Hygon Showcases Intelligent Inspection Robot and Other Applications, Accelerating the Development of Domestic AI Computing Infrastructure

Bot Telegraph July 21st report, the Guanghe Organization 2026 Intelligent Computing Application Conference was recently held in Zhengzhou. At the conference, Hygon Information Technology showcased multiple application achievements, including intelligent inspection robots, AI BOX edge-side integrated machines, and closed-scenario industrial autonomous driving solutions, covering typical scenarios such as industrial control, edge-side physical AI inference, and intelligent devices.

Du Xiawei, Assistant to the President and General Manager of the Intelligent Computing Product Department at Hygon Information Technology, stated that with the rapid expansion of the AI industry, the proportion of demand related to large models continues to increase. In the development of new-generation chips, Hygon has moderately tilted hardware computing resources towards AI computing units while fully retaining full-precision computing capabilities. This balances the needs of existing scientific computing customers and helps traditional research users accelerate the AI transformation of their businesses.

Currently, Hygon’s DCU has completed adaptation and optimization for multiple mainstream large models, covering model ecosystems such as GLM, DeepSeek, MiniMax, and Kimi, and continues to promote efficiency improvements in training, inference, and cluster management.

In terms of software architecture, Hygon employs a three-layer standardized software architecture to assist users in AI transformation: the first layer is the DTK basic software stack, responsible for standardizing and encapsulating the underlying hardware computing power; the second layer is the DAS AI-specific software layer, achieving efficient adaptation between large models and hardware computing power; the third layer is the DAP business platform layer, providing customers with a lightweight business-oriented operation interface.

Du Xiawei revealed that this year, the platform’s iteration focus has shifted to AI Agents, with the launch of the DAPClaw intelligent agent component. It outputs capabilities in a standardized Token format, helping enterprise customers quickly build intelligent agent applications in private environments.

Source: Bot Telegraph China