So, China currently is responsible for roughly 10–20% of Nvidia’s revenue indirectly (it’s hard to know the exact numbers, but it sits in this range). Now, as you know, China has been a great ally of Nvidia and was there when Nvidia needed the revenue for the beat and raise. Squeezing every dollar is needed for the “beat and raise,” and that shows how important selling to SEA is for Nvidia. It doesn’t matter whether it is direct or indirect. Direct sales would be great for Nvidia because they could have a clear indication of what China will order and also clear old inventory instead of facing future write-offs, instead of relying on subsidiaries and local companies to order Nvidia chips through other partners, which gives Nvidia less visibility.

In addition, it’s true that buying Nvidia chips and just working straight on models is way easier than getting local chips that may be cheaper but possibly inferior in performance. Not only that, but the language and tools to work with them are something you need to learn, and very few may know them compared to CUDA, which has a big community. However, the Chinese government has stepped in here and is forcing big players in the field (like DeepSeek) to work on local chips and not rely lazily on Nvidia. They do permit them to work on Nvidia, for example for inference, but they want independence in this area, especially for internal uses (government, army, and so on).

Does it make sense now?