China's Z.ai claims it trained a model using only Huawei hardware (www.theregister.com)

Chinese AI company Zhipu AI, operating as Z.ai, has announced the successful training of its new model, GLM-Image, entirely on Huawei hardware, marking what it claims to be the first advanced AI model developed in China without relying on Nvidia or AMD technology. The GLM-Image model employs a hybrid architecture of autoregressive and diffusion decoding, enabling the joint generation of images and text. Utilizing Huawei’s Ascend Atlas 800T A2 server and its proprietary Kunpeng 920 processors, Z.ai demonstrated the potential of a domestically produced computing platform, though specifics on the quantity of hardware or training speed remain undisclosed. This development is significant for the AI/ML community as it reflects China's growing capabilities in producing advanced AI solutions independently, potentially disrupting global hardware markets, particularly for Nvidia and AMD, amidst strict U.S. export controls affecting chip sales to China. The open-source nature of GLM-Image further raises concerns, as it could influence global technology standards and practices, particularly regarding digital governance and the spread of values through AI technologies. The implications extend beyond technical achievements, hinting at geopolitical dynamics as nations scrutinize the influence of AI-driven models in shaping cultural and operational norms worldwide.