Huawei Ascend

Enterprises are strategically diversifying away from Nvidia in the AI accelerator market due to over-reliance vulnerabilities including pricing, supply chains, and geopolitical risks. Alternatives like Huawei offer negotiating leverage, mitigate vendor lock-in, and provide access to alternative supply chains, especially in regions with Nvidia restrictions. Huawei’s Ascend platform excels in inference workloads, offering potential cost and power efficiency. This transition involves a risk assessment, weighing diversification benefits against Nvidia’s established ecosystem. For some, this realignment is crucial for competitiveness and future-proofing AI initiatives.

Huawei’s CloudMatrix 384, powered by Ascend 910C processors and the MindSpore framework, challenges Nvidia’s dominance in AI acceleration. Adopting Huawei’s ecosystem requires significant adaptation, including transitioning from PyTorch/TensorFlow to MindSpore and utilizing the CANN software stack. ModelArts, Huawei’s AI platform, supports the entire development lifecycle. While lacking the maturity of Nvidia’s ecosystem, Huawei aims to offer a viable alternative, reducing reliance on US-based technology. Transitioning requires personnel training and code re-architecting, but Huawei provides resources to facilitate the process.

At Huawei Connect 2025, Huawei revealed its Ascend chip roadmap, including the 950, 960, and 970 series for AI and HPC, challenging NVIDIA’s dominance. Despite semiconductor manufacturing challenges, Huawei focuses on domestic design, proprietary tech, and open-source strategies. New Ascend chips promise performance leaps with enhanced interconnects. Huawei’s SuperPoD and SuperCluster strategy, powered by UnifiedBus 2.0 (an open protocol), aims to provide scalable, high-performance computing, expanding into general-purpose computing with Kunpeng 950 processors and TaiShan SuperPod. Huawei claims significant performance advantages over competitors.

The 2025 WAIC features Huawei’s debut of the Atlas 900 A3 SuperPoD, built on the Ascend 384 Super Node. This super-node utilizes advanced bus technology for high-bandwidth, low-latency interconnection between 384 NPUs, addressing communication bottlenecks in large AI clusters. Huawei’s CloudMatrix 384 (CM384) AI cluster, built around Ascend chips, delivers 300 PFLOPs of dense BF16 compute power, reportedly surpassing NVIDIA’s GB200 NVL72. Analysts suggest Huawei’s scaled solution surpasses current market offerings from NVIDIA and AMD.