Nvidia's grip on the Chinese AI chip market has slipped below 60% for the first time, as domestic manufacturers delivered 1.65 million AI GPUs while government pressure pushes data centers toward Chinese alternatives.
This is exactly what US export controls were designed to prevent—and it's happening anyway. When the Biden administration banned exports of advanced Nvidia chips to China in 2022, the stated goal was to slow Chinese AI development. Instead, it accelerated the timeline for Chinese semiconductor independence.
Companies like Huawei, Moore Threads, and Biren Technology have ramped up production of chips that, while not matching Nvidia's top-end performance, are good enough for most AI training and inference workloads. And in a market this large, good enough plus government backing equals rapid market share gains.
The Chinese government isn't subtle about this. State directives are explicitly pushing data centers to use domestic chips, even when foreign alternatives might be technically superior. When you combine industrial policy, nationalist buying preferences, and genuine technological progress, you get exactly what we're seeing: Nvidia losing ground in what had been its fastest-growing market.
Here's what makes this geopolitically significant: the AI chip supply chain was supposed to be the West's insurmountable advantage. Advanced lithography from Netherlands, design tools from California, manufacturing from Taiwan. China couldn't break in, the theory went, because these dependencies were too complex.
But export controls created exactly the incentive structure needed to overcome those dependencies. When you can't buy the chips you need, you build them yourself. When that's not technically possible, you make it politically necessary.
The chips coming out of Chinese fabs still lag Nvidia's latest generation by a couple of years. But the gap is closing, and more importantly, Chinese AI companies are learning to optimize for the hardware they have rather than waiting for access to hardware they can't get.
