1. Core Enterprise Map
1. Cloud Training / Inference Chips
Cambricon (688256): The “pure AI chip first stock” in A-shares, the Siyuan series chips cover all scenarios from cloud to edge, with the Siyuan 590 (7nm) achieving a computing power of 471 TOPS, performance close to 80% of NVIDIA’s A100, with revenue of 1.7 billion yuan (+13 times) in Q3 2025, and a net profit of 570 million yuan, turning losses into profits.
Haiguang Information (688041): A leader in DCU (similar to GPU), deeply tied to Alibaba, with the Shensuan No. 3 entering ByteDance’s recommendation algorithm, a single order worth 980 million yuan; revenue for the first three quarters reached 9.49 billion yuan (+54.65%), with a net profit of 1.961 billion yuan (+28.56%).
2. Edge AI Chips
Rockchip (603893): A leader in AIoT chips, with the RK3588 holding a 37% market share in AR/VR devices, and the edge computing co-processor RK182X solving model deployment challenges.
Allwinner Technology (300458): A leader in smart terminal ASICs, with low-power AI chips adapted for IoT scenarios.
Juchip Technology (688049): A leader in edge AI audio chips, with increased penetration in the high-end speaker market, achieving record high revenue.
3. Other Core Enterprises
Lanqi Technology (688008): A leader in memory interface chips, benefiting from AI server demand with DDR5 products, with a net profit increase of 135% in Q1.
ZTE Corporation (000063): Synergy between communication and AI chips, with revenue up 13.13% in the first three quarters and net profit up 25.58%.
2. Core Competitiveness Perspective
1. Technological Breakthroughs
Architectural Innovation: Cambricon MLU, Haiguang DCU, and Birun BR100 (with computing power three times that of A100) all adopt self-developed architectures.
Process Breakthrough: Domestic 7nm chip mass production yield reaches 92%, and SMIC’s 14nm process meets 70% of domestic GPU demand.
Energy Efficiency Advantage: Cambricon Siyuan 590 has an energy efficiency ratio of 30 TOPS/W, while the energy efficiency ratio of the integrated storage and computing chip from Apple Core Technology is 27 TOPS/W (four times the industry average).
2. Acceleration of Domestic Substitution
NVIDIA’s market share in China has dropped from 95% in 2022 to 54% in 2025, while Huawei’s Ascend has increased from 23% to 28%, with domestic chips like Cambricon accounting for a total market share of 18% (training) and 25% (inference).
Strong Policy Support: National funding projects must phase out overseas AI chips, and new projects must be 100% compatible with domestic products.
Internet Giants “Switching Chips”: Alibaba and Tencent Cloud have initiated domestic substitution, with a compatibility rate of 98% between domestic chips and mainstream AI frameworks.
3. Performance Explosion and Valuation Bubble
1. Performance Surge
Cambricon: Revenue in the first half of 2025 is expected to reach 2.881 billion yuan (+2386%), with government smart computing center projects accounting for 65% of orders.
Haiguang Information: Rapid growth in DCU orders, with government sector orders accounting for 65%.
Edge AI chip companies: Juchip Technology, Rockchip, and Tailin Microelectronics have seen revenue growth of 50%-200% year-on-year.
2. Valuation and Risk Warning
Overvaluation Concerns: Cambricon’s price-to-earnings ratio is as high as 4000 times, with a market value once exceeding 580 billion yuan, which is severely mismatched with net profit.
Bubble Characteristics: The sales-to-market ratio of the AI computing power sector exceeds 30 times, approaching the peak of the 2015 GEM bubble.
Technological Gap: NVIDIA’s Blackwell series FP16 computing power is expected to exceed 3 PetaFLOPS, while domestic chips still have generational gaps.
Supply Chain Risks: The Fabless model relies on external foundries, and some companies have been placed on the “entity list”.
4. Investment Value Assessment
1. Investment Logic
Domestic substitution is a necessity: With high-end chips being restricted, domestic substitution is accelerating in key areas such as government, telecommunications, and energy.
Accelerated Scene Implementation:
Cloud: Cambricon’s Siyuan series and Haiguang DCU have been deployed in leading internet companies.
Edge: The penetration rate of AIoT, automotive, and security chips is increasing, with Juchip Technology’s audio chips and Bowatech’s AI cameras achieving large-scale applications.
Ecosystem Building: Huawei’s Ascend, Cambricon, and others are building independent software ecosystems, deeply adapting to domestic large models.
2. Risk Warnings
Valuation Reversion Risk: High valuations depend on sustained high growth in performance; if expectations are not met, there may be significant corrections.
Intensified Competition: International giants have technological advantages, AMD has obtained export licenses to China, and local companies face dual pressure.
Profitability Uncertainty: Some companies are still in losses, such as Moore Threads, which has invested 8 billion and is slow in commercialization progress.
5. Investment Strategy Recommendations
1. Focus on Four Types of Enterprises
Technologically and Performance Superior: Cambricon, Haiguang Information (already profitable, technologically advanced) Shanghai Securities Journal.
Edge Application Leaders: Rockchip, Allwinner Technology (fast scene implementation, stable performance growth).
Core of the Industry Chain: Lanqi Technology (memory interface, essential for AI servers).
Technological Breakthrough Types: Birun Technology, Muxi Co., Ltd. (high-performance GPUs, key for future domestic substitution).
2. Operational Strategies
Gradual Position Building: Avoid full positions in high volatility, focus on quarterly performance realization.
Focus on Scenarios: Prioritize layout in chip companies that have achieved large-scale applications in government, internet, and intelligent driving fields.
Pay Attention to Valuation Safety Margins: Be cautious with PE > 100 times, focusing on PS, order growth, and other leading indicators.
6. Outlook
AI + chips have become a national strategic focus, with the domestic AI chip market expected to reach $40 billion by 2025, and the domestic share of training/inference chips expected to reach 35%/45% by 2030.
Conclusion: Core AI + chip companies in A-shares are currently in a period of “accelerated domestic substitution + performance explosion + valuation bubble” overlapping. Investment needs to grasp: ① technical strength and product landing verification; ② sustainability of performance growth; ③ rationality of valuation. It is recommended to adopt a strategy of “selecting individual stocks, gradual layout, long-term holding, and wave operation”, focusing on leading companies like Cambricon, Haiguang Information, and Rockchip that have achieved technological breakthroughs and performance realization, while being alert to valuation bubble risks.