Hygon Information Technology Co., Ltd.
Development and sales of x86-based general-purpose CPUs and AI-oriented DCU co-processors
x86 instruction set compatibility + independent iteration capability, enabling seamless adoption of the mature software ecosystem
Company Overview
Founded in 2014, Hygon Information Technology was listed on the STAR Market in August 2022. Headquartered in Tianjin, the company is primarily engaged in the R&D and sales of high-end processors (CPUs) and coprocessors (DCUs) targeting artificial intelligence and high-performance computing. Within the AI industry chain, Hygon occupies the core computing and communication hardware layer, making it the player with the strongest "universal compatibility" in the second tier of domestic computing power chip vendors.
In its early days, Hygon Information Technology rapidly established independent iteration capabilities by introducing AMD's x86 Zen architecture license, and subsequently developed Hygon CPUs compatible with the x86 instruction set. Since 2023, the DCU (Deep Computing Unit) product line has accelerated its volume ramp-up, becoming one of the primary domestic substitute forces in AI inference/training scenarios.
Hygon CPUs are the only high-end processor series in China based on the x86 general-purpose architecture with independent iterative evolution, covering the 7000 series (cloud), 5000 series (edge), and 3000 series (desktop/embedded). Primary customers include government, finance, telecommunications, education, and other Xinchuang (IT application innovation) sectors, with further penetration into public cloud and enterprise data centers. In 2023, CPU business revenue reached approximately RMB 5.1 billion, with a gross margin of approximately 66%.
Core advantages: Native compatibility with the x86 ecosystem, enabling Windows/Linux and various enterprise-grade applications to run without software adaptation.
The DCU (Deep Computing Unit) adopts a CUDA-like general-purpose computing architecture, primarily targeting high-parallelism scenarios such as AI training/inference and scientific computing. Representative products include the Deep Computing 8000 (first generation) and Deep Computing 9000 (second generation, mass production in 2024). The DCU has been successfully adapted to mainstream AI frameworks including DeepSeek, Baidu PaddlePaddle, and MindSpore. In 2023, DCU revenue reached approximately RMB 900 million, with year-over-year growth exceeding 300%, and is expected to account for over 25% of revenue by 2024.
| Product Line | Revenue Share | Core Customers | Gross Margin |
|---|---|---|---|
| Hygon CPU (7000/5000/3000) | ~85% | Server manufacturers such as Inspur, H3C, and Lenovo; Xinchuang system integrators | ~66% |
| Hygon DCU (Deep Computing series) | ~15% | Telecom operators, financial institutions, intelligent computing center operators | ~55% (initial stage) |
File: database/companies/Hygon.md
Technology Moat
Moat 1: x86 Instruction Set Compatibility + Proprietary Microarchitecture Iteration
Through legal licensing, Hygon has secured perpetual rights to use and independently evolve the x86 instruction set, and has developed multiple generations of proprietary microarchitectures (e.g., "Chanmu," "Tianqu") on this foundation — enabling it to run the vast installed base of existing x86 software while avoiding the "comprador" label. Compared with domestic CPUs pursuing the ARM/RISC-V roadmap, Hygon holds an overwhelming advantage in application compatibility and migration costs.
Moat 2: Chiplet Advanced Packaging for Cost Reduction and Efficiency Gains
Against the backdrop of partial restrictions on advanced process nodes (7nm/5nm), Hygon was the first to adopt Chiplet (multi-die) technology, integrating functional modules manufactured on different process nodes (compute dies, I/O dies) into a single package via high-speed Die-to-Die interconnects. This not only reduces dependence on advanced process technology but also improves yield and flexibility. For instance, the Hygon 7000-series CPU adopts an architecture of four compute dies plus one I/O die, functionally equivalent to a monolithic large-die chip.
Moat 3: Full-Fledged Adaptation to the Domestic Large Model Ecosystem
Recently, domestic computing power vendors — including Moore Threads, Iluvatar CoreX, and Hygon Information — collectively announced support for open-source large models such as DeepSeek. Leveraging its HCCL high-performance communication library and native PyTorch support, Hygon's DCU has emerged as one of the preferred hardware platforms for domestic large-model inference. A software-hardware synergistic ecosystem barrier is taking shape.
In the domestic server CPU market, Hygon Information Technology holds approximately 20% market share by leveraging its x86 compatibility advantage, ranking first among domestic manufacturers; closely followed by Kunpeng (ARM, ~15%) and Phytium (ARM, ~10%). Globally, Intel and AMD together account for over 95% of the server CPU market, with Hygon still in the catch-up phase.
In the AI accelerator market, NVIDIA holds an 80%+ global share; among domestic players, Huawei Ascend leads with its full-stack CANN software advantage, while Hygon DCU competes with Cambricon, Biren, and others for the second tier.
| Dimension | Data |
|---|---|
| Domestic server CPU market share | ~20% (2023, by shipment volume) |
| Global server CPU industry ranking | 5th (after Intel, AMD, Kunpeng, and Samsung) |
| Major competitors | Intel, AMD (global); Huawei, Cambricon, Zhaoxin, Phytium (domestic) |
| Downstream customers | Inspur Information, H3C, Lenovo, ZTE, Xinchuang system integrators |
Financials & Growth
| Metric | Data |
|---|---|
| Revenue (2023A) | RMB 6.012 billion (+17.3% YoY) |
| Gross Margin | 64.8% (2023) |
| Net Margin | 21.0% (2023) |
| R&D Investment Ratio | 28.5% (2023) |
| Core Growth Drivers | ① Acceleration of IT innovation substitution (government agencies + eight key industries); ② Localization of AI computing power; ③ Rapid ramp-up of DCU shipments |
In Q1 2024, revenue reached RMB 1.592 billion (+47.5% YoY), with net profit of RMB 289 million (+59.2% YoY), as DCU contribution grew rapidly. Full-year 2024 revenue is projected to exceed RMB 8.5 billion.
Key Risks:
- Advanced Process Node Constraints: Although Chiplet technology can alleviate some pressure, processes below 7nm still rely on TSMC and SMIC. If the U.S. further tightens Entity List restrictions, it may impact production capacity supply.
- Intensifying Ecosystem Competition: Huawei Ascend, leveraging its self-developed CANN and framework ecosystem, has established strong stickiness in AI training scenarios; x86 solutions have advantages in AI inference scenarios, but the training side still needs to overcome equivalent cuDNN-level optimization.
- Product Iteration Pressure: Intel/AMD update their architectures every 1-2 years. Hygon needs to sustain high R&D investment (R&D expenses of RMB 1.7 billion in 2023) to narrow the generational gap.
Core Investment Logic / Industry Value Summary:
Hygon Information Technology is a scarce asset in the domestic computing power chip sector, driven by the dual engines of "general-purpose compatibility + AI acceleration." It directly captures the hundreds-of-billions-level replacement market for IT infrastructure in the domestic alternative ecosystem (CPU), while leveraging DCU to seize incremental opportunities from intelligent computing center construction. Despite challenges in process nodes and ecosystem, its "low-cost migration" moat formed by x86 compatibility is difficult to replace in the short term, positioning it as a key player in the hardware self-sufficiency drive of the AI industry chain.