China's AI Compute Progress Falls Short of Overtaking the West

China has been making significant strides in artificial intelligence (AI) compute, with hardware improvements closing the gap with Western counterparts. However, key bottlenecks in chip manufacturing and software ecosystem maturity suggest that China is not poised to leap ahead of the West anytime soon.

Why it Matters

The global AI race is intensifying, with major implications for technological leadership and economic dominance. While China has made notable progress, particularly in computational power and data transfer capabilities, several critical challenges remain. These obstacles highlight the complexity of achieving parity in high-performance computing and underscore the continued reliance on Western technology.

Key Metrics: Closing but Not Overtaking

Chinese hardware has seen exponential growth since 2017, with floating-point operations per second (FLOP/s) improving significantly. In 2018, there was an order of magnitude difference between Chinese and Western silicon. Today, this gap has narrowed to a 3x difference. For instance, Huawei’s Ascend 910C, China's leading GPU, is now comparable to NVIDIA’s B200, the leading Western GPU. Notably, the release of the BR100 datacenter GPU briefly positioned Chinese chips as the fastest on the market.

Chip Manufacturing: A Significant Bottleneck

One major hurdle for China is chip manufacturing. U.S. export controls on chipmaking equipment have made it more expensive and challenging for China to produce high-performance chips at scale. These restrictions limit access to advanced lithography tools, which are essential for fabricating cutting-edge semiconductors. As a result, Chinese developers continue to prefer NVIDIA’s chips when available, due to their superior performance and reliability.

Software Ecosystem: A Weak Link

Another critical issue is the immaturity of China's software ecosystem. Unlike NVIDIA’s well-established CUDA stack, which offers robust support and documentation, Chinese alternatives often suffer from bugs, poor documentation, and instability. This makes it difficult for developers to transition smoothly to Chinese hardware, even when performance metrics are comparable. The lack of a mature software environment is a significant deterrent for large-scale adoption.

The Opportunity: Bridging the Gap

Despite these challenges, China has not been idle. Companies like Alibaba, ByteDance, Baidu, and DeepSeek are investing heavily in AI research and development. These efforts aim to enhance both hardware capabilities and software ecosystems. However, it is likely that Chinese developers will continue to rely on Western chips for at least the next few years, particularly for high-stakes applications where performance and reliability are paramount.

Conclusion

While China has made substantial progress in AI compute, significant roadblocks remain. The combination of stringent export controls and an underdeveloped software ecosystem means that China is unlikely to surpass the West in the near term. However, ongoing investments and technological advancements may gradually narrow the gap over time.