The Nvidia Chip Deal Helps China Achieve Its AI Goals
US concerns about competition and security are traded away for a piece of the action.
Nvidia CEO Jensen Huang faces similar stress from Washington and Beijing.
Photographer: Na Bian/Bloomberg
In 2017, China outlined its plan to make the country the dominant global force in artificial intelligence by 2030. President Xi Jinping provided the will and the funds to move quickly, ramping up homegrown semiconductor manufacturing, data center construction and a power grid to support it. But one thing had been holding it back in its fight with the US — reduced access to the best American-designed semiconductors. The White House has given up at least some US advantage in return for a 15% cut of the sales of AI chips to China by Nvidia Corp. and Advanced Micro Devices.
While President Donald Trump is wrong when he says Nvidia’s H20 chip is “obsolete,” it’s right to note that it is not close to being the company’s most powerful one and would not be used by the likes of OpenAI to build their most sophisticated models. The H20 was designed specifically to avoid previous export limits and is better suited to “inference” — the running of AI — rather than “training” — the creation of new models. The fact the H20 was cleared for sale, then not, and is now back again is proof enough that much of this is “just theater and negotiation,” suggested leading semiconductor analyst Patrick Moorhead, who consulted with the current White House on this issue. The same appears to go for any concerns about national security threats.