en.Wedoany.com Reported - Ascend will showcase its latest full-stack AI capabilities at the 2026 World Artificial Intelligence Conference, including the first public display of the industry's largest-scale supernode, the Atlas 950 SuperPoD.

The rapid development of agent and AI programming technologies is placing higher demands on underlying large models and computing infrastructure. The core challenge facing current computing infrastructure is to support larger-scale MoE models, longer context sequences, and lower inference latency, while significantly reducing inference costs and improving computing efficiency. Supernodes are seen as a key solution to this challenge and have become the mainstream form of computing infrastructure in the Agentic AI era.

In July 2026, the annual global AI event—the 2026 World Artificial Intelligence Conference—will be held in Shanghai. Under the theme "Together with the Era, Ascend Together," Ascend will showcase its latest products, solutions, industry practices, and customer cases at Booth A101, Hall H2, Shanghai World Expo Exhibition and Convention Center.

At this conference, the Ascend exhibition area will demonstrate Ascend's full-stack AI capabilities from four aspects: hardware computing base, open software ecosystem, technological breakthroughs, and industrial deployment.

As a key highlight of the exhibition area, the Ascend 950 SuperPoD will make its public debut in its physical form as the industry's largest-scale supernode, marking new progress for Ascend in the field of ultra-large-scale computing infrastructure.

In addition to the flagship 950 series, Ascend will also showcase the physical Atlas 850E air-cooled supernode, which requires no liquid cooling retrofitting and can deliver supernode performance in standard air-cooled data centers.

In terms of the software ecosystem, Ascend AI basic software will showcase the latest progress in comprehensive open-source and open-access initiatives. The CANN heterogeneous computing architecture continues to be open-sourced, with deep compatibility for third-party acceleration libraries and development tools. The end-to-end experience from operator programming to inference deployment has been upgraded, driving development from "usable" to "easy-to-use." At the same time, Ascend fully embraces mainstream open-source ecosystems to continuously improve large model training and inference efficiency.

In terms of industry applications, the exhibition area will highlight the large-scale deployment results of Ascend supernodes across 11 core industries, including solutions and best practices, as well as benchmark cases in multiple fields. Additionally, training-inference integrated machine solutions, open-source ecosystems, and industry practices will also be showcased, allowing visitors to engage in on-site discussions with Ascend experts.