Peng Cheng Laboratory, a research lab in Shenzhen, said 10,000P of its Cloud Brain III system is now in trial use. Its website says it announced this at the World Computing Network Conference on October 15 and 16. The "P" means petaflops, a measure of calculations per second. The lab said the live part already helps train domestic large AI models and runs a research network for AI for science. The rest is still being installed. IT Home reported a planned total of 16,000P.
The lab describes a home-grown "supernode" design, built to cut delays in moving data around the system. It names no chip supplier. IT Home, citing public reports, linked the system to Huawei's Ascend 384 SuperPoD. The sources give no date for reaching 16,000P and no end date for the trial. They also name no users, report no real-world speed, and do not say how the 10,000P figure was measured.
A capacity number alone is hard to compare across vendors. A buyer would want to know who runs on the system, how fast real jobs finish, and when the trial ends. The lab also pictures computing as a shared network, moving from pooling resources toward on-demand use. How outside teams would get access is not stated.