At the Huawei Connect conference in Shanghai, rotating chairman Wang Tao outlined the company's AI strategy, which prioritizes computing power and hardware monetization. The strategy includes developing cluster-based computing infrastructure, fostering an open-source ecosystem for large models, and integrating the Pangu model into Huawei's product line.
Huawei has already deployed over 1,000 Ascend super-nodes, with the Ascend 950 model currently in large-scale commercial use. The company is also advancing its hardware roadmap with the upcoming Ascend 960 super-node, which will be the industry's first to utilize NPO technology.
The Atlas 860 air-cooled super-node is scheduled for release in the second quarter of 2027, followed by the Atlas 960 liquid-cooled version in the third quarter of 2027. These developments follow the introduction of the Ascend 950 architecture, which features enhanced vector computing, increased interconnect bandwidth, and proprietary HBM technology.
The Ascend 950 series includes the 950PR model, designed to boost inference prefill performance, and the 950DT model, which focuses on improving decoding and training capabilities. Huawei continues to emphasize the integration of these computing resources into communication networks to support broader AI adoption.
