Shenzhen Activates China's First 10,000-Card Huawei Ascend Cluster — Zero Nvidia Hardware

Shenzhen activated China's first 10,000-card fully domestic AI computing cluster on approximately April 1, using exclusively Huawei Ascend 910C chips, delivering 11,000 petaflops of compute power. Combined with a previously installed 3,000-petaflop cluster at the same Shenzhen facility, the total site capacity reaches 14,000 petaflops — equivalent to approximately 5.5 million personal computers. Nearly 50 organizations signed framework agreements for access to the new cluster; combined booking rate across both phases was 92%. The cluster contains zero Nvidia chips — described as 'full-stack domestically controlled' AI infrastructure. SCMP reported the cluster directly enables DeepSeek V4 training runs on Huawei Ascend hardware, which was simultaneously being prepared for deployment. The activation represents a major milestone in China's 15th Five-Year Plan AI compute infrastructure buildout: for the first time, China has demonstrated that domestically manufactured chips can be deployed at scale sufficient for frontier AI model training, without relying on US or allied semiconductor companies. The Shenzhen cluster's 92% booking rate also signals strong enterprise demand for domestic AI compute — a commercial validation that Chinese companies are willing to pay a premium for hardware independence.