When something gets into Jensen Huang’s head, he goes after it and often succeeds. This is what happened in July of this year when managed to convince Trump to let him sell his H20 chip in China. History has just repeated itself and has managed to the president lifts the veto on H200 chips (although keeping a part). The problem is China, which does not see it very clearly.
what has happened. China is preparing restrictions aimed at limiting access to NVIDIA’s H200 chips, according to Financial Times. If these restrictions end up being implemented, it will mean that the chips will not be available to any company that wants to buy them; They will first go through a pre-approval process, which includes explaining why chips from domestic companies do not meet their needs.
In addition, there is another fact that adds up: for the first time, China has put national chips from companies like Huawei and Cambricon in its official procurement list. This list is a kind of purchasing guide for public institutions and large state groups that move billions a year in contracts.
Why is it important. It is further proof that the Chinese government’s priority is not to depend on American technology for the development of its AI. Their bet is to favor the use of national chips even though they are not technologically at the level of NVIDIA chips.
It’s not the same. China has already responded with distrust when NVIDIA obtained permission to sell H20 chips months ago and it seems that now they want to follow the same path, but there is a big difference: the H20 chips were the most basic, the H200 GPUs are much more advanced and represent a greater technological advantage, especially in more demanding tasks such as training large language models.
What Chinese companies say. According to South China Morning PostAI companies in China such as ByteDance, Alibaba or Tencent continue to prefer to use H200s because they are much more powerful than the national alternatives offered by Huawei or Cambricon. Additionally, much of these companies’ code is based on NVIDIA’s Hopper microarchitecture, allowing them to use the chips without having to rewrite the code. On the other hand, developers who do not need maximum performance are wary of using American chips given the instability of the situation.
The energy. NVIDIA’s CEO has been around for a while pressing for the US to lift these restrictions. Their pitch is that if China does not have access to NVIDIA chips, then they will improve their domestic chips and win the AI race, but there is more. He has also warned that China has a huge energy advantagelargely thanks to government subsidies. He has already managed to convince Trump to sell chips and now the most difficult thing remains.
Image | Wikipedia
GIPHY App Key not set. Please check settings