Baidu, China's leading search engine, is intensifying its efforts in AI technology development to secure a global leadership position. In the Q1 2024 Earnings Call, CEO Robin Li highlighted significant advancements, particularly in the creation of a system that combines GPUs from different vendors into a single compute cluster for AI training. This innovation is crucial as Baidu transitions from an Internet-centric business to an AI-first enterprise, with its generative AI, ERNIE, expected to be central to its offerings by 2025.
However, Baidu faces substantial challenges due to US restrictions on exporting advanced technologies to China, including the latest Nvidia, AMD, and Intel chips, which are vital for AI development. This limitation has driven Chinese companies to innovate and seek alternatives to continue their AI progress. As American-built GPUs are less accessible, Chinese firms are developing domestic GPUs and finding ways to circumvent the restrictions, such as acquiring advanced GPUs through unofficial channels and creating new technological solutions.
A notable breakthrough from Baidu is its advanced GPU cluster management technology, announced during the earnings call. This technology allows the integration of GPUs from various vendors into a unified computing cluster, significantly enhancing the efficiency of training large language models (LLMs) despite hardware limitations. "Leveraging our technical expertise, we can now integrate GPUs from various vendors into a unified computing cluster to train an LLM," Li stated. "Our platform has demonstrated high efficiency with this setup on a GPU cluster composed of hundreds, even thousands of GPUs. This is an important breakthrough because of the limited availability of imported GPUs."
This innovative approach positions Baidu to mitigate the impact of US technology sanctions and maintain momentum in AI development, underscoring the company's resilience and adaptability in the face of international challenges.
