The Chinese DeepSeek AI model has shocked the world with its rapid ascent to top levels of performance at a fraction of the cost, in what some have called a āSputnik momentā. DeepSeek approaches OpenAIās ChatGPT performance at a fraction of the cost using more efficient coding and simpler computing hardware. DeepSeekās origins give a glimpse of a Chinese style of innovation that arises from decades of building up its education system and is one of the many results from a strategic focus on STEM education and entrepreneurship in China. DeepSeek should be a wake-up call to the US, since as China is building up its universities, higher education and research funding in the US is under attack.
The Education of Liang Wenfeng, DeepSeekās Founder
DeepSeekās founder, Liang Wenfeng, was the son of primary school teacher, and as a precocious child in Guangdong, in Southern Child, quickly mastered advanced math concepts in junior high school, and received bachelorās and masterās degrees in electronic information engineering and communication engineering from Zhejiang University, a mid-sized Chinese university ranked in the top 100 globally. After graduation, he launched a new company named High-Flyer that employed a machine learning model for tracking investments. Using some of the resources from High-Flyer, which by 2019 managed over $10 billion in assets, Liang developed a cluster of 10,000 of the Nvidia A100 GPU chips in 2022 before U.S. export restrictions took effect. Liang created DeepSeek in 2023 to work much like a Chinese version of Google, where a creative team of recent graduates from Chinaās top universities such as Peking University, Tsinghua University, and Zhejiang University could follow their curiosity in developing advanced AI technologies. Liang sought not only brilliant graduates in mathematics and STEM fields, many who won awards in international conferences, but also some poets and humanities graduates.
Building an AI with Chinese Characteristics
DeepSeekās algorithms for predicting stocks eventually grew into standalone AI models. Deepseekās highest performance model named R1 has been the top downloaded app on Apple and Google stores for over a week, with 16 million downloads in just the first 18 days. DeepSeekās V3 model was developed with 2,000 H800 chips from Nvidia, which outperforms OpenAIās GPT-4o in some tests, despite OpenAIās use of over 10,000 of the most advanced H100 Nvidia chips. The DeepSeek model gains efficiency from a machine-learning based training and reduced use of memory, and since it has been provided as an open-source code, will help other companies and countries develop their own new LLMās. The success of DeepSeek is also an example of a homegrown Chinese innovation center, where creative talent can spontaneously provide breakthroughs. Liang described the need to overcome a limitation of Chinese tech companies, which āisnāt capital but confidence and the ability to organize high-caliber talent for effective innovation.ā
Chinaās Silicon Valleys
DeepSeek is based in Hangzhou, home to Chinese tech companies such as Alibaba, and is just one of many emerging tech hubs in China. Liang himself grew up in Guangdong province, which is part of the Greater Bay Area, which has emerged as one of several Chinese regional concentrations of tech companies and talent. Guangdong province includes nine rapidly growing cities, including Shenzhen, Hong Kong and Macao, and accounts for 86 million people and a gross domestic product of $1.94 trillion. The Greater Bay Area aspires to be the Chinese version of Silicon Valley, combining its density of innovative tech companies with an unmatched supplier and manufacturing capacity for rapidly converting ideas into hardware. The region is also an educational hub, since Hong Kong is home to five universities in the global top 100 QS world ranking, nearly half of Chinaās total of top-ranked universities. In Shenzhen alone, the city government has committed to invest $23 billion to build 20 new universities. New university campuses within the Greater Bay Area include the Shenzhen Technology University, completed in 2017 and enrolling over 12,000 students, the Southern University of Science and Technology (SUSTech), founded in 2010, and already ranked among the top 200 of global universities, and a branch of the Hong Kong University of Science and Technology, developed jointly with Peking University. Additional new campuses in Shenzhen under construction include the Shenzhen Normal University, a Shenzhen Conservatory of Music, an Institute for Creative Design, and the Shenzhen Ocean University.
Chinaās Massive Scaling of Higher Education Bears Fruit
Chinaās massive scaling of its university system across the entire country in past decades has produced nearly 2000 new universities since 1997, as the total number of colleges and universities in China increased from 1,020 in 1997 to 2,822 public institutions in 2023. These new institutions have enabled Chinese university enrollment to grow by more than a factor of ten since 1997, with China producing over 12 million graduates in 2024. These new graduates are fueling a surging growth in tech companies from the increased emphasis on STEM subjects in Chinese universities. Data from 2020 shows that a third of Chinese undergraduates study engineering compared to 8% in the US, and China produced over twice the number of science and engineering graduates than the US.
Chinaās education system is not rapidly increasing in scale but also rising in quality to compete globally. Chinese universities included 13 of the top 200 in the THE global rankings in 2023, with Tsinghua University and Peking University ranking 12th and 14th place respectively. Since 2017, when China declared its goal to be the world leader in AI by 2030, AI has been a strategic priority for China, with 440 universities offering degrees in AI, and China has produced almost half of the worldās leading AI researchers, compared to 18% from the US.
Multiple Hubs of AI Talent in China
Where will this meteoric rise of tech talent take China and AI? Chinaās goal to become the world leader in AI by 2030 could be realized, despite US export controls, as it seems to be rooted in a broader strategy to develop its own base of creative technology talent. The massive government investment in new universities is already producing nearly twice as many PhDās in STEM fields than the US, with 77,000 STEM PhD graduates predicted for 2025 compared to 40,000 in the US. Chinaās focus on AI includes developing a $2.1 billion AI industrial park and creating multiple hubs for AI talent in the country. A report analyzing AI job openings in China shows three major AI technology hubs centered on the cities of Beijing, Shanghai and Shenzhen, creating the possibility of multiple Chinese āSilicon Valleys.ā For the US to maintain its lead and edge in AI and other technology fields, a similar commitment and investment in higher education as will be needed. DeepSeek provides a wake-up moment for the US if it is to retain leadership in STEM fields, or as Liang said in a recent interview, āWeāre done following. Itās time to lead.ā