DeepSeek has surged into global prominence as its AI chatbot app climbed to the top of the Apple App Store and Google Play charts, sparking widespread attention and debate. The Chinese AI lab, backed by High-Flyer Capital Management—a quantitative hedge fund that uses AI for trading—has rapidly gained recognition for its high-performing, cost-efficient models. Founded in 2023 by AI enthusiast Liang Wenfeng, who began his journey in trading while a student at Zhejiang University, DeepSeek started as a research initiative within High-Flyer before becoming an independent company. Despite U.S. export restrictions limiting access to advanced AI chips, DeepSeek built its own data centers and trained models using available hardware, including the Nvidia H800, a less powerful variant of the H100. The company’s technical team is known for being young and highly skilled, with a focus on recruiting top PhD researchers from leading Chinese universities. Notably, DeepSeek also hires individuals from non-technical backgrounds to help improve the model’s understanding of diverse real-world topics. DeepSeek first introduced its core models—DeepSeek Coder, DeepSeek LLM, and DeepSeek Chat—in November 2023. But it was the release of the DeepSeek-V2 series in early 2024 that drew serious industry attention. These models delivered strong performance in benchmarks while operating at significantly lower costs than competitors, prompting rivals like ByteDance and Alibaba to slash prices or offer free access to their models. The launch of DeepSeek-V3 in December 2024 further solidified its reputation. According to internal tests, DeepSeek-V3 outperforms both open-source models like Meta’s Llama and closed models such as OpenAI’s GPT-4o. Equally notable is the DeepSeek-R1 reasoning model, released in January 2025, which rivals OpenAI’s o1 model in key tests. R1 uses self-checking mechanisms to improve accuracy in complex domains like science, math, and physics—though it takes longer to generate responses. However, DeepSeek’s models are subject to China’s internet regulations. The company’s chatbot refuses to answer questions about sensitive topics like Tiananmen Square or Taiwan’s status, aligning with national content policies. By March 2025, DeepSeek had attracted over 16.5 million visits, ranking second in traffic despite a 25% drop from February. While still far behind ChatGPT’s 500 million weekly users, its growth has been remarkable. In May, DeepSeek released an updated version of R1 on Hugging Face, and later unveiled V3.2-exp, a model designed for efficient long-context processing with lower inference costs. DeepSeek’s business model remains unclear. The company offers its models at low prices or for free, avoids seeking venture capital, and does not appear to rely on traditional monetization. It attributes its cost advantage to technical innovations in model efficiency, though some experts question the validity of these claims. Despite the controversy, DeepSeek’s models have gained massive traction among developers. On Hugging Face, over 500 derivative models based on R1 have been created, amassing 2.5 million downloads. The rise of DeepSeek has triggered strong reactions. Nvidia’s stock dropped 18% in January, and OpenAI’s CEO Sam Altman publicly acknowledged the threat. U.S. government agencies, including the Commerce Department, have banned DeepSeek on official devices. South Korea and New York State have followed suit. Microsoft’s Brad Smith confirmed that DeepSeek is prohibited for use by Microsoft employees due to security and propaganda risks. Still, Microsoft has integrated DeepSeek into its Azure AI Foundry platform, signaling cautious adoption. Meanwhile, Meta’s CEO Mark Zuckerberg reiterated that AI infrastructure spending remains a key strategic priority. The U.S. government is reportedly considering a formal ban on DeepSeek, citing national security concerns. As the AI race intensifies, DeepSeek’s success has forced a reevaluation of global AI leadership and the role of state-backed innovation.