Huawei to Unveil Atlas 950 SuperPoD — the Industry's Largest-Scale Super Node — at WAIC 2026

Huawei China announced today that it will present its latest Ascend AI products, solutions, and industry deployments at the 2026 World Artificial Intelligence Conference (WAIC 2026), scheduled to run from July 17 to July 20. The showcase will be headlined by the first-ever public hardware demonstration of the Atlas 950 SuperPoD, which the company describes as the industry’s largest-scale super node.

IT-NEWS has learned that Huawei’s WAIC 2026 exhibition will revolve around four key highlights. The centerpiece is the Atlas 950 SuperPoD, making its physical debut after being announced in March of this year. Built on a single-cabinet, 64-card base unit, the system scales up to 8,192 NPU cards interconnected at high speed through innovations spanning base components, protocol algorithms, and optoelectronic technologies. It features global unified memory addressing designed to accelerate both training and inference for trillion-parameter large models.

Alongside the flagship, Huawei will also show the Atlas 850E air-cooled super node, which brings super-node capabilities to standard air-cooled data-center environments. Upgraded VCE thermal technology provides extreme stability while delivering high throughput and low latency tailored for agentic inference workloads.

On the software side, Huawei plans to highlight the open-source and open ecosystem around Ascend AI foundational software, which now offers deep compatibility with third-party training acceleration libraries, inference engines, and programming languages. The company says the focus has shifted from functional adequacy to ease of use, with continued efficiency gains for mainstream large-model training and inference.

Rounding out the exhibition, Huawei will demonstrate how its super-node architecture is already deployed at scale across 11 major industries — including internet services, telecommunications, finance, and energy — with real-world best practices and integrated solutions on display.

Compared to traditional clusters, the Atlas 950 SuperPoD delivers substantial improvements in training efficiency, reliability, and inference performance, positioning it as a compelling option for organizations operating at the frontier of large-scale AI workloads.