Zhipu AI's GLM-5.3-Flash Debuts Running Entirely on 100,000 Domestic Chinese Chips, Shares Surge 12%

Zhipu AI (Z.ai) revealed on August 27 that its new GLM-5.3-Flash model — which had circulated for a week under the code name 'Ox Alpha' — runs its entire global inference workload on a cluster of 100,000 domestically produced chips, with analysts pointing to likely Huawei Ascend hardware, though Zhipu declined to name its chip supplier. The model, a lower-cost open-weight variant of Zhipu's flagship GLM line with 320 billion total parameters (18 billion active), ranks 10th on the Artificial Analysis Intelligence Index — ahead of DeepSeek's V4 Pro Max — and is priced at roughly one-tenth the cost of comparable offerings. During its preview period the model processed 62 trillion tokens and set an OpenRouter platform record with 11 trillion tokens in its first three days, capturing nearly a third of the platform's weekly coding-model volume. Hong Kong-listed Zhipu AI shares closed more than 12% higher at HK$1,160 on the news, with analysts framing the launch as evidence China can now run large-scale global AI inference workloads entirely on home-grown silicon rather than Nvidia hardware.