Related articles
Credit: CFP
AsianFin -- China's top tech titans are unveiling how they're weathering intensifying U.S. semiconductor restrictions as they race to stay competitive in the global artificial intelligence boom.
During recent earnings calls, Tencent Holdings Ltd. and Baidu Inc. outlined contingency plans including chip stockpiling, software optimization, and adoption of domestic alternatives to mitigate the impact of Washington's export curbs on advanced chips made by Nvidia Corp. and Advanced Micro Devices Inc.
Tencent President Martin Lau said the company has accumulated a "pretty strong stockpile" of high-end GPUs to support ongoing AI training. These chips, crucial for powering large language models, were secured before the latest U.S. restrictions took effect.
"We believe we have enough high-end chips to continue training our models for several more generations," Lau said. He noted that Tencent has been able to achieve strong results with fewer GPU clusters than some of its U.S. counterparts, suggesting a more efficient training methodology.
On the inference side — where AI models are deployed to perform tasks — Tencent is turning to software optimization and smaller models to reduce computing demands. Lau also highlighted the firm's ability to leverage domestic custom-designed chips, reducing its reliance on U.S. technology.
"There are many avenues we can explore to meet growing inference needs — we just need to focus more on the software side rather than relying on brute-force hardware purchases," he added.
Baidu, meanwhile, emphasized the strength of its vertically integrated AI infrastructure. Dou Shen, head of Baidu's AI Cloud business, underscored the firm's "full-stack" approach, combining AI chips, cloud computing, models, and applications such as its ERNIE chatbot.
"Even without access to the most advanced chips, our unique AI capabilities allow us to deliver meaningful value and build strong applications," Shen said.
By controlling more layers of its technology stack, Baidu claims it can manage costs and GPU utilization more efficiently. Shen also pointed to China's growing domestic chip capabilities as a long-term strategic advantage.
"Locally developed, self-sufficient chips — paired with an increasingly mature homegrown software stack — will form a robust foundation for long-term innovation in China's AI ecosystem," he said.
China's efforts to develop a self-reliant semiconductor industry have intensified since the U.S. began tightening export controls. While Chinese-made GPUs still lag behind those from U.S. leaders in terms of performance, analysts say progress is undeniable.
"Stockpiling is just one piece of the puzzle," said Gaurav Gupta, a semiconductor analyst at Gartner. "China has built momentum across the semiconductor supply chain, from materials and equipment to design and packaging. While gaps remain, the pace and ambition have been impressive."
Gupta noted that although domestic chips can't yet match U.S. performance, they are rapidly improving — and in some cases are already viable for specific AI workloads.
The growing resilience of China's tech sector has prompted criticism from U.S. industry leaders. Nvidia CEO Jensen Huang recently called the Biden administration's chip curbs a "failure," warning that the restrictions are hurting American firms more than their Chinese rivals.
As the global AI arms race accelerates, China's internet giants appear increasingly determined to chart their own path — with or without access to U.S.-made silicon.
更多精彩内容，关注钛媒体微信号 （ID：taimeiti），或者下载钛媒体 App
