Despite Nvidia's whooping 90% share in the high-end AI GPU sector, China-based GPU chip makers are still keen on contracting their customers frequently to improve their products, eyeing to tighten the partnerships in the hopes of gaining their orders in the long term.
China's GPU companies are half worried and half joyful seeing Nvidia's success in championing the use of GPUs in global markets. On one hand, Nvidia is educating the market for them and preparing the customers for the era of accelerated computing. On the other, Nvidia is gobbling down more than 90% of the global market, and the Chinese GPU firms only get the crumbs in the domestic market. Nvidia's GPU chips, including A100, A800, H100, and H800 are all hot sales and out of stock in China.
Nevertheless, in the computing market, hardware is just a small part of the solution. Optimization and large-language model (LLM) services are where the profit is. Now with the US restricting the exports of high-end AI chips to China, while the Beijing authority is driving for import replacement with local products, the GPU companies in China said this is an opportunity they have been waiting for.
Chinese GPU makers including ILuvatar CoreX, Moore Threads, and BirenTech have started to engage their products with LLM by collaborating with local cloud computing service providers such as Bidu, while some others such as DenglinAI, Vast AI Tech, and MetaX already have products to support LLM applications.
Some other Chinese companies such as T-Head, an Alibaba subsidiary, and Enflame have used non-GPU solutions to compete in the AI-accelerated computing market.
Although CPU giant Intel has insisted that CPUs will still dominate AI workloads in the future, the latest Nvidia financial report said otherwise. Compared to Intel's second-quarter revenue of US$12.9 billion, Nvidia's revenue totaled US$13.5 billion in the same quarter, the first time in Nvidia's history.
Given Nvidia's revenue estimate of US$16 billion for the third quarter, there is a possibility that Intel may still lag behind in the quarter.
Article translated by Judy Lin
