Huawei Claims Ascend Has Overtaken Nvidia in China’s AI-Accelerator Market
Редакция ЦОД:Главное
Huawei says its Ascend processors have surpassed Nvidia in share of China’s AI-accelerator market, citing internal company data. The claim comes as US export restrictions and demand for domestic supply are reshaping accelerator procurement in China.
- Huawei claims Ascend has taken the leading share of China’s AI-accelerator market, but neither exact shares nor independent verification were provided.
- Huawei is prioritizing Chinese customers as domestic demand strains its available production capacity.
- Atlas 950 is being promoted at a scale of 256,000 compute cards; a 1 million-NPU configuration remains a possible future expansion.
- A domestic shift toward Ascend could increase the importance of non-CUDA software compatibility and large-scale AI infrastructure.
- Potential infrastructure demand effects on networking, cooling and power depend on actual deployment patterns and cluster architectures.
Huawei rotating chairman Xu Zhijun said on September 30 that the company’s Ascend series had moved into first place in China’s AI-accelerator market, ahead of Nvidia. The statement is based on Huawei’s internal data rather than an independently published market assessment, and Huawei acknowledged that Nvidia’s precise share in China is difficult to establish.
Huawei said it will prioritize deliveries to customers in China, citing production capacity that is under pressure from domestic demand. The company is positioning Ascend as a domestic alternative as US export restrictions constrain access to imported accelerators and place greater emphasis on supply-chain security for Chinese AI infrastructure operators.
A central element of Huawei’s scale-out strategy is the Atlas 950 SuperCluster, which the company is actively promoting with 256,000 compute cards. The project is intended to support training and inference for models with up to several trillion parameters. Huawei has also indicated that the platform could potentially expand to 1 million NPUs, although this is a future possibility rather than deployed capacity.
For data-center operators, the significance is not limited to accelerator market share. If Chinese AI deployments increasingly move to Ascend, software compatibility beyond CUDA could become more important across the domestic AI stack. Huawei says the software-ecosystem barriers around Ascend have declined over the past 18 months, including as AI-model architectures have changed.
The shift could also favour larger clustered deployments. If operators compensate for lower performance at the individual-accelerator level through device count, demand may rise for supporting AI infrastructure, including servers, high-performance networking, cooling and electrical supply systems. This remains an analytical inference rather than a confirmed procurement outcome.
The supplied source also reports that Nvidia H200 deliveries to China have resumed on a limited basis and account for less than 1% of Nvidia data-center revenue. That figure is not supported by primary financial documentation in the available material.
If Huawei’s assessment reflects actual shipment patterns, China’s AI data-center market could become increasingly distinct from the global accelerator market in both hardware and software platforms. Large-scale Ascend deployments could also increase demand for AI server infrastructure, networking, cooling and power systems.
- TechNews 科技新報 / КитайскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.