Chinese startup DeepSeek has introduced its latest AI models, DeepSeek-V3 and DeepSeek-R1, claiming they rival or surpass leading U.S. models at a significantly lower cost. The company gained attention after revealing that training DeepSeek-V3 required under $6 million in computing power using Nvidia H800 chips. The affordability and efficiency of these models have raised concerns about the rationale behind major U.S. tech firms investing billions in AI. DeepSeek's AI Assistant, powered by V3, has become the top-rated free app on Apple’s U.S. App Store, surpassing ChatGPT. Analysts have pointed out uncertainties in DeepSeek’s total training costs, with the company stating that R1 is up to 50 times cheaper to use than OpenAI's models.
According to Chinese corporate documents, Liang Wenfeng, co-founder of quantitative hedge fund High-Flyer, controls the Hangzhou-based company. The fund shifted its focus to AI research before DeepSeek’s establishment in 2023. Reports suggest High-Flyer has invested in AI hardware, including a cluster of 10,000 A100 chips. The Chinese government has taken note of DeepSeek’s rise, with Liang attending a symposium hosted by Premier Li Qiang. His presence indicates the firm’s potential role in China’s strategy to advance AI despite U.S. export controls.
