stop a breakthrough of the kind that DeepSeek appears to represent.
When did DeepSeek spark global interest?
The AI developer has been closely watched because the release of its earliest model in 2023. Then in November, it gave the world a glimpse of its DeepSeek R1 reasoning model, designed to mimic human considering. That model underpins its chatbot app, which exploded in popularity as a less expensive OpenAI alternative, with investor Marc Andreessen calling it “AI’s Sputnik moment.”
The DeepSeek mobile app was downloaded 1.6 million times by Jan. 25 and ranked No. 1 in iPhone app stores in Australia, Canada, China, Singapore, the US and the UK, based on data from market tracker App Figures.
What did we learn from the enormous stock market response?
For much of the past two-plus years since ChatGPT kicked off the worldwide AI frenzy, investors have bet that improvements in AI would require ever more advanced chips from the likes of Nvidia.
The DeepSeek breakthrough suggests AI models are emerging that may achieve a comparable performance using less sophisticated chips for a smaller outlay.
Investors offloaded Nvidia stock in response, sending the shares down 17% on Jan. 27 and erasing $589 billion of value from the world’s largest company — a stock market record. Semiconductor machine maker ASML Holding NV and other corporations that also benefited from booming demand for cutting-edge AI hardware also tumbled.
DeepSeek’s success calls into query the vast spending by corporations like Meta and Microsoft Corp. — each of which has committed to capex of $65 billion or more this yr, largely on AI infrastructure.
Shares in Meta and Microsoft also opened lower, though by smaller margins than Nvidia, with investors weighing the potential for substantial savings on the tech giants’ AI investments. Meta even recovered later within the session to shut higher. Chinese names linked to DeepSeek, reminiscent of Iflytek Co., also climbed.
Some industry watchers suggested the industry overall may gain advantage from DeepSeek’s breakthrough if it pushes OpenAI and other US providers to chop their prices, spurring faster adoption of AI.
How could DeepSeek affect the worldwide strategic competition over AI?
AI is the important thing frontier within the US-China contest for tech supremacy. Washington has banned the export to China of kit reminiscent of high-end graphics processing units in a bid to stall the country’s advances.
DeepSeek’s progress suggests Chinese AI engineers have worked their way around those restrictions, specializing in greater efficiency with limited resources. Still, it stays unclear how much advanced AI-training hardware DeepSeek has had access to.
Already, developers all over the world are experimenting with DeepSeek’s software and looking out to construct tools with it. This might help US corporations improve the efficiency of their AI models and quicken the adoption of advanced AI reasoning.
That in turn may force regulators to put down rules on how these models are used, and to what end.
DeepSeek’s progress raises an additional query, one that always arises when a Chinese company makes strides into foreign markets: Could the troves of knowledge the mobile app collects and stores in Chinese servers present a privacy or security threats to US residents?
The indisputable fact that DeepSeek’s models are open-source opens the likelihood that users within the US could take the code and run the models in a way that wouldn’t touch servers in China.
Who’s DeepSeek’s founder?
Born in Guangdong in 1985, engineering graduate Liang has never studied or worked outside of mainland China. He received bachelor’s and masters’ degrees in electronic and knowledge engineering from Zhejiang University. He founded DeepSeek with 10 million yuan ($1.4 million) in registered capital, based on company database Tianyancha.
The bottleneck for further advances will not be more fundraising, Liang said in an interview with Chinese outlet 36kr, but US restrictions on access to the very best chips. Most of his top researchers were fresh graduates from top Chinese universities, he said, stressing the necessity for China to develop its own domestic ecosystem akin to the one built around Nvidia and its AI chips.
“More investment doesn’t necessarily result in more innovation. Otherwise, large corporations would take over all innovation,” Liang said.
Liang has been in comparison with OpenAI founder Sam Altman, however the Chinese citizen keeps a much lower profile and rarely speaks publicly.
Where does DeepSeek stand in China’s AI landscape?
China’s technology leaders, from Alibaba Group Holding Ltd. and Baidu Inc. to Tencent Holdings Ltd., have poured significant money and resources into the race to amass hardware and customers for his or her AI ventures. Alongside Kai-Fu Lee’s 01.AI startup, DeepSeek stands out with its open-source approach — designed to recruit the biggest variety of users quickly before developing monetization strategies atop that giant audience.
Because DeepSeek’s models are cheaper, it’s already played a task in helping drive down costs for AI developers in China, where the larger players have engaged in a price battle that’s seen successive waves of price cuts over the past yr and a half.
What are DeepSeek’s shortcomings?
Like all other Chinese AI models, DeepSeek self-censors on topics deemed sensitive in China. It deflects queries concerning the 1989 Tiananmen Square protests or geopolitically fraught questions reminiscent of the opportunity of China invading Taiwan. In tests, the DeepSeek bot is able to giving detailed responses about political figures like Indian Prime Minister Narendra Modi, but declines to accomplish that about Chinese President Xi Jinping.
DeepSeek’s cloud infrastructure is more likely to be tested by its sudden popularity. The corporate briefly experienced a serious outage on Jan. 27 and can have to administer much more traffic as latest and returning users pour more queries into its chatbot.
–With assistance from Luz Ding, Zheping Huang, Claire Che, Ville Heiskanen, Mayumi Negishi, Olivia Solon and Jan-Patrick Barnert.
(Updates to incorporate more detail on DeepSeek founder. A previous version of this text corrected the reference to Meta’s share price)
Most Read from Bloomberg Businessweek
©2025 Bloomberg L.P.