Tencent Stockpiled Enough NVIDIA AI GPUs To Last Them A Couple More Generations

Chinese AI firms such as Tencent have revealed that they have stockpiled vast amounts of NVIDIA GPUs to last them for generations but also plan to switch to domestic solutions in the future.

According to a report by Reuters, Tencent has disclosed that the company has identified the potential impacts of the new US curbs, which have restricted them from the access of NVIDIA's higher-end H800 and A800 AI GPUs. In a third-quarter earnings report, Tencent says that it has enough of NVIDIA's equipment to train its custom "Hunyuan" model, which could last for several generations to come. For a certain period, Tencent will have no effect on its "working model" and the company will continue to develop its existing AI-based resources.

We will have to figure out ways to make the usage of our AI chips more efficient. And we will also try to look for domestic sources for these training chips

-Tencent's President Martin Lau

However, Tencent isn't completely "invulnerable" to the new US sanctions, since the company believes that its ability to sell "cutting-edge" cloud services to clients will be put at stake. This shows how deep Team Green was when it comes to market penetration, since its AI GPUs were vital for the industry's development, especially in the case of cloud computing. With a domination of more than 90% in the Chinese market, the recent US ban will have a huge impact in slowing down the pace of growth of the local industries.

While there are rumors revolving around that NVIDIA might introduce more Hopper GPUs for the affected markets, the consistent disruptions have bothered Chinese firms to a greater extent, which is why they are looking for alternatives. Tencent will reportedly join the likes of Baidu and other Chinese firms in the desire to look for "in-house" solutions to counter US influence on the markets, and the most viable alternative is seen as Huawei's Ascend AI chips, with its flagship model reaching around 80% inference and 70% training performance compared to the NVIDIA's A100 accelerator. Tencent plans on utilizing homegrown equipment for "less priority" tasks in an attempt to diversify its product portfolio.

We'll try to see whether we can offload a lot of the inference capability to lower-performance chips so that we can retain the majority of our high-performance AI chips for training purposes.

It will be interesting to see how the market evolves since Chinese tech firms and their AI solutions are now the only alternative for China's huge domestic market which hungers for more chips. This could apparently squeeze out AMD and NVIDIA from the regional AI markets if the sanctions under the Biden government continue to harshen.

News Source: Reuters