Chinese AI startup DeepSeek is focusing on research rather than revenue, taking advantage of a sudden surge in sales by billionaire founder Liang Wenfeng, who has decided not to follow the example of his Silicon Valley rivals.
The Hangzhou-based company has had to adapt to a surge in demand for its free consumer services on its website and app, as well as paid services from business users. Revenue last month was enough to cover operating expenses for the first time, the Financial Times reported, citing two people familiar with the matter. Interest in DeepSeek grew in January, when the company released a low-cost reasoning AI model called R1, which performs on par with leading US and Chinese models but cost significantly less to develop.
Corporate clients from the medical and financial industries have been buying DeepSeek’s API access to develop their own applications based on the V3 and R1 models, and the demand has been so high that the startup has had to temporarily suspend customer registrations due to a lack of resources for non-research purposes. Mr. Liang has shown little intention of immediately making money from DeepSeek. Instead, the company has focused most of its resources on developing new models and creating strong artificial intelligence (AGI) with cognitive abilities on par with humans. The head of the independent company has also so far refused to cooperate with Chinese tech giants, venture capital funds, and government funds that have expressed a desire to invest in DeepSeek. On the contrary, it has been difficult to even schedule a meeting with the reclusive businessman.
DeepSeek’s arrival has upset investors, who have questioned whether American tech leaders Google and OpenAI can maintain their advantage, and whether the tech giants’ massive spending on AI infrastructure is justified. The Chinese company’s approach is significantly different from that of many Silicon Valley startups: OpenAI has used its status as an AI leader to build an ecosystem of commercial services around ChatGPT and begin to generate significant revenue from access to its AI models via APIs. The American company has raised about $20 billion in funding since 2019, and is currently in talks with a group of investors led by SoftBank to raise another $40 billion at a valuation of $260 billion. OpenAI employs just over 2,000 people, while DeepSeek has about 160.
DeepSeek’s lack of commercial aspirations has played into the hands of Chinese tech giants Alibaba and Tencent, whose established infrastructure and service packages have attracted large numbers of corporate clients in their home country, raising doubts about the sustainability of the startup’s revenue streams. Apple, for example, chose Alibaba’s Qwen model over DeepSeek to deploy AI features on iPhones in China; Tencent has boosted sales by opening up DeepSeek’s open-source models to its cloud infrastructure, with about half of the tech giant’s cloud clients choosing them, and about 20% requesting improvements. Because DeepSeek has refused to make the effort to market its own products to the mass market, Tencent has integrated the startup’s models into its popular consumer apps.
In recent years, Liang Wenfeng, who also runs a hedge fund, purchased about 10,000 Nvidia A100 AI accelerators and a similar number of H800 models before they were banned from entering China. The company plans to buy hardware from other suppliers in the long term, and has received support from the Chinese government, which has given it access to government-funded data centers to help meet some of its computing needs. In the long term, DeepSeek may find its lack of access to Nvidia’s cutting-edge hardware a problem and explore other partners. And to gain additional political support, industry experts say the company may have to open itself up to government funding. The company’s engineers are now hard at work developing the R2 and V4 models, which were originally scheduled for release in May but are now moving faster.