How Tech Start-Up DeepSeek Emerged as an Unlikely Game Changer in the US-China AI War

The DeepSeek logo displayed on a phone screen reflects a significant innovation in the tech industry. Recently, DeepSeek shocked the global tech market, especially during the period leading up to the Lunar New Year. The Chinese start-up's advancements led to a noticeable drop in the stock value of major companies like the semiconductor giant Nvidia, impacting many players in the booming artificial intelligence (AI) sector.

DeepSeek, based in Hangzhou, has recently rolled out two robust AI models known as DeepSeek-V3 and DeepSeek-R1. These models were developed with considerably less financial expenditure and computational resources compared to those utilized by larger tech companies for creating large language models (LLMs)—the foundation for generative AI technologies such as ChatGPT.

Social media discussions among AI professionals highlighted the exceptional performance of DeepSeek's open-source models. Industry benchmarks indicated that these models either matched or even surpassed those of competitors, despite stringent U.S. regulations limiting China's access to cutting-edge semiconductors and technology. Notably, Sam Altman, the CEO of OpenAI, praised the R1 model on X, remarking on its impressive capabilities for the cost.

The excitement around DeepSeek peaked recently when it was reported that its chatbot, powered by the R1 model, became the leading free app on both the U.S. and Chinese versions of the Apple App Store.

A Visionary Leader

Liang Wenfeng, the founder and CEO of DeepSeek, is typically discreet about his public engagements. However, in the past couple of years, he has shared insights through a couple of interviews with Chinese tech media, discussing his aspirations for LLM development, the future of AI, and his personal ambitions.

When Liang established DeepSeek in May 2023, his focus was on researching artificial general intelligence (AGI), aiming to democratize the advancements in this field. He believed that the technology should not be monopolized by just a select few individuals or corporations. Liang prioritized recruiting team members who shared his vision, emphasizing a culture of curiosity and creativity that fuels innovation in AI.

Team of Young Geniuses

The team at DeepSeek primarily consists of gifted young scientists from leading universities such as Tsinghua University and Peking University. Although the firm tends to hire fresh graduates, they also seek out PhD candidates and emerging AI professionals with some work experience.

The technical documentation for the R1 model reveals that around 200 researchers contributed to its development, with 18 of them being core contributors. This collaborative effort highlights the talent and dedication present within DeepSeek's workforce.

Funding and Resources

Despite having a registered capital of only 10 million yuan, DeepSeek benefits significantly from resources provided by High-Flyer Quant, a hedge fund company founded by Liang. In July of last year, Liang indicated that financing was less of a concern than the restrictions imposed by the U.S. on advanced chip exports to China. It’s important to note that High-Flyer Quant has amassed over 10,000 Nvidia graphics processing units which are crucial for DeepSeek's operations.

A Startup Mindset

Liang explained that DeepSeek operates without rigid hierarchies and allows for flexible deployment of computing resources and talent recruitment. Each employee is encouraged to follow their individual paths, with a natural division of labor guiding the company's work processes. This structure enables the company to quickly respond to technological advancements and pursue unique innovations in the AI field.

DeepSeek, AI, Innovation