USA vs China – Small Time Smuggler

Chip smuggling is pretty irrelevant.

• A lot of attention is being paid to the smuggling of Nvidia chips into China, but the reality is that if the US really wants to stop China from using its chips, it needs to prevent China from importing compute from overseas data centres.
• Nvidia is being far more cautious these days about who it sells its AI systems to, which has meant that the customer list in Asia has almost halved.
• This has no bearing whatsoever on Nvidia’s financials, as Asia is still a small portion of sales and demand so robust that any orders that get cancelled can easily be sent elsewhere.
• However, it does mean that the companies that have been buying the equipment and re-exporting it to China are going to find it harder to continue this practice in the future.
• This follows the embarrassing discovery that Supermicro employees had been charged with aiding the re-export of Nvidia equipment to China through the company’s channels.
• This makes for a great story and an entertaining read, but in the grand scheme of things, it is irrelevant and is going to have very little, if any, impact on China’s ability to develop and sell AI services at scale.
• This is because the scale of smuggling is very small when considered in the light of the global AI roll-out.
• The Supermicro smuggling operation is thought to have been part of $1bn of illicit imports into China between May and July 2025.
• If I annualise that and round it up a bit for the operations that have gone undiscovered, I would estimate that this activity might have made up about $5-6bn on an annual basis.
• When one puts this in the context of USA capital expenditure on data centre systems alone (~$560bn) in 2026, one begins to realise that this smuggling is irrelevant.
• China is expected to spend around $75bn on data centres in 2026, which RFM thinks is very far from the kind of scale that China needs to roll out AI services at scale to 1.2bn citizens, 80% of whom are pretty keen on the idea of AI.
• Even against China’s modest investment, the smuggling operation is small and is not going to meaningfully help China develop leading-edge models, which it has already demonstrated that it is quite capable of doing.
• Hence, I think that the furore around chip smuggling is mostly so that powers that be can be seen to be doing something while at the same time ignoring the elephant in the room.
• This elephant is Chinese companies renting entire data centres in overseas countries such as Thailand, Malaysia or Vietnam, which are not subject to the restrictions and running their workloads there.
• RFM Research has concluded that Chinese AI silicon is so far behind that it makes no economic sense to use a Chinese data centre when imports are available at much better tokens per dollar.
• Furthermore, China is simply unable to make the chips fast enough to meet demand, and so it has no choice but to rely heavily on compute coming from overseas.
• If one is serious about holding China up, restricting these imports would have a far greater effect than playing cat and mouse with smugglers.
• Hence, the extent of the USA’s intentions will become clearer the next time it updates its import regulations, which should happen in September or October this year.
• This could 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.