Despite the U.S. government’s attempt to limit China’s development of artificial intelligence (AI) technology through microchip export controls, the nation’s tech sector continues to thrive.
U.S. restrictions on shipments of Nvidia and Advanced Micro Devices (AMD) chips aimed to curb Chinese advances in supercomputers and AI systems, such as ChatGPT. However, these efforts have had only a minimal impact on China’s tech industry.
U.S. microchip controls and China’s adaptation
Nvidia has created slowed-down versions of its chips for the Chinese market to comply with U.S. regulations.
Industry experts predict that the latest chip, the Nvidia H800, will take 10% to 30% longer to complete certain AI tasks and could double some costs compared to Nvidia’s fastest U.S. chips. Nevertheless, even these slowed chips represent an upgrade for Chinese firms.
Tencent Holdings, one of China’s largest tech companies, estimates that using Nvidia’s H800 will reduce the time required to train its most extensive AI system by over half, from 11 days to just four days.
According to Shanghai-based analyst Charlie Chai of 86Research, the AI firms he has spoken to see the handicap as relatively small and manageable.
Slowing China’s progress without hurting U.S. companies
The U.S. faces a difficult balancing act as it tries to limit China’s technological progress without damaging its own companies.
When implementing the microchip restrictions, the U.S. aimed to avoid such a significant disruption that China would abandon U.S. chips and intensify its domestic chip development efforts.
One aspect of the U.S. strategy was to draw a line that would gradually degrade China’s capability over time without causing immediate disruption.
Despite the constraints, Nvidia continues to sell its H800 chips to major Chinese technology firms such as Tencent, Alibaba Group Holding Ltd, and Baidu Inc for AI-related tasks.
China’s AI industry: Overcoming restrictions and adapting to challenges
While the U.S. restrictions have limited chip-to-chip transfer speeds and impacted AI performance, experts in the AI industry believe that there are ways to overcome these limitations.
Naveen Rao, the CEO of MosaicML, a startup specializing in optimizing AI models for limited hardware, estimates a system slowdown of 10-30%. However, he believes that algorithmic workarounds can help circumvent the issue for at least a decade.
Money is another factor that can soften the blow of the restrictions. Despite higher costs and longer training times for AI systems, large Chinese companies like Alibaba and Baidu can afford the financial burden.
Innovations in AI: Reducing impact of U.S. speed limits
AI researchers are also working on reducing the size of their systems to lower the cost of training AI models like ChatGPT. By using fewer chips, they can decrease chip-to-chip communication and minimize the effects of U.S. speed limits.
Cade Daniel, a software engineer at Anyscale, a San Francisco startup that provides AI software solutions, believes that if the industry continued on its previous trajectory of creating larger AI models, the U.S. export restrictions would have had a more significant impact.
China’s AI industry has demonstrated its resilience by adapting to U.S. restrictions and finding ways to thrive. The nation’s tech sector continues to innovate and grow, even in the face of challenges posed by international regulations.
