China Has Large-Scale But Weak AI Computing Power, Experts Say(Yicai Global) Feb. 15 -- China has advantages in the scale of computing power but still has a long way to go when it comes to the uses related to artificial intelligence, experts told Yicai Global.
"China lags behind the US in intelligent computing power, and chip issues also affect its intelligent computing centers," an AI expert told Yicai Global. The world's second economy has advantages in computing power scale but still needs to learn how to better use and improve computing power, the person added.
With the rise in popularity of ChatGPT, an AI chatbot developed by OpenAI, the demand for computing power, an essential element of AI together with algorithms and data, has surged.
"China has more ultra-large internet data centers than the US but is weaker in computing power, so it must focus on improving the quality of its IDCs," said Xu Feng, executive deputy director of the AI Development Research Center at the Institute of Scientific and Technical Information of China.
China's intelligent computing power reached 155.2 ExaFlops in 2021, according to data from the 2022-2023 Evaluation Report on the Development of China's AI Computing Power. The figure will likely rise 52 percent annually to 1,271.4 EFlops in 2026.
"AI needs strong computing power for model training, reasoning, and operations, which needs to be more accurate as training intensifies and operations become more complicated," according to Zou Debao, executive deputy general manager of the AI Industry Research Center at CCID Consulting.
"The demand for intelligent computing power began to increase when AlphaGo, a computer program that plays board game Go, was released in 2016," said He Renlong, chief expert at the Shanghai Academy of Next Generation Information and Communications Technology. "But the launch of ChatGPT urges a new round of demand for AI computing power."
ChatGPT consumes a total of around 3,460 petaflop-days in computing power, which means performing 10 billion calculations per second for 3,640 days. This would require at least seven data centers, each having an investment of CNY3 billion (USD439 million) and a computing power of 500 pfs-days.
As of the end of 2021, China had 5.2 million standard IDC racks, 19 million IDC servers with a total computing power of over 140 EFlops, over 450 large and ultra-large IDCs, as well as more than 20 intelligent computing centers in use.
Chinese areas with developed digital economies, such as Beijing, Shanghai, and Guangzhou, recently released policies to advance the construction of intelligent computing centers. Beijing said on Feb. 13 that it would support leading companies to create ChatGPT-like models, hike investment in AI computing power infrastructure, and accelerate the supply of basic AI data.
Editor: Futura Costaglione