Huawei's Ascend AI Chips Drive China's Push for Semiconductor Independence

China’s ambition to develop a fully independent semiconductor industry has found a key player in Huawei, whose Ascend AI chip ecosystem is rapidly scaling as part of the country’s wider industrial strategy. While Huawei’s chips currently lag behind global leaders in raw efficiency and performance, their extensive deployment and growing ecosystem underpin China’s efforts to reduce reliance on foreign technology, especially amid ongoing U.S. export restrictions.

Context: China’s Semiconductor Independence Drive

China’s semiconductor independence is a central goal of its Made in China 2025 policy and subsequent industrial strategies, aiming to increase domestic content and self-sufficiency in critical technologies such as AI and chip manufacturing. The government has invested heavily through programs like The Big Fund, supporting domestic firms including Huawei, Semiconductor Manufacturing International Corporation (SMIC), and others to build a vertically integrated supply chain covering chip design, manufacturing, packaging, and materials. The target includes achieving roughly 50% self-sufficiency in semiconductor equipment by 2025, up from just over 13% in recent years.

Huawei’s Ascend AI Chip Ecosystem

Huawei’s Ascend chips serve as the backbone of a China-first AI hardware ecosystem. The company produces its chips via an enhanced 7-nanometer process in partnership with SMIC, though this technology still trails the 4nm nodes used by competitors like Nvidia. Despite this, Huawei compensates through large-scale deployment and integration. For example, the CloudMatrix 384 system stacks 384 Ascend processors with an optical interconnect network delivering nearly 300 petaFLOPS of compute power.

This system, however, consumes about 559 kilowatts at peak load—nearly four times the power of Nvidia’s latest comparable systems—reflecting lower power efficiency. Yet, China’s data centers benefit from lower power costs and fewer regulatory constraints, making this trade-off viable for domestic large-scale AI training.

Performance and Software Ecosystem

While Huawei’s Ascend chips do not yet match Nvidia’s Hopper-based H100 or Blackwell-based B200 GPUs in raw performance or efficiency, internal tests suggest competitive advantages in specific AI model classes. The Ascend ecosystem includes a maturing software stack with Huawei’s CANN programming environment and MindSpore AI framework, which support popular AI models and can translate PyTorch and TensorFlow graphs. Huawei is also moving to open-source parts of its toolchain to foster local and international development under export control limitations.

Expansion and Roadmap

Huawei plans to double Ascend 910C chip production to approximately 600,000 units in 2026, with total Ascend chip output potentially reaching 1.6 million dies across models. The company has revealed a multi-year roadmap extending to 2028, introducing new Ascend 950, 960, and 970 series chips featuring improved compute power, bandwidth, and integration with high-bandwidth memory. These advancements align with Huawei’s goal to double compute capabilities annually and maintain competitiveness in AI compute demand.

Additionally, Huawei is building large-scale AI clusters, such as the Atlas 950 SuperPoD, designed to link thousands of Ascend chips into single logical machines capable of advanced AI reasoning and learning, aiming to challenge Nvidia’s dominance in AI infrastructure.

Strategic Significance and Challenges

Huawei’s Ascend AI ecosystem exemplifies China’s broader strategy to create a self-reliant semiconductor supply chain that can operate independently of U.S. and European technology. This is increasingly critical given tightening export controls and geopolitical tensions. The ecosystem’s vertical integration—from chip design and packaging to AI frameworks—represents a credible fallback plan should sanctions intensify.

However, challenges remain. Huawei’s chips currently trail leading-edge GPUs in efficiency and raw performance. Overcoming these gaps will require continued innovation in chip architecture, manufacturing processes, and software optimization. China’s ambitions also hinge on expanding domestic chip equipment production and materials supply, areas where it currently lags behind global leaders.

Conclusion

Huawei’s Ascend AI chips are central to China’s pursuit of semiconductor independence, reflecting significant progress in building a domestic AI hardware ecosystem. While performance and efficiency gaps persist compared to Western competitors, Huawei’s strategy of scale, ecosystem integration, and government support positions it as a key player in China’s evolving semiconductor landscape. The coming years will be critical as Huawei rolls out next-generation chips and expands production to meet China’s growing AI compute needs.

Written by Deepak Periyasamy.

Call to action: Visit TNT PCB and TNT E-Comp.