Huawei Cloud

Full briefing

Huawei Cloud Computing Technologies Co., Ltd. operates the cloud computing and AI infrastructure division of Huawei Technologies. It is China's third-largest public cloud provider by revenue after Alibaba Cloud (Aliyun) and Tencent Cloud, and is notable as the only major cloud globally that runs on entirely proprietary domestic AI accelerators at scale. Huawei Cloud's AI infrastructure is built on the Ascend AI chip family, designed by HiSilicon (Huawei's in-house chip design arm). The Ascend 910 (first-generation, fabricated by TSMC before 2020 export restrictions) and Ascend 910B (second-generation, fabricated by SMIC at approximately 7nm-class using DUV multi-patterning) form the GPU-equivalent compute backbone of Huawei Cloud's AI clusters. The ModelArts platform (Huawei Cloud's AI development platform) enables training and deployment of large language models, computer vision models, and speech recognition models on Ascend hardware — competing with AWS SageMaker, Azure ML, and Google Vertex AI. ModelArts is the primary deployment platform for Pangu Models, Huawei's foundation model family (NLP, image, scientific). The supply chain closed-loop is strategically significant: HiSilicon designs → SMIC fabricates → Huawei Cloud deploys. This is the only major cloud AI stack that does not depend on US-origin chips (no NVIDIA, no AMD), US-origin EDA tools (Cadence/Synopsys licenses are restricted), or TSMC fabrication for its current AI deployment. However, it comes with performance constraints: the Ascend 910B achieves roughly 60–70% of H100 performance on typical LLM training benchmarks, and SMIC's 7nm-class process has lower yields than TSMC's N7. Huawei Cloud is central to China's 'AI sovereignty' narrative — demonstrating that a large-scale AI cloud can operate entirely within a domestic supply chain. It is the primary beneficiary of ByteDance, Baidu, and other Chinese AI companies being denied access to NVIDIA H100/H200 hardware.