Implications of H20 Sale Resumption to China

Technology denial hits a turning point

Authors

Arjun Kharpal of CNBC spoke to me about the geopolitical implications of the resumption of H20 sales to China. I am reproducing my answers to his questions below.

You can read Arjun’s article on this topic here

Q1: Jensen has argued that for the American tech stack to “win” it needs to be present in China. But even as the H20 might be allowed entry into China, isn’t the country moving into a future direction of travel that will largely try to replace American tech, including Nvidia hardware? I.e. is this going to buy China time?

It is going to buy China time. But it will also buy the US companies some respite. China is Nvidia’s largest market and is home to 50 per cent of AI developers according to Jensen Huang. If that path is completely closed, American firms like Nvidia will find it difficult to raise revenues and re-invest them in the next round of research and development. It might be justifiable to restrain access to the most advanced chips but to expand the scope of the restrictions doesn’t make strategic sense.

China cannot replace Nvidia easily. Intel didn’t win the processor wars because it built the best chips - Sun’s SPARC and DEC’s Alpha were technically just as good. Intel won because they created an ecosystem that was simply too valuable for developers and manufacturers to abandon. The switching costs became prohibitive. That’s the advantage Nvidia has with its CUDA ecosystem.

I think this move marks a turning point for technology denials. Here’s my take.

Q2: Is China going to struggle so much to catch up that they’ll still want/need Nvidia.

China will still need Nvidia for deployment outside of China because of CUDA’s centrality. Developers work on CUDA and are familiar with that ecosystem. Just like China hasn’t been able to replace Intel’s x86, replacing a globally popular ecosystem is not easy.

Q3: What does this mean for companies like Alibaba, Baidu and Tencent who are trying to race ahead with their models?

Chinese AI models are competitive despite all the GPU restrictions. They might not be the best on some parameters but are good enough for domestic use from a strategic perspective. The crucial element is which models and standards are used outside of China. That’s where US companies have an advantage.