China’s Z.ai Debuts First Major Model Trained Fully on Huawei’s AI Chip Stack

Chinese AI startup Z.ai has unveiled GLM-Image, an advanced multimodal AI model that represents a shift in the ongoing tech war between superpowers.

Z.ai (formerly known as Zhipu AI outside China until its rebranding in 2025) disclosed that GLM-Image is reportedly the first top-tier multimodal model trained entirely on Ascend chips developed by Huawei Technologies.

The announcement says it was built using Huawei’s Ascend Atlas 800T A2 servers and the MindSpore framework, with servers equipped with Huawei’s Kunpeng processors and Ascend AI chips, this represents China’s first fully domestically-trained AI model.

Chinese hardware

Training a competitive AI model on domestic Chinese hardware represents a technical achievement.

Just 2.5 weeks ago, Z.ai’s GLM-4.7 claimed the top position among open-source AI models worldwide, scoring 68 points on the Artificial Analysis Intelligence Index and edging past competitors. Now, with GLM-Image trained entirely on Chinese chips, Z.ai has demonstrated that technological independence isn’t just possible—it’s already happening.

Performance metrics from late December reveal decent capabilities. GLM-4.7 achieved 73.8% on SWE-bench and 66.7% on SWE-bench Multilingual in coding capabilities. Mathematical and reasoning capabilities saw even more improvements, with the model scoring 42.8% on the challenging Humanity’s Last Exam benchmark—a significant 12.4 percentage point jump from its predecessor.

While global semiconductor tensions have intensified over the past year, Z.ai has quietly built one of the world’s most competitive AI ecosystems using entirely domestic infrastructure. This alters the calculus for both international tech companies and policymakers who assumed Chinese AI development would be constrained by hardware limitations.

Money talks

The broader financial picture reveals just how ambitious this company has become.

Two weeks ago, the startup launched a major fundraising initiative, raising HK$4.35 billion ($560 million).

Z.ai aims to become the first large language model developer listed in Hong Kong, positioning itself at the forefront of the AI revolution. Financial data from 2.5 weeks ago shows revenue climbing from 57.4 million RMB ($8.2 million) in 2022 to 312.4 million RMB ($44.9 million) in 2024—a 130% compound annual growth rate.

Substantial backing from China’s tech giants has poured in, including Alibaba, Tencent, Ant Group, Meituan, Xiaomi, and HongShan, with a notable 2023 funding round bringing in 2.5 billion yuan ($350 million).

Looking ahead

Global expansion strategies are already in motion. The company rebranded internationally in 2025, positioning itself as a direct global competitor to established AI platforms like ChatGPT and Claude. This isn’t about domestic market dominance—it’s about challenging American AI hegemony on the world stage.

However, geopolitical tensions continue to complicate this landscape. In January 2025, the U.S. Commerce Department added Z.ai to its Entity List over national security concerns, potentially limiting access to certain American technologies.

But Z.ai’s latest announcement suggests those restrictions may have actually accelerated China’s push toward technological self-reliance, rather than slowing it down.

Some experts offer a more down-to-earth perspective on whether China will push ahead in the AI race.

The post China’s Z.ai Debuts First Major Model Trained Fully on Huawei’s AI Chip Stack appeared first on eWEEK.