US efforts to slow AI development are failing

US efforts to slow AI development are failing

US microchip export management presented last year to restrict China’s technological development of supercomputers or AI systems like ChatGPT has had minimal impacts on China’s technology sector.

Those rules limit shipments of Nvidia and AMD (Advanced Micro Devices) chips that have developed the global technology industry standard for developing chatbots and different AI systems. However, Nvidia has made variants of its chips for the Chinese market, which complete tasks more gradually and thus fulfill US regulations.

IT industry experts informed Reuters that the latest chips would take 10 to 30 percent longer to complete AI tasks, while some may cost twice as much as Nvidia’s quickest US chips.

Even more lagging Nvidia chips are an advance for Chinese companies. In April, Tencent Holdings, one of China’s largest technology companies, calculated that systems using the new Nvidia H800 chip would cut the time it takes to train its largest AI systems from 11 to four days.

An analyst at Shanghai-based 86Research development, Charlie Chai, expressed the AI companies he talked to saw the hurdle as rather small and manageable.

Challenges around AI negotiations

The negotiations between government and industry emphasize a major challenge facing Americans, how to restrict China’s high-tech advances without hurting its own companies? Part of the American strategy in developing the new rules was to avoid such a blow if the Chinese completely outstripped American chips with their production. The strategy would involve doubling the efforts and resources to create its chips in that case.

There are two methods for choosing US chip exports to China. The first lodging a limit on the chip’s capacity to compute extremely precise numbers, a measure designed to restrict supercomputers that can be employed in military research. Chip industry sources expressed this was an adequate action. But computing extremely precise numbers is less relevant in AI-powered tasks like large language models where the quantity of data the chip can decipher is more significant.