As revealed in a report presented by Hong Kong’s SCMP newspaper, the local Beijing government suggests giving companies some subsidies for purchases made with Enemy-made processors. The draft policy focuses on GPU chips, which the US export ban has most hurt. Chinese industries relied heavily on this product to advance the country’s AI applications.
China’s GPU subsidy push
The Bureau of Economy and Information of the Municipal Bureau of Beijing will provide a specific amount of funds to acquire domestically made GPUs to “alleviate the situation of controlling computing resources through an intelligent, proactive way,” says the draft policy document. The aim, lamely speaking, is that by 2023, China should become independent of computing infrastructure, relying only on “made-in-China” products. Admittedly, in the past few years, numerous domestic solutions have been presented and observed in close-ups.
The competitive landscape of the GPU market worldwide was made publicly available last December (18 active firms), and we discovered that, in particular, China had increased the development number of GPU designer firms at a rapid pace. In most cases, the Chinese firms, which are very optimistic in their pre-launch calls regarding the efficiency and performance of their vehicles, don’t achieve the promised results after the launch.
This happens in the Western part of the world, too, but companies like AMD and Nvidia are already exquisite, making some of the coolest graphic cards for consumers. We have experienced reports of China indirectly delivering hands-on-sanctioned technologies and goods like Nvidia H100. Nevertheless, the supply pathways aren’t stable and reliable enough to be durable for serious trade. China tries to economize the application of domestic resources to make up for the shortcomings mentioned above.
Fostering local AI prowess
The top Chinese makers with a good chance of benefiting from the GPU and AI processing subsidies are maybe firms like Biren Technology, Moore Threads, Innosilicon (PowerVR), MetaX, and Zhaoxin (Glenfly). The Longson, who made the GPU, also debuted its own, which is as good in FP32 performance as the 4th gen. Likely, elsewhere in China, there are labs and factories in the cradle, maybe out of the petri dishes in the research institutions and universities.
The location of most of the LLMs developed in China, Beijing, is said to be way more than fifty percent of this kind of robot. The SAMP story shall choose GPUs and projects supported by subsidies for training comprehensive and industry-specific large language models (LLMs). This endeavors to extend the reach of AI and cognitive technologies with Research and Development (R&D) in AI-supportive technologies such as software, AI-specific processors, silicon photonics, and quantum computing chips that also have continuous development.
This article originally appeared in Hong Kong’s SCMP newspaper.