($) Second Order Effects of the Expanded US AI Chips Ban
The yard is getting bigger, the fence is getting higher, with no end in sight.
This is premium member only content. Thank you for being a paid subscriber of Interconnected Premium.
The US Commerce Department released an anticipated set of new export control rules that expand upon the AI chips ban to China released last October. Although most of the rules won’t take effect until next month, and the typical public comment period is still in order, it is clear that in a supposedly “small yard, high fence” world, the yard is getting bigger, the fence is getting higher, with no end in sight.
Here is my high level, plain language summary of the expanded rules:
The types of chips that will require notifications and licenses have expanded considerably. In Nvidia’s case, in addition to its topline A100 and H100 chips, which are outright banned since last October, its A800 and H800 chips (the two modified types specifically designed to be compliantly sold to China), as well as its L40, L40S, and RTX 4090 are now all inside the “small yard”. (For readers unfamiliar with the L40 series and RTX 4090, the L40 series are GPUs designed for data center workloads and RTX 4090 is for gaming and creative graphic designs.) This expansion also affects AMD and Intel, both of which have made similar modified products to be compliantly sold to China.
The jurisdiction of this licensing requirement is expanding to all 22 countries, where the US currently has an arms embargo, to close the loophole where China could obtain the banned chips from a third country – a case of the fence getting higher.
Similar restrictions are being placed on more types of semiconductor manufacturing equipment, including the expanded jurisdiction over all 22 countries with an arms embargo – another case of the yard getting bigger.
The government entity list added two more Chinese companies, Biren and Moore Thread, and all their subsidiaries. Both companies develop semiconductors and seek to compete with companies like Nvidia. This part of the rule expansion is taking effect immediately.
The official press release from the Bureau of Industry and Security (BIS) which contains links to the nearly 300-page-long rules text has all the gory details.
There are some obvious first order effects of this expanded AI chips ban.
China’s AI progress and ambition will stall further. (Here, I’m talking specifically about the large-language-model-generative AI ilk of AI, not necessarily self-driving or other less GPU-intensive forms of machine learning applications.) This is on top of a cooling AI tech scene when it comes to VC investment and startup activities, additional “red teaming” regulatory burden on all generative AI service providers, and “Great Firewalling” Hugging Face, a platform where almost all open source AI development and collaboration happens.
American chip companies – Nvidia, AMD, Intel – will also get hurt, as these three leading companies all derive a solid portion of their revenue from China. Here are each companies’ China revenue as a percentage of their total revenue, according to their most recent 10K filing: