China has unveiled a plan to increase its total computing power by over 50% by 2025. This move underscores Beijing’s heightened emphasis on advancements in supercomputing and artificial intelligence. It also comes in the midst of escalating competition between China and the United States across various high-tech sectors, including semiconductors, supercomputers, and artificial intelligence. Additionally, the plan is a response to U.S. export controls imposed on chipmaking equipment.
A plan jointly issued by six Beijing departments, including the Ministry of Industry and Information Technology (MIIT), has outlined an objective for China’s total computing power to reach 300 EFLOPS by 2025. EFLOPS, which corresponds to one quintillion floating-point operations per second, serves as a measure of a computer’s speed.
The MIIT reported in August that China’s computing power had reached 197 EFLOPS in the current year, a rise from 180 EFLOPS in 2022. The ministry acknowledged that China ranks second in computing power globally, with the United States leading, although specific details about the U.S. computing power were not provided.
Expanding the availability of computing power is now a growing priority for Beijing due to the substantial computational demands of AI training.
A recent Google blog post highlighted that maintaining training times of several weeks or less for the world’s leading generative AI models will necessitate tens of EFLOPs of AI supercomputing.
As outlined in the plan, China intends to establish additional data centers throughout the nation to enhance businesses’ access to computing resources.
To meet the demands of the rapidly advancing AI industry, Beijing also intends to enhance the computational infrastructure in western China.
Regions like southwestern Guizhou, which are extensive but less populated, have been designated to establish large-scale data centers to support the country’s internet needs. For instance, Apple has partnered with a local entity to set up data centers in Guizhou, catering to its users in China.
Another key focus is to enhance the speed and efficiency of the computing network. The plan stipulates that the transmission speeds between critical computing facilities must not exhibit a latency of more than 5 milliseconds.
(Note: The corrected growth target in the headline and paragraph 1 is 50%, not 30%.)
Reporting by Josh Ye; Editing by Sonali Paul
Our Standards: The Thomson Reuters Trust Principles.