Chinese state-owned company China Telecom has announced the development of two LLMs fully trained on domestically produced chips.
In a statement from the Institute of AI at China Telecom, published on WeChat and reported by the South China Morning MailThe open-source TeleChat2-115B, which has more than 100 billion parameters, and a second unnamed model, which reportedly has 1 trillion parameters, were trained using tens of thousands of locally manufactured chips.
The statement claims that this development “indicates that China has truly achieved total self-sufficiency in domestic LLM training,” a challenging goal for the country since the US imposed strict export regulations blocking access to high-end GPUs like the Nvidia H100 and A100 . .
Turn to local suppliers
Although China Telecom did not specify who supplied the chips used to train its LLMs, it is likely that Huawei supplied most, if not all, of them. The company is positioning itself as a domestic alternative to Nvidia South China Morning Mail notes that China Telecom “previously revealed that it is developing LLM technology using Ascend chips developed by the Shenzhen-based telecom equipment giant.”
Huawei recently started sending samples of its new Ascend 910C processor to Chinese server and telecom companies for testing, targeting major Nvidia customers in China in the hope that they will at least part of their business change.
Although there is a thriving black market in China for Nvidia’s high-end GPUs, many companies, including ByteDance and Alibaba, prefer to remain compliant and use lower-spec, permitted GPUs such as Nvidia’s H20 to avoid legal and avoid reputational risks and maintain access to Nvidia Support. These companies are increasingly turning to Huawei for their AI needs. It was recently reported that TikTok owner ByteDance had placed an order for 100,000 Ascend processors.
The South China Morning Mail also reports that China Telecom, in addition to Huawei, is exploring hardware from Cambricon, a local AI chip startup, to further diversify its chip offering.