"China's State-Led AI Strategy Offers Long-Term Advantage Over the US"

One Year Since DeepSeek's Launch in China: Observations of "Competitive Advantage" Emerge
Streamlined Regulations, Energy Availability, and Raw Materials Highlighted
FT: "State-Led Industrial Strategy Offers Long-Term Benefits"

As of January 20, it has been one year since DeepSeek's artificial intelligence (AI) model 'R1' was introduced in China. Amid this milestone, there are growing observations that China, with its state-led strategic development in the global AI race, holds a long-term competitive advantage over the United States.

On the 20th, it has been one year since DeepSeek's AI model 'R1' was introduced in China. Photo by Reuters-Yonhap News

According to Yonhap News, Tej Parikh, an economist and economic columnist for the Financial Times (FT) in the United Kingdom, wrote on January 18 (local time) in an article titled "China Will Win the AI Race" that "the United States is currently leading in large language models (LLMs), but in the AI competition, which can be likened to a long-distance marathon, China will have the upper hand." He emphasized that "winning the AI race is not just about the models themselves, but also about adoption and deployment in the real economy," adding, "China's long-term, state-driven industrial strategy offers significant advantages in this regard."

Since designating AI as a strategic industry in 2016, the Chinese government has made substantial investments in research, talent, and infrastructure. According to the Australian Strategic Policy Institute (ASPI), private sector AI investment in the United States is larger than in China, but when considering the Chinese government's financial support, the actual capital gap is not significant. Additionally, as AI usage increases, more electricity is required to operate data centers. In this context, China is seen as having an advantage over the United States under Donald Trump's administration, which opposes alternative energy sources such as solar and wind power.

People are standing at the Nvidia booth at the International Supply Chain Expo held in Beijing, China, last July. Photo by Reuters Yonhap News

Although China currently lags behind the United States in data center construction, it is expected to rapidly scale up thanks to streamlined regulations and energy availability. Lia Pei, a China economist at the economic analysis firm Capital Economics, noted, "In China, the capital-intensive business of building data centers is more often handled by state-owned telecommunications companies," meaning "Chinese technology companies bear less investment risk."

Columnist Parikh also pointed out that China holds a dominant position in the supply chains for raw materials critical to high-tech industries. Furthermore, since China is currently ahead in fields such as humanoid robots and electric vehicles, it is expected to have strengths in the future implementation of physical AI as well.

In January last year, DeepSeek founder Liang Wenfeng revealed in a co-authored paper that training the R1 model cost $294,000 (about 408 million won), using 512 Nvidia H800 chips. In comparison, OpenAI, the developer of ChatGPT, disclosed that even basic model training in 2023 cost over $100 million (about 138.9 billion won), highlighting a stark difference in cost-effectiveness that shocked the market.

However, although DeepSeek carried out seven updates since January last year, their impact has been limited. Analysts attribute this to the United States' export restrictions on advanced chips to China. CNBC, the American business news channel, reported that "while Chinese authorities have encouraged the use of domestic chips amid ongoing Nvidia chip export restrictions, access to advanced computing resources is essential for developing cutting-edge models."