As of October 2026, Huawei reports that its Ascend AI chips have surpassed Nvidia in sales volume within China. Despite this, leading Chinese AI laboratories continue to utilize Nvidia hardware for training large language models, citing the established software ecosystem of the CUDA platform.
DeepSeek serves as a primary example of this dual-track strategy, planning to deploy at least 160 000 Huawei Ascend 950DT chips for inference tasks while maintaining Nvidia hardware for training workloads. Transitioning training processes to Huawei’s CANN platform is estimated to require 50% more time and capital.
Market projections for 2026 suggest that domestic manufacturers could secure approximately 50% of the Chinese AI accelerator market. Conversely, Nvidia’s market share in the region is expected to decline to roughly 8%, compared to approximately 40% in 2025.
The semiconductor landscape remains influenced by US export controls, which include ongoing legal actions against individuals involved in smuggling Nvidia chips into China. While Washington enforces restrictions, it continues to issue limited approvals for specific Nvidia models, creating a complex supply environment for local firms.
