A few days ago, there was news in the AI industry that NVIDIA banned AI software companies from using domestic AI chips, compatible with NVIDIA's CUDA platform, which said that the helplessness of domestic AI chips, because NVIDIA's CUDA platform has monopolized the AI market, resulting in many domestic companies encountering compatibility problems when purchasing domestic AI chips.
Domestic software and hardware products are compatible with overseas related products, in fact, it has always caused a lot of controversy, when Kingsoft's WPS compatible with Microsoft's Office, which caused a lot of disputes, and finally Microsoft agreed to WPS compatible with Office in order to keep the Chinese market; In the early years, the domestic operating system was compatible with Google's Android, which also caused a lot of controversy.
This kind of ecological monopoly is a sharp weapon for American companies, they have spent decades to establish ecological advantages, resulting in latecomers are often difficult to surpass, in the operating system market, domestic companies like to say that it is not difficult to do operating systems, but ecological construction is too difficult, although the domestic operating system has made great efforts for many years, but so far the domestic PC operating system has not been able to break the monopoly of Microsoft Windows.
The same goes for AI software, as early as 2005, NVIDIA began to build the CUDA platform, in that more than 10 years, many chip companies have not recognized the problem of AI chip hardware ecology, and even the server chip powerhouse Intel has failed to promote the compatibility of AI chips as soon as possible, until the AI industry has exploded in recent years, Intel, AMD, etc. have not realized the big problem of AI chip ecology.
Now AMD claims that the AI chip it developed can completely crush NVIDIA in terms of performance, after all, the two companies are the only two companies in the world with the strongest GPU chips, and it is not surprising that AMD quickly developed a super-powerful AI chip, but AMD also encountered the monopoly problem of the CUDA platform, which caused AMD to hesitate in the AI chip market.
The pace of domestic chip companies is even slower, after all, they lack the technical foundation of GPU chips, until the past 4 years, the AI market has gradually become popular, and domestic chip companies have developed AI chips, and a domestic chip company that claims that its AI chip performance can be comparable to NVIDIA's A100 is the first to launch AI chips in 2019.
Before 2022, when the United States also allowed NVIDIA to give China's high-end AI chips, domestic AI companies basically did not consider domestic AI chips, which led to the difficulty of domestic AI chip companies.
In 2022, the United States will prevent NVIDIA from China's ** high-end chips A100 and H100, and NVIDIA will customize chips A800 and H800 for the Chinese market.
Last year, the United States further restricted nvidia to China's ** high-end AI chips, nvidia to China's ** AI chips further castrated eighty percent of the performance, and ** extremely expensive, domestic companies finally began to buy domestic AI chips to replace nvidia chips, in such a short period of time to purchase domestic AI chips, compatible with NVIDIA's CUDA platform has become their best choice.
In the face of such a situation, NVIDIA must of course wield its CUDA ecological knife to prevent domestic companies from adopting domestic AI chips. In fact, even if it is compatible, it will lead to a lot of problems, such as running Windows on Linux, the performance of domestic AI chips is not strong enough, and now it has to be compiled at least once in operation and then run, which further affects the performance of domestic AI chips, it can be said that NVIDIA's move is a killer feature.
However, people in the industry believe that NVIDIA's approach is difficult to work, the enterprise market is different from the consumer market, the consumer's usage habits are very dependent, and enterprises are more concerned about security, cost and other issues, PC operating system has been difficult to break the monopoly of Windows, and in the server operating system market, several domestic server operating systems have achieved advantages, Windows share in the server system market has been squeezed to two percent, AI chips certainly belong to the enterprise market, I believe that NVIDIA can't stop the development of domestic AI chips.