China’s tech conglomerate, Baidu, confirms a looming constraint on its procurement of NVIDIA and AMD graphic accelerators (GPUs), spurred by US export restrictions. Despite this, the company remains confident in its software stack’s ability to sustain competitive strength, even with less powerful CPUs.
“Over the long term, we may not have access to the most advanced GPUs,” Baidu’s CEO, Robin Li, acknowledged during a discussion with investors. However, he assures that with their incredibly efficient software stack, “the user experience will not suffer.”
This confidence stems from the fact that the existing GPU stocks and remaining procurement capabilities are sufficient to sustain current AI services, AI-based applications, and at least some future projects. This also includes the support for Ernie, the most powerful AI model in China developed by Baidu as a counter to ChatGPT.
Baidu has traditionally deployed NVIDIA GPUs to train and operate its AI models. A few years ago, Baidu introduced its proprietary chip, Kunlun II, which can compete with NVIDIA A100 GPU. And last year, Baidu began using Huawei’s Ascend 910 processors to reduce reliance on NVIDIA products.
With Baidu’s inability to procure A100 and H100 processors, and void of access to B100 GPUs upon release, the company may have to rely on domestically developed and manufactured processors, which lag behind NVIDIA’s counterparts in performance. In this scenario, Baidu is pinning hopes on its software stack innovations.
“We have a great deal of room for innovation at the application, model, and framework levels,” claimed Li. Baidu’s robust team of researchers and developers, coupled with its unique four-tier AI architecture, is expected to efficaciously utilize less powerful chips for training and operating their AI models, granting Baidu a competitive edge in the Chinese market.
Baidu’s cost-effective AI infrastructure, combined with innovative AI architecture, positions the company favorably compared to domestic competitors. While all Chinese tech companies are wrestling similar difficulties accessing NVIDIA products, international cloud services such as AWS, Azure, and Google Cloud can procure NVIDIA GPUs in larger quantities. The future will reveal how this situation will impact Baidu and its competitors vis-à-vis Western tech firms.
This post was last modified on 03/02/2024