Tencent claims advanced AI chip sufficiency ‘for generations’ despite risk of U.S. sanctions

Tencent disclosed that it had enough high-end GPUs to train new AI models for years, with the company’s president claiming they had “a pretty strong stockpile of chips” that were acquired previously from Nvidia. 

Chinese companies, including Tencent Holdings, Alibaba Group, and ByteDance, placed at least $16B in orders for Nvidia’s H20 server chips in the first three months of the year—before export control was imposed on the chip in April.

The U.S. blockade of the flow of the most advanced GPUs to China, which was decided in 2022 by the Biden administration and then amplified by Trump, has succeeded in slowing and complicating Beijing’s technological rise. However, Nobel Laureate in Economics Thomas Sargent said, “China has lots of really good engineers. If the US completely cuts China off from chips … You can’t stop it permanently.”

Lau says Tencent and other Chinese chipmakers aim for self-sufficiency 

Tencent’s President pointed out that it actually helped to look at the company’s existing inventory of high-end chips and said Tencent should have enough high-end chips to continue its training of models for years to come. Lau noted that agentic AI and chain of thought workloads needed more GPUs but said software optimization also offered Tencent “quite a bit of room” to keep on improving the inference efficiency.

Huawei founder Ren Zhengfei on February 17th told CCP Chairman Xi Jinping that the lack of domestic production of advanced chips and the harmful impact of U.S. export controls were alleviated thanks to recent breakthroughs by Huawei and its partners. Ren also said he headed a network of more than 2K Chinese companies working collectively to ensure that China achieved more than 70% self-sufficiency in the semiconductor value chain by 2028.

“So it’s a very dynamic situation, and we just … have to manage the situation in a completely compliant way, but also trying to figure out the right solution for us to make sure that our AI strategy … can still be executed.”

– Martin Lau,  Tencent President

Lau also said Tencent was looking at alternatives to GPUs, saying the company could potentially make use of other chips—compliant chips available in China or available to be imported—as well as ASICs and GPUs in some cases for smaller models. Therefore, Tencent revealed that it intended to invest in efficiency initiatives, including training smaller models tuned to the needs of certain applications requiring fewer resources.

Tencent claims it does more in AI with fewer GPUs

Tencent reported a slowdown in GPU deployment, attributing it to a prioritization of chip efficiency over raw numbers, a strategy made clear by artificial intelligence firm DeepSeek. 

Tencent Chief Strategy Officer James Mitchell said the company was getting much higher productivity on a large language model training from existing GPUs without needing to add additional GPUs at the pace previously expected. He added that Tencent had integrated its HunYuan-T1 reasoning model and DeepSeek-R1 model into its Yuanbao chatbot app, pointing out that adding more capabilities will not necessarily require exponential growth in chip usage for each new AI model iteration.

The Chinese tech company’s approach to squeezing chips for extra performance became clear in January with the rollout of DeepSeek R1. Tencent disclosed that its training costs amounted to a few million and its chip usage to just ~2K older generation Nvidia GPUs, which made U.S. tech stock prices tumble. 

Tesla CEO Elon Musk previously said the AI race depended on chip control. He warned that Taiwan produced nearly all advanced AI chips, putting the global supply at risk if China invaded. Musk urged the U.S. to ramp up domestic production for national security.

KEY Difference Wire : the secret tool crypto projects use to get guaranteed media coverage