On March 26, 2026, China achieved a milestone in AI computing infrastructure development. The 11,000P intelligent computing cluster invested and built by Shenzhen was officially brought online. Combined with the 3,000P cluster put into operation earlier last year, the facility has now reached a total capacity of 14,000P, becoming the country’s first 10,000-card-level full-stack autonomous and controllable AI computing cluster built entirely with domestically produced advanced chips.
According to reports, the cluster is also the first 10,000-card Ascend 910C supernode AI computing cluster in China. Based entirely on domestic advanced chips and leveraging full-stack autonomous and controllable technologies, it has created a competitive “Ascend + CAAN” indigenous software and hardware ecosystem, effectively driving independent innovation in AI underlying technologies and accelerating the localization of critical infrastructure.
In terms of technical performance, the cluster has achieved several key indicators that rival global leaders. Real-world measurements show that the Phase I 3,000P cluster has an average daily equipment failure rate of only 0.3‰ — significantly lower than that of top-tier enterprise clusters such as Meta’s. Meanwhile, its training linearity for the Pangu-718B large model reaches 93.12%, putting its performance on par with leading clusters from Google and other international frontrunners.
The AI computing project has received strong market response. The utilization rate of Phase I capacity hit 100%. For the newly launched Phase II capacity, nearly 50 enterprises, universities, and research institutions have signed framework agreements for computing cooperation, bringing the overall utilization rate of the two phases to 92%. This has formed an efficient operation model driven by demand, led by key players, and supported by ecosystem collaboration.
As AI technology iterations accelerate and intelligent agents become a new paradigm for large-scale model deployment, token-based billing is rapidly closing the commercial loop, driving explosive growth in computing demand. With the goal of becoming a global AI pioneer city, Shenzhen is accelerating the development of a full-stack autonomous and controllable software and hardware ecosystem, operating major AI infrastructure such as the Pengcheng Cloud Brain III at a high standard, and deepening the “AI+” initiative to promote high-level AI applications across all domains, around the clock, and throughout all industries.
All Rights Reserved © 2026 MemoryMarket.com
