China Telecom, one of the country’s leading state-owned telecom giants, has achieved a significant milestone in its quest for technological self-reliance by training two large language models (LLMs) using exclusively domestically-produced chips. This breakthrough represents a critical step in China’s ongoing drive to reduce its dependence on foreign technology, particularly in light of tightening US restrictions on access to advanced semiconductors.
One of the models, named TeleChat2-115B, and another unnamed model, were trained on tens of thousands of Chinese-made processors, marking a major achievement in the face of stringent US export controls. These limitations have severely restricted China’s access to high-end chips from companies like Nvidia, widely regarded as industry leaders in AI hardware. In a statement posted on WeChat, China Telecom’s Institute of AI hailed the development as evidence of the country’s ability to independently advance AI technology, setting the stage for a new era of innovation and self-sufficiency.
A Model of Epic Proportions
The scale of the achievement is underscored by the complexity of the unnamed AI model, which boasts an impressive one trillion parameters. In AI development, parameters are key variables that the model adjusts during training to improve its learning. The more parameters a model has, the more capable and sophisticated it becomes. For comparison, the TeleChat2-115B model, with its 100 billion parameters, is positioned to compete with some of the most advanced AI platforms globally.
China’s drive to stay competitive in AI comes amid growing efforts to counter US trade restrictions, particularly on cutting-edge AI chips such as Nvidia’s A100 and H100 processors. These restrictions have forced Chinese tech companies to accelerate the development of domestic alternatives, with the ultimate aim of reducing reliance on foreign semiconductors.
Domestic Chips Take Centre Stage
While China Telecom did not disclose the exact source of the chips used to train its models, many speculate that Huawei’s Ascend processors played a key role. Huawei, which has faced significant sanctions from the US, has increasingly focused on its AI capabilities and recently began testing its latest AI processor, the Ascend 910C. The chip has already attracted interest from major Chinese server manufacturers and internet companies, many of which previously relied on Nvidia’s hardware.
Huawei’s Ascend processors have become one of the few viable alternatives to Nvidia’s cutting-edge GPUs, making them a critical part of China’s AI strategy. The company’s ongoing development in AI processors highlights China’s broader ambition to build a fully homegrown ecosystem for AI and technology, insulating itself from the impact of international trade barriers.
In addition to Huawei, China Telecom has been working with other domestic chipmakers, such as Cambricon, a start-up specialising in AI processors. These collaborations reflect a concerted effort across China’s tech sector to build a resilient, self-sufficient supply chain that can weather the effects of US export controls.
Navigating Global Restrictions
China’s increasing focus on domestic technology development comes as US sanctions continue to create hurdles for Chinese companies seeking access to advanced foreign-made chips. Nvidia’s GPUs, in particular, have become more difficult and costly to obtain, prompting some Chinese companies to resort to purchasing previous-generation models to avoid the risks of the black market while maintaining access to Nvidia’s support services.
By reducing its reliance on foreign semiconductors, China is gradually gaining greater control over its AI capabilities. The development of high-performance domestic chips represents a crucial shift towards technological independence, a move that aligns with China’s broader ambition to become a global leader in AI and semiconductor innovation.
A Path Towards Self-Reliance
China Telecom’s achievement of training a one trillion-parameter AI model on domestically-produced chips symbolises the country’s growing focus on self-sufficiency in a rapidly evolving global tech landscape. As China continues to face escalating restrictions from the US, its determination to innovate and build domestic capabilities is shaping the future of AI development.
By cultivating partnerships with local tech companies and investing in homegrown chipmaking technologies, China is taking significant steps towards creating a robust and independent AI ecosystem. This development signals a broader shift in the global AI race, with China positioning itself as a formidable competitor in the years to come.
