China Telecom’s Breakthrough in AI Technology

Significance of Domestic Chip Usage

China’s Domestic AI Chip Development

China Telecom, one of the country’s state-owned telecom giants, created two large language models (LLMs) and trained them solely on domestically-produced chips. This milestone represents a significant step in China’s drive towards self-reliance in AI technology, particularly as the United States imposes stricter controls on access to advanced semiconductors.

Training AI on Domestically-Produced Chips

AI Institute’s Efforts and Models

Challenges of US Restrictions

According to China Telecom’s Institute of AI, the Institute of AI trained two models, TeleChat2-115B and another unnamed model, using tens of thousands of Chinese-made chips. This accomplishment is remarkable, especially as US export restrictions escalate, limiting China’s access to high-end processors from Nvidia and other foreign companies.

In a statement on WeChat, the AI institute stated that this achievement demonstrates China’s ability to train LLMs independently, signaling a new era of innovation and self-reliance in AI technology. The scale of these models is impressive—China Telecom confirmed that the unnamed LLM has one trillion parameters. In AI, parameters are variables that enable models to learn; having more parameters makes AI more complex and powerful.

China’s Push for AI Independence

Overcoming US Export Restrictions

Domestic Alternatives for AI Processors

Chinese companies are working to keep pace with global AI leaders, many of whom are based outside the country. Due to Washington’s export restrictions on Nvidia’s AI chips, such as the A100 and H100, China sought domestic alternatives. Consequently, To reduce reliance on Western technologies, Chinese companies focused on developing their own processors. The TeleChat2-115B model, with approximately 100 billion parameters, can reportedly perform as well as mainstream AI platforms.

Domestic Chips Powering AI Innovation

Huawei and Domestic Partnerships

Huawei and Cambricon’s Contributions

China Telecom did not specify which company supplied the chips used to train these models. However, it is believed that Huawei’s Ascend chips play a significant role in China’s AI development strategy. Huawei, which has faced significant US penalties in recent years, is increasing its focus on AI by testing its latest processor, the Ascend 910C, with potential domestic clients.

Chinese server companies and internet giants, which previously used Nvidia chips, are now testing Huawei’s latest AI processors. The Ascend processors are seen as a vital part of China’s strategy to reduce its dependence on foreign technology. In addition to Huawei, China Telecom is partnering with other domestic chipmakers, including Cambricon, a start-up specializing in AI processors. These partnerships demonstrate China’s goal to build a domestic ecosystem of AI solutions, shielding it from US trade controls.

A Shift Towards Self-Sufficiency

By developing its own AI chips, China reduces its reliance on foreign hardware, especially Nvidia’s costly and restricted GPUs. While it remains challenging for Chinese companies to access the latest Nvidia hardware, some have resorted to sourcing older generation models to maintain official support while avoiding black market risks.

This achievement marks a significant shift in China’s approach to AI and semiconductor technology. This shift underscores self-sufficiency and resilience, particularly in response to American protectionist trade policies.