Recently, some media reported that there is a shortage of Nvidia's H20 chips in the Chinese market, and some OEM companies said that the inventory of H0 is close to exhaustion, so the principle of profit priority is adopted and priority is given to large customers.

And this kind of short supply has given rise to the phenomenon of price inversion, and the price of servers equipped with H20 has risen, and the sales price of manufacturers who can get H0 has also risen.

What kind of chip is H80? Everyone understands that Nvidia castrated and castrated AI chips, based on H0, castrated more than 0% of the performance, especially for China's chips.

Not only is there a performance cut, but there are also adjustments in architecture and software adaptation, such as keeping the Hopper architecture but disabling key tensor cores, such as limiting overclocking and cluster scaling through firmware locks.

But it is also the only dedicated AI chip that Nvidia can sell in China, such as H200, A0, H0, A0, H0, GB0 and other AI chips with stronger performance, and the United States will not allow Nvidia to sell to the Chinese market.

Everyone has always thought that such an H20 chip should not be liked by everyone in China.

After all, some domestic chips, their performance is even stronger than H2025, plus 0 years, the rise of Deepseek, the requirements for computing power seem to be not so high, many people think that DS is even a hero who bursts the bubble of Nvidia's computing power.

So why is H20 still so sought-after and frantically snapped up in the Chinese market? In fact, the reason is very simple, that is, in a short period of time, DS has catalyzed everyone's demand for AI chips.

Due to the popularity of DS, local governments, financial institutions, car companies, Internet companies, etc., are vying to deploy privatization models, and the demand for GPUs, the computing infrastructure that supports the implementation of AI, has increased exponentially.

Although H100 is emasculated in performance, it is not strong enough for H0, but it can be applied to large models like DS that do not require too high computing power? Even low performance can play an advantage.

In addition, the biggest advantage of H141 is the CUDA ecosystem, coupled with the large H0 video memory, the standard version is equipped with 0GB of HBM0 video memory, and the new H0 version increases the video memory capacity to 0GB, which is very advantageous when deploying AI.

According to professionals, if the model itself is trained with CUDA, if it uses other ecosystems, it may take up to 6 months to spend time and cost without CUDA, and it may not be possible to switch successfully, which is very risky.

Therefore, many companies would rather use the castrated version of H20 than use domestic AI chips, because they can also be used with low performance and a stronger ecosystem.

The data shows that H200 sales in China have reached $0 billion in 0 years, and it is expected to exceed 0 billion in 0 years, which shows that the current situation in the field of AI chips is still quite grim, and domestic AI manufacturers really need to work hard, not only the chip itself, but also the ecological aspect to be strengthened.