DeepSeek, a Chinese AI startup that’s just over a year old, has stirred both awe and anxiety in Silicon Valley by showcasing advanced AI models that rival the world’s best chatbots at significantly lower costs.
The rise of DeepSeek challenges the widely held belief that developing the future of AI requires ever-increasing power and energy resources.
By the end of January, global tech stocks saw a decline as the buzz around DeepSeek’s innovations grew, prompting investors to assess its implications for U.S.-based competitors and their hardware suppliers.
What is DeepSeek?
Founded in 2023 by Liang Wenfeng, a former head of an AI-powered quant hedge fund, DeepSeek develops open-source AI models. This means developers worldwide can inspect and enhance its software. After launching its mobile app in early January, it quickly topped the iPhone download charts in the U.S.
What sets this app apart from other chatbots like OpenAI’s ChatGPT is its ability to explain its reasoning before responding to prompts. DeepSeek claims that its R1 model matches the performance of OpenAI’s latest version and has already begun licensing its technology to developers interested in building chatbots.
How does DeepSeek R1 compare to OpenAI or Meta AI?
While detailed information remains limited, training and developing DeepSeek’s models reportedly cost only a fraction of what it takes to develop similar products by OpenAI or Meta Platforms. This efficiency questions the necessity of massive capital expenditures on cutting-edge AI accelerators, like those provided by Nvidia Corp.
Moreover, it highlights the impact of U.S. export restrictions on advanced semiconductors to China, which were meant to curb breakthroughs like DeepSeek’s. Despite these restrictions, DeepSeek’s success demonstrates that Chinese AI engineers have innovated around these limitations, focusing on achieving more with fewer resources.
DeepSeek asserts that its R1 model performs at par with or exceeds other top models on benchmarks such as AIME 2024 for mathematical tasks, MMLU for general knowledge, and AlpacaEval 2.0 for Q&A performance. It also ranks among the top performers on UC Berkeley’s Chatbot Arena leaderboard.
Overcoming Challenges
The U.S. has banned the export of advanced technologies like GPU semiconductors to China, aiming to slow the nation’s progress in AI—a key battleground in the U.S.-China tech race. However, DeepSeek’s achievements suggest that trade restrictions haven’t been entirely effective, as the company has prioritized efficiency over brute computational power. Although the extent of DeepSeek’s access to high-end AI training hardware is unclear, its accomplishments prove that China remains a formidable player in AI innovation.
When did DeepSeek gain global attention?
Since the release of its initial models in 2023, DeepSeek has been closely watched by AI developers. In November, it unveiled its DeepSeek R1 reasoning model, designed to mimic human thought processes. This model powers its mobile chatbot app, which, along with a web interface launched in January, has gained global recognition as a much cheaper alternative to OpenAI.
Investor Marc Andreessen even referred to DeepSeek as the “Sputnik moment” for AI.
According to market tracker App Figures, the DeepSeek mobile app was downloaded 1.6 million times by January 25 and ranked number one on the iPhone App Store in countries like the U.S., UK, Canada, Australia, China, and Singapore.
Who is the founder of DeepSeek?
DeepSeek was founded by Liang Wenfeng, born in 1985 in Guangdong, China. Liang earned his bachelor’s and master’s degrees in electronic and information engineering from Zhejiang University. According to the database Tianyancha, he established DeepSeek with a registered capital of 10 million yuan ($1.4 million).
In an interview with Chinese outlet 36kr, Liang said that the main challenge for further progress isn’t raising more funds but the U.S. restrictions on accessing advanced chips. Most of DeepSeek’s top researchers are fresh graduates from China’s leading universities. Liang emphasised the need for China to develop its own ecosystem around advanced technologies like Nvidia’s AI chips. He stated, “More investment doesn’t necessarily lead to more innovation. Otherwise, large companies would monopolise all innovation.”
Where does DeepSeek stand in China’s AI landscape?
While major Chinese tech giants like Alibaba, Baidu, and Tencent have poured significant resources into hardware and customer acquisition for their AI ventures, DeepSeek has taken a different path. Its open-source approach is designed to quickly attract a large user base and later build monetisation strategies around it.
DeepSeek’s cost-efficient models have already helped lower expenses for AI developers in China, where big players have been engaged in price wars, resulting in repeated cost reductions over the past 18 months.
What does DeepSeek’s success mean for the global AI market?
DeepSeek’s rise could pressure OpenAI and other U.S. providers to reduce their prices to maintain their competitive edge. It also raises questions about the massive spending by companies like Meta and Microsoft, which have committed over $65 billion this year, primarily for AI infrastructure. If more efficient models can compete with significantly lower investment, the spending strategies of these tech giants may come under scrutiny.
DeepSeek has also impacted global stock markets. As investors shifted focus, shares of companies like Nvidia and ASML Holding NV, which benefit from AI-driven demand, saw declines. Meanwhile, stocks of Chinese companies linked to DeepSeek, such as Iflytek Co., experienced gains.
Developers worldwide are already experimenting with DeepSeek’s software, exploring ways to build tools with it. This could accelerate the adoption of advanced AI reasoning models but might also raise concerns about the need for stricter regulations around their use.
What are DeepSeek’s limitations?
Like other Chinese AI models, DeepSeek self-censors on topics deemed sensitive in China. For instance, it avoids questions about events like the 1989 Tiananmen Square protests or geopolitical issues such as the possibility of China invading Taiwan. While DeepSeek’s chatbot provides detailed responses about global political figures like Indian Prime Minister Narendra Modi, it avoids giving similar information about Chinese President Xi Jinping.
Additionally, the sudden surge in popularity of DeepSeek’s cloud infrastructure has tested its limits. On January 27, the company experienced a significant outage and will need to manage even more traffic as new and returning users continue to interact with its chatbot.
