China Telecom Develops LLMs Using Huawei's AI Chips Amid US Export Restrictions

The announcement from China Telecom’s Institute of AI highlights the training of TeleChat2-115B, an open-source model with over 100 billion parameters, alongside a second unnamed model boasting 1 trillion parameters. These models were developed using tens of thousands of locally manufactured chips, primarily from Huawei, which is positioning itself as a viable alternative to Western chipmakers. The U.S. has implemented strict export regulations that have hindered access to high-performance GPUs, prompting companies in China to explore local solutions. As a result, major players like ByteDance are increasingly turning to Huawei’s Ascend processors, signaling a shift in the landscape of AI chip supply in China.

Implications of U.S. Export Restrictions

The U.S. government has been tightening its grip on the export of advanced AI chips, with recent proposals targeting specific countries in the Middle East and beyond. These measures are part of a broader strategy to limit the technological capabilities of nations deemed a security risk, particularly in relation to China. The Biden administration’s focus on curbing AI advancements in these regions underscores the ongoing geopolitical tensions surrounding technology supply chains.

Huawei’s Role in China’s AI Landscape

Huawei has emerged as a key player in China’s AI chip market, actively developing alternatives to Nvidia’s offerings. With the U.S. sanctions limiting access to high-end GPUs, Huawei’s Ascend chips are gaining traction among Chinese tech firms looking to maintain compliance and avoid reputational risks associated with using smuggled technology. This shift not only supports China’s ambitions in AI but also highlights the growing importance of domestic technological development in response to international pressures.