On July 9, the Guanhe Organization 2026 Intelligent Computing Application Conference opened in Zhengzhou. China's first fully domestic 100,000-card AI supercluster, Sugon 8000 (Dengfeng), was officially completed. The system is equipped with a computing power base of domestic chips such as Hygon, verifying Hygon chips' capability to support large-scale Token production and industrial-grade AI applications.
Meanwhile, Sugon Information Industry unveiled its complete "Cloud-Edge-Device" computing power system for the first time, bringing its full-scenario AI computing layout from data centers to edge computing and physical device sides to the surface.
CPU+DCU Dual-Chip Strategy
Hygon possesses China's only "C86+GPGPU" self-developed product matrix. The CPU is compatible with the x86 architecture, allowing existing industrial software, configuration tools, and business systems on Windows and Linux platforms to run natively. The DCU is based on the GPGPU route, adopting a general-purpose parallel computing architecture that can well adapt to mainstream international commercial computing software and AI (Artificial Intelligence) software. Currently, Hygon DCU has completed adaptation and optimization for multiple mainstream large models such as GLM, DeepSeek, MiniMax, and Kimi, covering 99% of non-closed-source large models globally, and has gathered over 6,000 ecosystem partners relying on the Guanhe Organization.
As tokens evolve, AI's technical capabilities and products are increasingly integrating with diverse business scenarios. Especially in the directions of AI for Science and AI for Engineering, there is an inherent native demand for high precision. Hygon's full-precision route can not only release the computing power of new AI features but also retain support for traditional numerical computing, achieving 100% business process coverage.
Currently, Hygon DCU can support full-scenario needs from billion-level model inference to hundred-billion-level model training. The full-precision route retains the maximum possibilities in the face of market changes and technological iterations.
Perfecting the DCU Software Stack: Domestic Computing Power Moves from "Usable" to "User-Friendly"
Du Xiawei, President Assistant and General Manager of the Intelligent Computing Product Department at Sugon Information Industry, delivered a speech at the conference focusing on "Full-Stack Software Collaboration, Empowering Industry Tokenization." He emphasized, "The industrialization of AI cannot only focus on chip computing power, but more importantly, whether the computing power is user-friendly."
Photo: Du Xiawei, President Assistant and General Manager of the Intelligent Computing Product Department, Sugon Information Industry
As AI applications evolve from Q&A and generation to intelligent agents and complex task processing, the scale of Token invocations has rapidly increased. What enterprises truly need is a computing power platform that can be rapidly deployed, operate stably, and continuously reduce costs.
Centered on this demand, Hygon continues to perfect its DCU software stack. Sugon Information Industry has built a three-layer DCU software stack comprising DTK, DAS, and DAP. The DTK basic layer keeps in step with the chip side, completing support for new architectures and instruction sets. The DAS AI layer advances with rapid, iterative steps, quickly adapting to the latest inference engines such as vLLM, SGLang, and Tailor. The DAP platform layer is business-demand oriented.
In Du Xiawei's view, the core of the software layer lies in encapsulating computing power into underlying fundamental capabilities.
Du Xiawei stated, "You cannot communicate with a business expert using computational language, because while they are discussing their engineering implementation, you are talking about your FLOPS computing power; the two can never be on the same wavelength. Users only need to invoke the capabilities without worrying about the underlying precision. This 'business-oriented encapsulation' approach is precisely the critical leap for domestic computing power from being 'usable' to 'user-friendly'."
Multi-Dimensional Advantages Empowering the New "Cloud-Edge-Device" Battlefield
As AI applications accelerate their extension into industry, energy, transportation, robotics, and smart terminals, Hygon is actively deploying embedded native AI technologies. Through new technologies such as AI and intrinsic security, it upgrades and transforms traditional industrial control scenarios like oil and gas, rail transit, security, ports, and aviation, promoting better adaptation of AI capabilities to device-side and edge-side application scenarios.
The competitiveness of this layout lies in its triple advantages:
Compatibility: The C86 architecture enables the migration of legacy software at almost zero cost. Developers do not need to relearn instruction sets and toolchains, making it the optimal solution for replacing traditional industrial control computers.
Inference Capability: The GPGPU many-core SIMT architecture has obvious advantages in parallel data processing and floating-point computing power. It is compatible with mainstream frameworks such as PyTorch and TensorFlow, and the algorithm migration and deployment processes are mature.
Intrinsic Security: Hygon CPUs support national cryptographic algorithms, trusted computing, and privacy computing from the underlying architecture. Without the need for external security devices, it achieves native security capabilities from the "chip root to the device."
The "Cloud-Edge-Device" system means providing an integrated solution from data centers to production sites. This capability can significantly reduce customers' system migration and deployment costs, improve the implementation efficiency of domestic substitution solutions, thereby accelerating the penetration rate of domestic computing power in these key industries. From a business logic perspective, this is an upgrade from selling "chips" to selling "computing power services" and "solutions." A complete "Cloud-Edge-Device" full-stack service has stronger attractiveness and bargaining power for customers, enabling device-side solutions to quickly adapt to industry applications and shortening project delivery cycles.
The Flywheel is Spinning
Currently, Hygon's "Cloud-Edge-Device" panoramic computing power system has taken shape. Relying on the Guanhe Organization ecosystem, Hygon will further unite partners in basic software, databases, complete machines, and industry applications to promote the deep integration of AI computing power with industrial scenarios, providing underlying support for the intelligent transformation of the real economy.