Tencent Says It Has a Stockpile Of NVIDIA AI Chips But Needs Domestic Suppliers

(CTN News) – Chinese technology giant Tencent Holdings said on Wednesday that it has a large stockpile of AI chips from US chip maker Nvidia. However, the recent enactment of broadened US restrictions on high-end chip sales to China will eventually apply to its cloud services.

In an analyst call following Tencent’s third-quarter earnings report, Tencent’s president Martin Lau stated that the US’s decision to ban the export of more chips for artificial intelligence (AI) in China last month will force the company to use existing stock more efficiently and seek domestically produced alternatives to use.

The company has enough NVIDIA chips to continue the development of its “Hunyuan” AI model for at least a couple more generations, so Lau said the curbs will not have a significant impact on the company’s AI capabilities in the near future.

It will be important for us to figure out ways to make the use of our AI chips more efficient in the future,” he explained. I would also like to mention that we will also be seeking out domestic sources for these training chips as well.

Tencent’s new rules will impact its cloud services in which it sells computing power to its clients. Our company believes that the chip ban does indeed affect our ability to resell (use) these AI chips through our cloud services since we are not able to resell them.

In fact, NVIDIA dominates about 90% of the Chinese AI chips market, so Chinese companies with AI ambitions are scrambling to deal with US export restrictions. In response, some have turned to domestic chipmakers, such as Huawei Technologies, in a trend that analysts say threatens NVIDIA’s dominance.

Last week, Reuters reported that Tencent rival Baidu had ordered 1,600 Huawei Ascend 910B chips.

According to the new rules, Tencent has a stock of H800 AI chips that NVIDIA developed specifically for China last year in response to earlier export curbs.

To optimize their use, the company suggested reserving the H800 chips for the most crucial part of AI model development, namely the training process.

Hopefully, we can offload most of the inference capability to lower-performance chips so that we can still use most of our high-performance AI chips for training,” Lau said.

Last week, the industry newsletter SemiAnalysis reported that NVIDIA would announce new AI chips destined for China on Nov. 16.

As a result of the new rules, SemiAnalysis said the chips are equipped with most of Nvidia’s newest AI features, but their computing power has been reduced.

SEE ALSO:

The Dollar Rises As Data Indicates That The Federal Reserve Will Be More Patient With Rate Cuts