Following recent actions by the US to restrict device supplies to China, Hong Kong announced on Wednesday its intention to explore alternative sources for AI education chips.
During a conference call with analysts, Martin Lau, the president of Tencent, addressed the White House’s recent decision to prohibit further exports of AI chips to China. Lau emphasized the need for Tencent to optimize the utilization of its existing chips and seek out domestically manufactured artificial intelligence (AI) chips.
Lau stated, “We must enhance the efficiency of our AI chips and explore potential private suppliers for these educational chips.”
These comments from Lau coincide with the challenges faced by foreign companies aspiring to incorporate AI technologies due to the expanding trade restrictions imposed by the United States on AI chip exports to China.
Prior to these restrictions, the US semiconductor giant Nvidia had a stronghold on the Chinese AI chip market. However, an increasing number of Chinese tech firms are now turning to local chip manufacturers like Huawei Technologies for their AI chip needs.
Recent reports from Reuters revealed that Baidu, a competitor of Tencent, has placed an order for 1,600 Ascend 910B chips from Huawei.
Despite having accumulated a sufficient supply of Nvidia chips to sustain the advancement of its Hunyuan AI model for several generations, Tencent anticipates minimal immediate impact from the new chip restrictions on its AI capabilities.
Nevertheless, Lau expressed concerns about the implications of the new regulations on Tencent’s cloud services, through which the company offers computing resources to clients.
Lau mentioned, “We foresee challenges in resuming the sale of AI chips through cloud-based services.”
Tencent currently boasts one of the largest inventories of AI chips in China, having secured early access to Nvidia’s H800 chips, ensuring a substantial supply of L800 chips.
In response to a previous US ban on AI chips destined for China, Nvidia introduced the H800 chips specifically tailored for the Chinese market. However, these new chips are also affected by the latest regulatory measures introduced last month.
To maximize the utility of the remaining H800 chip inventory, Lau suggested reserving them primarily for model training, a critical phase in AI development.
In an effort to preserve the high-performance AI chips for training purposes, Lau stated, “We will explore outsourcing the majority of inference tasks to lower-capacity chips.”
The article credits Josh Ye as the reporter, with Toby Chopra and Jane Merriman serving as editors, adhering to the Thomson Reuters Trust Principles.
