China Telecom, a state-owned Chinese carrier, has developed two large language models (LLMs) entirely trained on domestically produced chips as US restrictions on advanced semiconductors intensify.

The unnamed model has 1 trillion parameters, while TeleChat2-115B contains over 100 billion. A higher number of parameters generally improves computational capabilities.

China Telecom has not disclosed specific suppliers for its domestic chips, but it has confirmed the use of Ascend chips from Huawei Technologies in developing its LLM technology.

In addition to Huawei’s chips, China Telecom is using chips from the Chinese startup Cambricon to reduce reliance on foreign components.

Recent US export restrictions have also limited access to advanced Nvidia GPUs, which has prompted Chinese companies to seek lower-powered alternatives.