Sales by Nvidia in China have plummeted, despite the company selling versions that met US restriction rules, allowing local rivals like Huawei to make inroads into its market.
But since China is pushing for a lifting of the export ban, its investments in chip design and manufacture do not seem to have allowed it to catch up with US capabilities. Donald Trump‘s plan to replace Joe Biden’s attempt to control AI proliferation with a global licensing system is more pragmatic.
AI diffusion is a multinational endeavour, and trying to keep countries out will not work. Building safeguards into the process of technology development is a better approach.
So long as countries adopt, and operate within, protocols for the responsible use of AI, there is little to justify the denial of computing power, which can anyway be developed locally. Any anti-proliferation regime must be guided by the principle of encouraging commercial development of AI while discouraging its weaponisation.
Denial of computing power limits both and is, thus, a blunt tool against AI proliferation, a point Huang makes about Nvidia’s lost sales in China. The intensity of AI development in the country, Huang points out, makes local chip development an inevitability.
Generative AI needs to train on large amounts of data beyond the capacity of countries to generate. Since countries are unwilling to share data, the technology must be applied locally, by domestic agents. Computing power rationed globally according to use case works if there is control over high-end chip supply. Yet, it is unlikely that the global chip supply response will remain inert.
The US can retain its technological edge in AI by building more muscle in chip-making, reducing the need for other countries to develop their capabilities that could eventually overtake it.
