China Telecom trains AI model with 1 trillion parameters on domestic chips

China Telecom has created two large language models trained solely on domestic chips, including a 100 billion parameter model and a trillion-parameter model. This achievement, using thousands of Chinese-made chips from companies like Huawei and Cambricon, demonstrates progress in China's AI self-reliance efforts amid US export restrictions on advanced semiconductors. It aligns with China's strategy to develop a homegrown AI ecosystem and reduce dependence on foreign technology, especially Nvidia's GPUs.