WAIC 2026 | Harnets.AI Appears at the Yidian AI Computing Forum as 23 Companies Form an AI Computing System Architecture Alliance
On July 19, the “Next-Generation AI Computing Cloud Industry Technology Innovation Forum” was successfully held at WAIC 2026. Li Dan (李丹), Professor at Tsinghua University's Department of Computer Science and an IEEE Fellow, and Xiong Dian (熊典), founder and chairman of Harnets.AI, were invited to speak about the new ZCube AI-cluster networking architecture. During the forum, the AI Computing System Architecture Alliance was launched with 23 companies across the AI-computing value chain to build an open, collaborative, and independently controllable domestic full-stack AI-computing ecosystem.
01AI Computing System Architecture Alliance Officially Established
Guided by the principle “pool ecosystem strength and build an open architecture,” the alliance was initiated by Shanghai Yidian AI Computing (上海仪电智算) together with 23 upstream and downstream AI-computing companies, including StepFun, MiniMax, Iluvatar CoreX, Biren Technology, T-Head, Cambricon, Metax, Moore Threads, Enflame, ThinkSilicon, ZTE, H3C, Inspur, Suanfeng Information, Sugon, Lightelligence, Harnets.AI, Qingcheng Jizhi, Qulian Technology, Jiliu Technology, SiliconFlow, Yunmai Xinlian, and Yunhe Zhiwang. The alliance deeply integrates full-stack AI-computing ecosystem resources, bringing together hardware and software participants in the Yangtze River Delta and Shanghai to build a collaborative industry ecosystem and broaden industry partnerships.
02Harnets.AI Explains the ZCube AI-Network Architecture
At the WAIC Yidian AI Computing Forum, Professor Li Dan and Harnets.AI founder and chairman Xiong Dian shared research on new AI-cluster networking architectures and systems. They introduced the team's independently developed ZCube architecture, using underlay innovation to address the challenges facing domestic AI-computing clusters.
Today, mainstream AI-computing centers commonly use conventional tree-like, multi-tier Clos fabrics. Their complex forwarding patterns readily create structural link congestion, constraining large-model training and inference performance while requiring extensive switch and optical-transceiver investment. This is a core bottleneck for deploying large-scale, cost-effective domestic AI-computing clusters.
Addressing these common industry challenges, Harnets.AI's core team has focused on innovation in AI-network underlay technology. Its self-developed ZCube architecture breaks through the inherent limits of conventional Clos fabrics. ZCube abandons multi-tier topology for a fabric-wide flat interconnect, avoiding structural congestion in AI-computing traffic at the architectural level and greatly improving GPU-cluster communication efficiency.
ZCube delivers key breakthroughs in both performance and cost: it substantially improves overall AI-cluster throughput and large-model inference experience while reducing the number of required core networking components such as switches and optical transceivers. ZCube deployments show a 33% reduction in network-hardware cost and a 15% increase in inference throughput, offering a new technical path and mature deployment approach for low-cost, high-throughput, highly adaptable, independently controllable domestic AI-computing clusters.
03Advancing AI-Network Underlay Innovation
The WAIC Yidian AI Computing Forum was a concentrated showcase of Harnets.AI's technical capabilities and ecosystem strategy. With large-scale cluster validation at the Shanghai Yidian AI Computing Center, ZCube has completed its journey from theory and laboratory validation to a real 1,000-GPU computing cluster.
Going forward, Harnets.AI will use the AI Computing System Architecture Alliance as an important vehicle to continually iterate and optimize ZCube networking technology, expanding its capabilities for larger-scale, multimodal, and agentic scenarios.
Innovate the computing foundation; advance through ecosystem collaboration. The appearance of ZCube and establishment of the AI Computing System Architecture Alliance at this forum mark a new stage of “unified architecture and co-built ecosystem” for Shanghai's domestic AI-computing industry. Harnets.AI will remain committed to underlay technology innovation and work with every alliance partner through an open architecture to advance the high-quality development of domestic AI-computing infrastructure.