Driving the Digital Economy with Computational Power

The booming development of global large models and generative artificial intelligence (AIGC) has become one of the hot topics at this conference, and discussions around the computational resources, computing power, and infrastructure supporting AIGC applications have naturally become the focus of attention.

Computational power forms the foundation of innovation in the artificial intelligence industry, and the continuous innovation of large models has driven an explosive growth in the demand for computational power. It can be said that the efficiency of large model training and the speed of innovation fundamentally depend on the size of computational power. As of January 2023, ChatGPT’s official website had a total of 616 million visits, with a monthly operational computational power of about 4874.4 Pflop/s. GPT-4 reached a scale of one million trillion parameters, resulting in a significant increase in computational power demand. Data indicates that the growth of large model parameters is directly proportional to the square of computational power.

In 2022, the “14th Five-Year Plan for the Development of the Digital Economy” proposed accelerating the implementation of the “Eastern Numerical, Western Computational” project, enhancing data center inter-networking and cross-regional data exchange capabilities, strengthening edge computing capabilities for specific scenarios, and enhancing computational power coordination and intelligent scheduling. Efforts were made to promote the construction of green digital centers, accelerate energy-saving transformations in data centers, and continuously improve the utilization of renewable energy in data centers. In the ten national data center clusters across the country, the data center’s scale reached 540,000 standard racks, with computational power exceeding 1,350 trillion floating-point operations per second, roughly equivalent to the computational power of 27 million personal computers, driving investments in various aspects of over 190 billion yuan.

This year is a critical year for the effective implementation of the “Three-Year Action Plan for the Development of New Data Centers (2021-2023),” and many local governments, including nodes of the “Eastern Numerical, Western Computational” project, are paying close attention to the construction and layout of intelligent computing (super-computing) centers. Combining the advantages of local green new energy resources, they aim to promote the landing and rooting of emerging industries and optimize the industrial structure layout. The new data centers are commonly referred to as super-computing or intelligent computing centers in the industry, serving as both the infrastructure for the digital economy and the core driving force behind general artificial intelligence. For each 1% increase in the computational power index, the digital economy and GDP are expected to grow by 3.5‰ and 1.8‰, respectively.

In the future, data centers will undoubtedly provide green computational power based on new energy sources, aligning with the “dual-carbon” indicators. They will develop in two directions: “universalization” and “intelligence.” “Universalization” means that more enterprises and institutions can conveniently access computational resources and afford the corresponding costs  according to their needs. “Intelligence” means that more intelligent application scenarios will emerge, leading to exponential growth in computational power demand, necessitating timely and robust support from computational resources. The end-users will penetrate multiple industries, primarily applied in internet, finance, software, power, industrial, medical, and other sectors.

AI large models will undoubtedly become the “new engine” driving the core development of the artificial intelligence industry. IDC predicts that by 2026, China’s artificial intelligence software and application market will reach $21.1 billion, and AI will enter a crucial period of large-scale practical applications. Computational power centers or other specialized fields will undoubtedly be the trend in the future.