Last Updated: 9 November 2024

China Telecom, one of the largest wireless carriers in mainland China, says that it has developed two large language models (LLMs) relying solely on domestically manufactured AI chips. The state-owned company did not disclose the specific chips used for training the LLMs, but the South China Morning Post reported that China Telecom previously announced plans to use Huawei Ascend AI chips for LLM training. It is likely, therefore, that Huawei’s processors are behind this groundbreaking AI.
If accurate, this achievement represents a crucial milestone in China’s quest for semiconductor independence, especially amidst the ongoing U.S.-China chip war, which has seen the U.S. tighten and ban the supply of cutting-edge chips to China.
Huawei, mostly banned from the U.S. and other allied nations, is a leader in China’s domestic chip industry. It has been making considerable progress with its Ascend AI chip series, currently offering the Ascend 910B. Recent reports suggest that Huawei is sending samples of its next-generation chip, the Ascend 910C, to customers for testing.
Beijing has been urging companies to move away from Nvidia’s AI chips in favor of locally-produced alternatives. This aligns with rumors of U.S. plans to sanction Nvidia’s H20, the company's most powerful accelerator to date, despite its current compliance with Washington’s restrictions. Should Nvidia face further sanctions, Huawei may be well-positioned to fill the resulting gap, meeting both China’s demand and the evolving U.S. regulations.
If China Telecom’s LLMs have indeed been trained exclusively using Huawei chips, it would be a major success for both Huawei and the Chinese government. Huawei has reiterated its commitment to advancing AI chip technology, regardless of U.S. bans and sanctions.
Additionally, such an achievement would underscore the success of Beijing’s investments in semiconductor technologies, reinforcing President Xi Jinping’s assertion that China can progress without ASML’s cutting-edge lithography technology.