Why China’s Biggest Tech Firms Are Still Wary of Huawei’s AI Chips
Huawei’s ambition to dethrone NVIDIA in the Chinese AI hardware space has hit a wall – and not just because of geopolitics. Despite bold claims and powerful new chips like the Ascend 910C, China’s tech giants such as ByteDance, Alibaba, and Tencent are holding back from embracing Huawei’s hardware en masse. Why? The reasons are a mix of technical limitations, market dynamics, and strategic hesitations.
One of the biggest roadblocks is NVIDIA’s entrenched CUDA ecosystem. Over the years, companies have poured vast resources into optimizing their software stacks around NVIDIA’s platform. Huawei’s alternative, the Compute Architecture for Neural Networks (CANN), is simply not there yet in terms of features and ease of use. Switching would be costly, time-consuming, and risky – especially for firms running critical AI workloads.
Then there’s the issue of competition. Huawei isn’t just a hardware vendor; it’s also a direct rival in several sectors. For many Chinese tech firms, relying on Huawei’s chips would feel like strengthening the hand of a competitor. It’s one thing to support domestic technology – it’s another to hand a rival the keys to your infrastructure.
Technical shortcomings also play a role. Reports have emerged about overheating issues with the Ascend 910C, raising concerns over performance stability in high-intensity AI training scenarios. While Huawei’s chips offer theoretical parity with NVIDIA’s H100 in FP16 compute, features like FP8 support – critical for memory-efficient AI models – remain lacking or require clunky workarounds.
Furthermore, the US Department of Commerce’s strict export control guidance has made Huawei’s chips toxic for Chinese firms with international ambitions. Using these chips without special clearance could expose companies to serious legal risks abroad.
Lastly, most of these companies still have hefty inventories of NVIDIA GPUs stockpiled – bought when the threat of sanctions loomed large. Until those run out, there’s little incentive to switch, especially when NVIDIA continues to deliver with innovations like its Blackwell-based supercomputers and robust AI infrastructure pipeline.
Ironically, the very ecosystem lock-in Huawei wants to break is the same reason it’s struggling to get traction. Until its chips mature, the software support ecosystem improves, and geopolitical concerns subside, the status quo – NVIDIA dominance – is likely to hold, at least among China’s tech elite.