On May 27, Zheng Jun, CTO of Huawei Technologies Co., Ltd.’s Financial Systems Division, delivered a keynote speech at the ‘2026 Phoenix Greater Bay Area Finance Forum – Financial Summit’ in Shenzhen. Citing a Stanford University report, he noted that the performance of Chinese AI models lags behind U.S. models by just 2.7%. Since February 2026, the volume of calls made to Chinese AI models has consistently exceeded that of U.S. models. He emphasized that China remains committed to open-source AI development; currently, the capabilities of open-source models in China are approaching those of closed-source U.S. models. Meanwhile, China’s ecosystem of domestic AI models and computing power is rapidly maturing, reaching ‘world-leading levels’.
Zheng also outlined the current state of AI development from a generational perspective: over the past two years, domestic token consumption has surged 1,400-fold, signaling that AI is transitioning from early-stage experimentation to large-scale implementation. He pointed out that in the first 47 days of 2026 alone, 30 major AI models have been released in China — averaging one new model every 1.5 days. Among them, DeepSeek V4, which is natively compatible with Huawei’s Ascend chips, has drawn significant industry attention. Leveraging China’s robust computing power and electricity infrastructure, domestic AI models enjoy notable cost advantages compared to their foreign counterparts.