Analysis of the AI Chip Industry Chain

This is purely an analysis of the relevant industry, and no specific stocks are recommended.

The following is an analysis of the core segments of the AI chip industry chain and related stocks based on the latest industry dynamics in 2025, covering subfields such as chip design, edge chips, edge storage, servers, and EDA:

1. Chip Design: The Core Battlefield for Domestic Substitution

1. Cloud Training Chips

  • Cambricon (688256): A leading domestic AI chip company, the Siyuan 590 chip has a computing power of 256 TOPS (FP16), suitable for training large models with hundreds of billions of parameters. In Q1 2025, new orders exceeded 1.5 billion yuan, with government smart computing center procurement accounting for over 40%. The technical advantage lies in the integrated ecosystem of cloud and edge, but attention should be paid to the progress of narrowing losses.
  • Hygon Information (688041): A dual leader in domestic x86 architecture CPU/DCU, the DCU Deep Computing No. 2 performance reaches 80%-90% of NVIDIA A100, with a market share in government cloud exceeding 40%. After merging with Zhongke Shuguang, it forms a “chip-machine” vertical closed loop, with contract liabilities reaching 3.2 billion yuan in Q1 2025, indicating high certainty for domestic substitution.
  • Jingjia Micro (300474): A leading high-reliability GPU chip company, the JM9 series supports 4K rendering and AI computing, with a domestic rate of over 90% in military and security applications, and civilian orders expected to double in H1 2025, entering automotive and edge AI scenarios.

2. Edge Inference Chips

  • Rockchip (603893): A leading edge AI chip company, with a year-on-year revenue growth of 62.95% in Q1 2025, edge inference efficiency improved by 40%, benefiting from open-source models like DeepSeek driving the AIoT explosion, with significant growth in automotive electronics and machine vision fields.
  • Allwinner Technology (300458): A leading RISC-V architecture processor company, with a year-on-year net profit increase of 979.75% in Q1 2025, products applied in smart home and automotive electronics, with industry-leading cost control capabilities.
  • Espressif Technology (688018): A low-power Wi-Fi/Bluetooth SoC chip manufacturer, with a gross margin of 43.37% in Q1 2025, suitable for IoT edge devices, supporting multimodal interaction.

2. Edge Storage: High Bandwidth and Low Power Demand

1. Storage Chips

  • Lanqi Technology (688008): Among the top three global memory interface chip manufacturers, DDR5 technology increases bandwidth by 50%, with a year-on-year net profit increase of 153.4% in Q3 2025, suitable for AI server high concurrency demands.
  • Biwei Storage (688525): Focused on HBM and storage-computing integration technology, with a year-on-year revenue increase of 62.64% in Q3 2025, stacking technology breaks the single-chip capacity of 1TB.
  • GigaDevice (603986): Among the top three in global NOR Flash market share, products used for edge device storage, with significant stock price increase in 2025, but risks of price wars should be monitored.

2. Storage Solutions

  • Jiangbolong (301308): Integrated storage module and chip packaging, products have high reliability, applied in smart terminals and automotive devices.

3. AI Servers: The Hardware Carrier for Computing Power

1. Complete Machine Manufacturing

  • Inspur Information (000977): The global leader in AI server market share (over 45%), with liquid cooling technology achieving a PUE as low as 1.05, with ByteDance and Alibaba’s orders exceeding 10 billion yuan scheduled for production until 2026. Risks include reliance on NVIDIA chip supply interruptions.
  • Zhongke Shuguang (603019): Liquid cooling server market share exceeds 60%, with immersion solutions achieving a PUE of 1.08, participating in the construction of 80% of national smart computing centers, with order growth of 120% after merging with Hygon Information, and clear policy benefits.
  • Foxconn Industrial Internet (601138): Core foundry for NVIDIA H100, with a yield rate of 99.99%, AI server revenue increased by 230% year-on-year, deeply bound to North American cloud vendors.

2. Liquid Cooling Technology

  • Invid (002837): Dual technology route of cold plates and immersion, leading in bidding shares for Tencent and Alibaba, with a year-on-year increase of 456.88% in non-recurring net profit in Q1 2025.

4. EDA/IP: The Underlying Support for Chip Design

1. EDA Tools

  • Galen Electronics (688206): A leading domestic EDA company, focusing on device modeling and simulation, supporting designs below 5nm process.
  • Huada Empyrean (301269): Full-process coverage of analog circuit EDA, participating in the construction of the domestic substitution ecosystem.

2. IP Licensing

  • Chipone Technology (688521): A leading domestic IP licensing company, NPU IP used for 5nm chips, with IP reuse rate doubling in the Chiplet era, with a stock price increase of 48.61% in 2025.

5. Emerging Technology Directions

1. Chiplet Technology

  • Changdian Technology (600584): 3D packaging yield rate reaches 98%, providing Chiplet testing for Cambricon.
  • Tongfu Microelectronics (002156): Bound to AMD Chiplet technology, with a year-on-year revenue increase of 45% in H1 2025.

2. Optical Communication and Networks

  • Inspur (300308): Global optical module market share exceeds 45%, with 1.6T silicon optical modules sent to NVIDIA, with a net profit growth expectation of over 45% in 2025.

6. Comparison of Core Enterprises and Risk Alerts

Segment Representative Enterprises Technical Advantages Risk Alerts
Cloud Chips Cambricon, Hygon Information Improved computing power density, depth of domestic substitution Loss pressure, technology iteration
Edge Chips Rockchip, Allwinner Technology Low power design, edge inference efficiency Homogenization competition, gross margin fluctuations
Edge Storage Lanqi Technology, Biwei Storage HBM technology, high bandwidth support Price wars, overcapacity
Servers Zhongke Shuguang, Inspur Information Liquid cooling technology, vertical integration capability Geopolitical risks, supply chain dependence
EDA/IP Chipone Technology, Galen Electronics Chiplet IP reuse, maturity of domestic toolchain Ecological barriers, suppression by international giants

Risks and Trends:

  • Policy Driven: The State Council’s “AI+” action promotes the domestic substitution rate target to 50%, with surging demand for liquid cooling and optical modules.
  • Technical Risks: U.S. chip restrictions (40% probability) impact the manufacturing side but accelerate the replacement of domestic equipment; Chiplet and storage-computing integration technology may break through computing power bottlenecks.
  • Market Opportunities: Edge computing (Rockchip), inference chips (Cambricon), and computing power scheduling (Zhongke Shuguang) are key directions for the second half of 2025.

Conclusion: The domestic AI chip industry chain has entered a “dual drive stage of orders and profitability”. It is recommended to focus on “depth of domestic substitution + technical barriers” high targets (such as Hygon Information, Zhongke Shuguang, Rockchip), while also paying attention to the flexible opportunities in computing infrastructure and edge computing driven by policy catalysts.

Leave a Comment