It’s inevitable that China will catch up on AI and chip production in general. First, it’s not just about having faster chips. The software matter even more here. Take Deepseek v3 for example, it outperforms top models from the US, but the training cost was only a fraction of models like GPT4 because they used an innovative approach to training and model design.

Meanwhile, on the hardware front, the reality is that we’re starting to hit limits of what you can do with silicon. There’s no where to go past 1nm or so. And it’s a game of diminishing returns, where you have to put ever greater investments in to get increasingly meagre returns. While China is catching up with their fabs, the western companies can’t advance at the same rate.

China is also investing in alternative substrates, and there already positive results like the ability to produce standard 12 inch wafers using novel materials. Even a crude chip built using such materials has potential to make silicon look like vacuum tubes overnight, and then there would be decades of optimizations to follow.

There are no comparable efforts in the west, and I’d argue this kind of fundamental research can only be accomplished with state level funding. You have to commit to invest into this for many years without any clear returns, and private companies will never do that.