On January 20, the little-known Chinese AI research lab DeepSeek made waves by releasing an open-source AI model. According to their research paper, DeepSeek-R1 outperforms leading models like OpenAI’s o1 in several math and reasoning benchmarks. The model also excels in key areas such as capability, cost and openness, challenging Western AI giants.

A Surprising Outcome of the Tech Cold War

DeepSeek’s success highlights an unintended result of the ongoing tech rivalry between the US and China. US export controls have limited Chinese companies from scaling up traditionally; buying more chips and extending training times. Most Chinese firms have shifted focus to downstream applications, avoiding direct competition with Western leaders. However, DeepSeek’s strategy proves there’s another way to win: revamping AI model structures to use resources more efficiently.

A Focus on Software Optimisation

“Unlike many Chinese AI firms that rely heavily on advanced hardware, DeepSeek focuses on software-driven resource optimisation,” says Marina Zhang, an associate professor at the University of Technology Sydney.

“By embracing open-source methods, they pool collective expertise and foster collaborative innovation. This approach overcomes resource constraints and accelerates technological development, setting DeepSeek apart from its competitors.”

DeepSeek’s Roots in Finance

DeepSeek began as Fire-Flyer, a research branch of High-Flyer, a leading Chinese quantitative hedge fund. High-Flyer, founded in 2015, quickly gained prominence, managing over 100 billion RMB ($15 billion) at its peak. By 2023, Liang Wenfeng, a computer science master’s graduate, redirected the hedge fund’s resources into DeepSeek. His bold vision was to build cutting-edge models and work toward artificial general intelligence.

“DeepSeek represents a new generation of Chinese tech companies that prioritise long-term innovation over quick profits,” says Zhang.

Read more: UNESCO Dedicates World Education Day to Artificial Intelligence

Driven by Curiosity, Not Profit

Liang told 36Kr, a Chinese tech publication, that scientific curiosity motivated him to start DeepSeek. “Basic science research has a very low return-on-investment ratio,” he said. “When OpenAI’s early investors gave it money, they weren’t thinking about profits. They simply wanted to pursue the idea.”

Unlike other Chinese AI firms, DeepSeek operates without funding from giants like Baidu, Alibaba, or ByteDance.

A Team of Young Innovators

Liang’s hiring strategy focused on PhD students from top Chinese universities, such as Peking University and Tsinghua University. Many had published in prestigious journals and won international awards but lacked industry experience.

“Our core technical positions are filled by recent graduates,” Liang told 36Kr. This created a collaborative environment where researchers had access to ample resources for unorthodox projects. The contrast with other firms, where teams compete for resources, is stark.

“Most young people can commit fully to a mission without utilitarian concerns,” Liang explained. His pitch to recruits was simple: “Solve the hardest questions in the world.”

Necessary Innovation

US export controls on advanced chips like Nvidia’s H100 posed a challenge for DeepSeek. Although the company began with a stockpile of 10,000 H100s, they needed more to stay competitive.

DeepSeek adapted by finding more efficient ways to train models. “They used engineering tricks, custom chip communication schemes and memory-saving techniques,” explains Wendy Chang, a policy analyst at the Mercator Institute. DeepSeek also excelled in Multi-head Latent Attention (MLA) and Mixture-of-Experts designs, creating cost-effective models requiring less computing power. Their latest model, for instance, used one-tenth the resources of Meta’s Llama 3.1.

DeepSeek has emerged as a bold innovator, proving that resource constraints can inspire groundbreaking approaches. By focusing on efficiency, collaboration, and open-source methods, it has reshaped global AI dynamics. The rise of DeepSeek signals a new era where innovation is driven not just by resources but by creativity and determination.

Stay tuned to Brandsynario for the latest news and updates.