(Bloomberg) — Deepseek, a Chinese AI startup just over a year old, has sparked fear and consternation in Silicon Valley after demonstrating breakthrough artificial intelligence models that offer performance comparable to the world’s best chatbots at a seemingly fraction of the cost.
Most read from Bloomberg
The emergence of Deepseek could offer a counterpoint to the widespread belief that the future of AI will require ever-increasing amounts of power and energy to develop.
Global tech stocks fell as hype around Deepseek’s innovation snowballed and investors began to digest the implications for its U.S.-based rivals and their hardware suppliers.
What exactly is Deepseek?
Deepseek was founded in 2023 by Liang Wenfeng, head of AI-led hedge fund High-Flyer. The company develops AI models that are open-source, meaning the developer community can inspect and improve the software. Its mobile app rose to the top of the iPhone download charts in the United States after its release in early January.
The app sets itself apart from other chatbots like Openai’s Chatgpt by articulating its reasoning before providing a response to a prompt. The company says its R1 version offers performance on par with the latest OpenAI and has licensed it for those interested in developing chatbots using the technology to take advantage of it.
How does Deepseek R1 compare to Openai or Meta Ai?
Although not fully detailed by the company, the cost of training and developing Deepseek’s models appears to be only a fraction of what is required for top products from Openai or Meta Platforms Inc. The model’s much better efficiency calls into question the need for vast capital expenditures to acquire Nvidia Corp’s newer and more powerful AI accelerators. intended to prevent a breakthrough of the type that Deepseek appears to represent.
Deepseek claims that R1 is close to or better than rival models in several leading benchmarks such as AIME 2024 for math tasks, MMLU for general knowledge, and alpacaeval 2.0 for question-answering performance. He also ranks among the top performers on a UC Berkeley-affiliated leaderboard called Chatbot Arena.
Suqian, China – January 27, 2025 – An illustrative photo shows the logo of Deepseek and Chatgpt in Suqian, Jiangsu province, China, January 27, 2025. (Photo credit should read Cfoto/Future Publishing via Getty Images) ·Cfoto via getty images
What is causing alarm in the United States?
Washington has banned the export of high-end technologies like GPU semiconductors to China, aiming to block the country’s advances in AI, the key frontier in the U.S.-China contest for technological supremacy. But Deepseek’s progress suggests that Chinese AI engineers have worked their way around the restrictions, focusing on greater efficiency with limited resources. While it’s unclear how much of the advanced AI training material Deepseek has had access to, the company has demonstrated enough to suggest that trade restrictions have not been entirely effective in thwarting China’s progress.
When did Deepseek gain global interest?
The AI developer has been under scrutiny since releasing its first model in 2023. Then in November, it gave the world a glimpse of its R1 Deepseek reasoning model, designed to mimic human thinking. This model underpins its mobile chatbot app, which along with the web interface in January made its way to global fame as a much cheaper open alternative, investor Marc Andreessen calls it “the moment Ai’s Sputnik.
Deepseek mobile app was downloaded 1.6 million times before January 25 and ranked #1 in iPhone app stores in Australia, Canada, China, Singapore, USA and UK United, according to market tracker app figures data.
Who is the founder of Deepseek?
Born in Guangdong in 1985, Liang received bachelor’s and master’s degrees in electronic and information engineering from Zhejiang University. He founded Deepseek with 10 million yuan ($1.4 million) in registered capital, according to the company’s Tianyancha database.
The bottleneck for further advances is not more fundraising, Liang said in an interview with Chinese Outlet 36KR, but U.S. restrictions on access to the best chips. Most of its top researchers were new graduates from China’s top universities, he said, highlighting the need for China to develop its own domestic ecosystem similar to the one built around Nvidia and its AI chips.
“More investment does not necessarily lead to more innovation. Otherwise, big companies would take over all the innovation,” Liang said.
Where does Deepseek fit into China’s AI landscape?
Chinese technology leaders, from Group Alibaba (BABA) Holding Ltd. and Baidu (Bidu) Inc. to Tencent (0700.HK) Holdings Ltd., have poured significant money and resources into the race to acquire hardware and customers for their AI companies. Alongside Kai-Fu Lee’s startup 01.ai, Deepseek stands out with its open source approach – designed to quickly recruit the largest number of users before developing monetization strategies on top of this large audience.
Because Deepseek’s models are more affordable, it has already played a role in helping lower costs for AI developers in China, where the biggest players have engaged in a price war that has seen waves successive price drops over the past year and a half.
What are the implications for the global AI market?
Deepseek’s success may push Openai and other US suppliers to reduce their price to maintain their established lead. He also questions the vast spending by companies like Meta and Microsoft Corp. – each committed to CAPEX of $65 billion or more this year, largely on AI infrastructure – so more efficient models can compete with a much smaller outlay.
This disrupted global stock markets as investors sold companies like Nvidia Corp. and ASML holding NV which have benefited from the booming demand for AI services. Shares of Chinese names linked to Deepseek, such as Iflytek Co., climbed.
Already, developers around the world are experimenting with Deepseek’s software and looking to create tools with it. This could accelerate the adoption of advanced AI reasoning models – while potentially addressing an additional concern around the need for guardrails around their use. Deepseek’s advances can accelerate regulation to control how AI is developed.
The Deepseek logo and the Chinese flag are seen in this illustration taken on January 27, 2025. Reuters / Dado Ruvic / Illustration ·Reuters / Reuters
What are the shortcomings of Deepseek?
Like all other Chinese AI models, Deepseek autocensors on topics deemed sensitive in China. He deflects questions about the 1989 Tiananmen Square protests or geopolitically fraught issues such as the possibility of China invading Taiwan. In tests, the Deepseek bot is able to give detailed answers about political figures like Indian Prime Minister Narendra Modi, but refuses to do so for Chinese President Xi Jinping.
Deepseek’s cloud infrastructure is likely to be tested by its sudden popularity. The company briefly experienced a major outage on January 27 and will have to handle even more traffic as new users and users return more queries to its chatbot.
– with the help of Luz Ding, Zheping Huang, Claire Che, Ville Heiskanen and Mayumi Negishi.
(Adds additional market context in 17th paragraph)
