Advertisement

Warning of glut in AI computing centres as China readies for return of Nvidia’s H20 chip

Industry demands are changing and only about 30 per cent of the country’s intelligent computing capacity is being used

Reading Time:3 minutes
Why you can trust SCMP
8
Only about 30 per cent of China’s intelligent computing capacity is being used, according to an industry report. Photo: Shutterstock
China is facing a glut of AI computing power and the booming service sector needs to adapt to changing demand, state media has warned as US tech titan Nvidia prepares to resume exports of its made-for-China chips.
Citing data from the Inspur Institute of Artificial Intelligence, Science and Technology Daily reported on Thursday that despite the rapid adoption of artificial intelligence, only about 30 per cent of the country’s intelligent computing capacity was being used.

As of November, nearly 150 intelligent computing centre projects were up and running throughout China, and another 400 or so projects were planned or under construction, according to the newspaper, the mouthpiece of the Ministry of Science and Technology.

“The rapid development of artificial intelligence technology has given rise to a boom in the construction of intelligent computing power. However, amid this boom, many blindly launched intelligent computing centres are running idle, and a large amount of idle computing power is waiting to be activated,” the report said.

“Blind investment has led to waste of resources and market imbalances, which have had a negative impact on the high-quality development of [China’s] computing power industry.”

01:02

Nvidia to resume selling H20 graphic processing chips to China in boon for AI

Nvidia to resume selling H20 graphic processing chips to China in boon for AI
AI is evolving quickly and so too is computing demand, as some industry processes shift from training models to inference models such as Hangzhou-based DeepSeek’s R1.
Advertisement
Select Voice
Choose your listening speed
Get through articles 2x faster
1.25x
250 WPM
Slow
Average
Fast
1.25x