China vs. USA – No Difference

Chinese AI chip makers make no difference to Nvidia et al.

• Every few months, another Chinese chip maker announces that it will be making an AI chip, but seeing as all of them will almost certainly be stuck with SMIC, this is unlikely to make any difference to Nvidia or anyone else.
• This time it’s DeepSeek that is reportedly developing a chip for inference that will be optimised for running its models, and it will be used as part of DeepSeek’s in-house infrastructure.
• Chinese companies have already proven themselves to be competent chip designers, as there are already a handful of companies making chips for vehicles that are competitive against Qualcomm, Nvidia, Mobileye and so on a chip-by-chip basis.
• However, these companies are only competitive because they are able to manufacture their chips at leading-edge foundries, as they have yet to be blocked by the US Department of Commerce.
• Huawei is a special case as it has both links to the Chinese military and, as a national champion, suffered from being used as a political football.
• This resulted in Huawei being placed on the Entity List some years ago, which resulted in much heavier restrictions than those that exist for Chinese companies more generally.
• This is how many Chinese chip designers can use leading-edge foundries while Huawei cannot.
• Consequently, if DeepSeek and other AI chip designers can access advanced manufacturing, then they have a very good chance of producing competitive silicon.
• However, this seems unlikely as it appears that DeepSeek will be forced to use SMIC, which RFM Research has already concluded produces silicon that is very uncompetitive.
• For example, to produce the same amount of compute as a 1GW Blackwell, a data centre built using Huawei’s 2026 Ascend chip (made by SMIC) would cost 3x more and consume 9x the amount of power.
• This disadvantage is mostly caused by the difference in manufacturing, where the gap between TSMC and SMIC is wide and will widen further as time passes.
• This means that no rational builder of a data centre would choose Huawei over Nvidia or any of its non-Chinese competitors.
• This is why I have long argued that Nvidia is not really missing out by being excluded from China, as whatever it loses there, it will more than make up in overseas markets where there will now be no real Chinese competition.
• Hence, for as long as the Chinese chip makers are stuck with SMIC, they will be unable to compete overseas and will only be competitive in their home market.
• The other problem that they face is that at its current build-out rate, the home market is not nearly large enough to support all of the home-grown variants, and there is not enough manufacturing capacity to support them even if there were a lot more spending going on.
• Hence, as far as Nvidia or any of its peers are concerned, it does not matter how many Chinese competitors emerge, as they will all be unable to compete due to their manufacturing disadvantage.
• This, combined with the fact that China is not spending nearly enough on compute capacity to meet its ambitions of rolling out AI services at scale, is going to leave it dependent on importing compute from nearby countries like Vietnam or Malaysia.
• This is a risky proposition as the US is well aware of this workaround, and I suspect that when the Department of Commerce updates its rules in Q3 or Q4 of 2026, it will target this practice.
• This will cause real consternation and further constriction of the compute capacity in China, meaning that China will fall further behind.
• Its models are still likely to see a lot of use outside of the USA, but China’s ability to subsidise compute to drive usage is already limited and may become much more so.
• This is why I continue to think that the West and the USA in particular have the advantage in the AI race that is currently being run.