Domestic AI chips are accelerating the replacement of NVIDIA, and Huang Jenxun regrets it.
Although the United States has been banning Nvidia from China's advanced AI chips in order to curb the development of Chinese artificial intelligence, Chinese chipmakers are working hard to develop AI chips to replace NVIDIA, and recently another Chinese chip company has also received good news.
Haiguang Semiconductor, which is Haiguang Semiconductor, began to research high-performance server chips after obtaining the license of AMD, because it can match the current x86 architecture, so it occupies a large market share and has become an alternative for Haiguang Semiconductor.
In addition to continuing to research server chips, Haiguang has also conducted research on artificial intelligence chips, which have become more and more popular in recent years, and developed DCUs, which can compete with NVIDIA. It is said that Haiguang's first-generation DCU is only about 50% of that of NVIDIA A100, and now, the good news from Haiguang is that its second-generation DCU has reached the level of 80% of NVIDIA.
Compared with other domestic semiconductor companies, Haiguang Semiconductor's leading position is that their server chips can be fully compatible with Intel and AMD servers, which greatly alleviates the urgent needs of domestic manufacturers; If Haiguang's DCU chip is added, it can better improve the computing power of AI big data.
This is also the main goal of the two companies, Intel and AMD, both of which have launched their own AI chips, and want to occupy a place in the hot AI chips, because the AI chip industry is developing rapidly, and NVIDIA has surpassed TSMC and Intel to become the number one in the world.
However, Chinese semiconductor companies such as Haiguang, as well as Intel, AMD and other companies, are facing the same problem, NVIDIA has established a relatively strong CUDA ecosystem after more than ten years of development, and the CUDA ecosystem has become a major obstacle in front of these chip companies.
It is precisely because of this that NVIDIA will reduce the price of A800 and H800 chips without price reduction, and Chinese manufacturers have to compromise, and this time, NVIDIA will weaken the function of H20 by 80%, which makes Chinese companies unbearable, and instead uses a large number of domestic AI chips, it is said that domestic AI chips have received 5000 orders, with a profit of up to 500 million.
Haiguang and other Chinese AI chip companies released artificial intelligence chips, in terms of performance has far exceeded the H20 provided by NVIDIA for China, due to the huge difference in performance, coupled with the gap between the two, so that many companies have chosen independent research and development of AI chips, and the artificial intelligence chips produced by Haiguang can be seamlessly connected with the existing data center, so as to break through NVIDIA's CUDA ecosystem and accelerate the domestic AI chip pair Nvidia replacement.
For the development of domestic AI chips, it is said that NVIDIA has felt tremendous pressure, changing its previous posture of not falling but rising, and the price of H20 processors supplied to China has been greatly reduced, showing that only through fierce competition will American semiconductor manufacturers give fair treatment to the requirements of Chinese companies, otherwise NVIDIA is likely to still stand on top of the world.
The semiconductor industry in the United States has always been very arrogant, once relying on its monopoly position, greatly reduced its own **, but since last year's domestic chips were replaced by a large number of domestic chips, China's chips have fallen by 90%, and now the same is true in the field of artificial intelligence, which shows how rapid China's chip technology has progressed, and at the same time, it is also proving to the United States that China is no longer the only semiconductor company, and they have enough strength to manufacture products that can replace them.