Chinese company Zhipu AI has trained image generation model entirely on Huawei processors, demonstrating that Chinese firms can build competitive AI systems without access to advanced Western chips. The model, released on Tuesday, marks the first time a state-of-the-art multimodal model completed its full training cycle on Chinese-made chips, Zhipu said in a statement. The Beijing-based company trained the model on Huawei’s Ascend Atlas 800T A2 devices using the MindSpore AI framework, completing the entire pipeline from data preprocessing through large-scale training without relying on Western hardware. The achievement carries strategic significance for Zhipu, which the US Commerce Department last year added to a list of entities acting contrary to US national security or foreign policy interests over its alleged ties to China’s military. The designation effectively cut the company off from Nvidia’s H100 and A100 GPUs, which have become standard for training advanced AI models, forcing Chinese firms to develop alternatives around domestic chip architectures. Followign that listing, Zhipu began collaborating with Huawei on GLM-Image. Huawei’s Ascend processors have become the primary alternative for Chinese AI companies restricted from purchasing Nvidia’s hardware. The model’s successful training on Ascend chips provides a data point that Chinese firms can develop competitive AI systems despite restricted access to Western chips. “This proves the feasibility of training high-performance multimodal generative models on a domestically developed full-stack computing platform,” Zhipu’s statement added. Zhipu has made GLM-Image available through an API for 0.1 yuan (approximately $0.014) per generated image. The company released the model weights on GitHub , Hugging Face , and ModelScope Community for independent deployment. The pricing positions GLM-Image as a cost-effective option for enterprises generating marketing materials, presentations, and other text-heavy visual content at scale. Technical approach and benchmark performance GLM-Image employs a hybrid architecture combining a 9-billion-parameter autoregressive model with a 7-billion-parameter diffusion decoder, according to Zhipu’s technical report . The autoregressive component handles instruction understanding and overall image composition, while the diffusion decoder focuses on rendering fine details and accurate text. The architecture addresses challenges in generating knowledge-intensive visual content where both semantic understanding and precise text rendering matter, such as presentation slides, infographics, and commercial posters. On the CVTG-2K benchmark, which measures accuracy in placing text across multiple image locations, GLM-Image achieved a Word Accuracy score of 0.9116, ranking first among open-source models. The model also led the LongText-Bench test for rendering extended text passages, scoring 0.952 for English and 0.979 for Chinese across eight scenarios including signs, posters, and dialog boxes. The model natively supports multiple resolutions from 1024×1024 to 2048×2048 pixels without requiring retraining, the report added. Hardware optimization strategy Training GLM-Image on Ascend hardware required Zhipu to develop custom optimization techniques for Huawei’s chip architecture. The company built a training suite that implements dynamic graph multi-level pipelined deployment, enabling different stages of the training process to run concurrently and reducing bottlenecks. Zhipu also created high-performance fusion operators compatible with Ascend’s architecture and employed multi-stream parallelism to overlap communication and computation operations during distributed training. These optimizations aim to extract maximum performance from hardware that operates differently from the Nvidia GPUs most AI frameworks target by default. The technical approach validates that competitive AI models can be trained on China’s domestic chip ecosystem, though at what cost in development time and engineering effort remains unclear. Zhipu did not say how many processors or how long it took to train its model, nor how the requirements compared to equivalent Nvidia-based systems. Implications for global AI development For multinational enterprises operating in China, GLM-Image’s training on domestic hardware provides evidence that Chinese AI infrastructure can support state-of-the-art model development. Companies with Chinese operations may need to evaluate whether to develop strategies around platforms like Huawei’s Ascend and frameworks like MindSpore. The release comes as Chinese companies invest in domestic AI infrastructure alternatives. Whether export controls will slow or accelerate the development of parallel AI ecosystems remains a subject of policy debate. This article first appeared on Infoworld .