China wants to build generative AI models and would like to power them with its own hardware. Thus, it has to redraw the supply chain now dominated by American chip maker Nvidia.
On account of export ban, Nvidia for some time stopped its H20AI chip made for the Chinese market. China asked its companies to stop buying H20AI chips. DeepSeek released an upgrade of its model in August 2025 powered by the next generation indigenous chips.
The language models use chips for two purposes — one is to train the model, which is a complicated task. However, there is another purpose to use the chips i.e. to draw inferences which created responses from pre-trained models. This is a bigger market for chips. The inferences actually produce the text as well as images for people using generative AI models. Inference making chips are less complex. H20s made in the US are suitable for this purpose. Over a period of time, China could design such chips. It will be filling the performance gap with Nvidia.
Cambricon, founded in 2016, went public four years later, is due for listing as a tech startup in AI chip space. It has become a stock market darling since investors consider it a good candidate to take market share in inference chips. It also is investing in software development.
China is revising its plan of funding the next round of AI development. China does not want to encourage funding from less developed regions.
These are good signs. Young startups should be encouraged to go public as soon as possible. Fundraising in stock-market is simpler than supporting the startups financially. Investors are seeking their own version of Nvidia and therefore will exercise due diligence on behalf of the government. They will promote those companies which champion such worthy causes.
Leave a Reply