Enflame Technology, a Shanghai-based AI chipmaker, has been taking part in building large-scale computing clusters to provide services for training large models, laying a solid foundation for the development of China’s domestically produced large AI models.
Zhang Yalin, founder and general manager of Enflame Technology, said they tailor-make dedicated GPU chips according to the specific needs of different major internet companies and different computing centers.
The company is eyeing clusters with 100,000-GPU-scale computing power.
"In fact, our 10,000-GPU computing cluster already opened in Qingyang, Gansu Province, two year ago. This year, we will move towards the goal of a cluster with 100,000-GPU-scale computing power," Zhang said.
Compared with the world's top computing chip companies, China's computing chip industry is still in its infancy, he said, and hopes that the construction of computing power clusters can better serve the development of domestic super-large-scale models.
"The goal of building a 100,000-GPU cluster is to provide more services. Besides large-scale inference, such as inference services like WeChat, we will also serve the training of large models, moving towards a training and inference capability of 100,000 GPUs. We hope to achieve a significant breakthrough in clustering this year," he said.
In 2025, China's computing chip procurement reached 220-250 billion yuan (about 32.6 to 37 billion US dollars), representing a year-on-year increase of approximately 45 percent. Industry insiders said that the growth rate will be even higher in 2026.
Chinese AI chip maker targets large computing clusters for model training
