DeepSeek has recently gained widespread attention.
This Chinese AI laboratory has entered the spotlight after its chatbot application reached the top position in the Apple App Store, as well as on Google Play. DeepSeek’s AI models, developed with compute-efficient techniques, have prompted analysis from Wall Street experts and technology specialists regarding whether the U.S. can maintain its leadership in the AI industry and if the demand for AI chips will remain steady.
The rapid rise of DeepSeek to global recognition raises questions about its origins and journey to fame.
DeepSeek’s Background in Trading
DeepSeek is supported by High-Flyer Capital Management, a Chinese quantitative hedge fund that leverages AI for trading decisions. AI enthusiast Liang Wenfeng co-founded High-Flyer in 2015 and launched the hedge fund in 2019, concentrating on AI algorithms. In 2023, High-Flyer initiated DeepSeek as a separate entity focused on AI tool research. DeepSeek built its own data centers for model training and faced challenges due to U.S. export bans, leading to the use of less powerful Nvidia H800 chips.
DeepSeek’s technical team is notably young and reportedly recruits PhD AI researchers aggressively from leading Chinese universities. Additionally, it employs individuals without computer science backgrounds to enhance its technology’s understanding of varied subjects.
DeepSeek’s Advanced Models
Initially unveiling models such as DeepSeek Coder, DeepSeek LLM, and DeepSeek Chat in November 2023, DeepSeek captured industry attention with the subsequent release of DeepSeek-V2 series. This general-purpose system, effective in text and image analysis, performed proficiently in AI benchmarks and was more cost-efficient compared to other models. It prompted competitors like ByteDance and Alibaba to reduce prices or offer models for free.
The introduction of DeepSeek-V3 in December 2024 further increased DeepSeek’s prominence, outperforming notable models like Meta’s Llama and OpenAI’s GPT-4o, according to internal benchmarks. DeepSeek’s R1 "reasoning" model, released in January, claims parity with OpenAI’s o1 model on key benchmarks by efficiently self-fact-checking, beneficial in fields such as physics, science, and math.
Despite their strengths, DeepSeek’s AI models adhere to regulations by China’s internet regulator, influencing their responses on topics like Tiananmen Square or Taiwan’s autonomy.
Innovative Business Approach
While DeepSeek’s exact business model remains unclear, its strategy includes pricing products below market rates and offering free services. Despite significant interest, the company has not sought investor funding. Efficiency improvements reportedly maintain its competitive costs, though some experts question the provided figures. Developers have widely adopted DeepSeek’s models under licenses allowing commercial use. Hugging Face’s platform, hosting these models, has facilitated the creation of over 500 derivative models, totaling 2.5 million downloads.
DeepSeek’s impact has been described as transformative yet over-hyped, causing a notable drop in Nvidia’s stock in January and prompting responses from industry leaders like OpenAI CEO Sam Altman. Reports suggest that the U.S. government intends to ban DeepSeek on official devices, primarily due to national security concerns.
Microsoft announced the availability of DeepSeek on its Azure AI Foundry service. Despite this, some companies and countries, including South Korea, have imposed bans due to data risk concerns. New York state also prohibited the use of DeepSeek on government devices.
The future of DeepSeek remains uncertain, with expectations of improved models. However, the U.S. government is increasingly cautious about foreign influence and is likely to ban DeepSeek on governmental devices, according to reports.
This report, first published on January 28, 2025, will continue to be updated.
