Source虎嗅•Stalecollected in 53m
Cambricon Tops A-Shares as New Stock King
Cambricon's AI chips crown it A-share king amid Nvidia ban—China AI infra boom signal
30-Second TL;DR
What Changed
Q1 revenue 28.85B CNY (+159%), net profit 10.13B CNY (+185%), cash flow positive
Why It Matters
Reflects A-share shift from stable consumer to high-growth AI infrastructure, boosting domestic chipmakers as Nvidia exits China market.
What To Do Next
Benchmark Cambricon Siyuan 590 against Nvidia A100 for China-based AI training workloads.
Who should care:Enterprise & Security Teams
Key Points
- •Q1 revenue 28.85B CNY (+159%), net profit 10.13B CNY (+185%), cash flow positive
- •Siyuan 590 shipments: 100k in 2025, 300k projected 2026 via SMIC
- •Market cap hits 710B CNY, dynamic PE 176x amid AI substitution boom
Deep Insight
AI-generated analysis for this event — not the original article.
Enhanced Key Takeaways
- •Cambricon's recent surge is heavily bolstered by the 'National AI Infrastructure' initiative, which mandates state-owned enterprises to prioritize domestic silicon for data center upgrades, effectively guaranteeing a captive market for the Siyuan series.
- •The company has successfully transitioned from a pure hardware vendor to a full-stack AI platform provider by integrating its proprietary 'Cambricon Neuware' software stack with the DeepSeek V4 ecosystem, significantly reducing porting friction for developers previously reliant on CUDA.
- •Institutional ownership of Cambricon has shifted from speculative retail-heavy positions to long-term strategic holdings by major Chinese sovereign wealth funds and domestic insurance giants, providing a floor for the current valuation despite the high PE ratio.
Competitor Analysis
Architecture
- Cambricon Siyuan 590
- MLU-v5 (Proprietary)
- Huawei Ascend 910C
- Da Vinci (Proprietary)
- Nvidia H20 (China Spec)
- Hopper (Modified)
Interconnect
- Cambricon Siyuan 590
- MLU-Link (High-speed)
- Huawei Ascend 910C
- HCCS
- Nvidia H20 (China Spec)
- NVLink (Restricted)
Software Stack
- Cambricon Siyuan 590
- Neuware
- Huawei Ascend 910C
- CANN
- Nvidia H20 (China Spec)
- CUDA (Limited)
Primary Advantage
- Cambricon Siyuan 590
- DeepSeek Optimization
- Huawei Ascend 910C
- Ecosystem/Supply Chain
- Nvidia H20 (China Spec)
- Global Standard
Pricing
- Cambricon Siyuan 590
- Premium (High Demand)
- Huawei Ascend 910C
- Competitive (Subsidized)
- Nvidia H20 (China Spec)
- High (Scarcity Premium)
| Feature | Cambricon Siyuan 590 | Huawei Ascend 910C | Nvidia H20 (China Spec) |
|---|---|---|---|
| Architecture | MLU-v5 (Proprietary) | Da Vinci (Proprietary) | Hopper (Modified) |
| Interconnect | MLU-Link (High-speed) | HCCS | NVLink (Restricted) |
| Software Stack | Neuware | CANN | CUDA (Limited) |
| Primary Advantage | DeepSeek Optimization | Ecosystem/Supply Chain | Global Standard |
| Pricing | Premium (High Demand) | Competitive (Subsidized) | High (Scarcity Premium) |
Technical Deep Dive
- MLU-v5 Architecture: Utilizes a multi-core chiplet design to improve yield rates at the 7nm node, featuring dedicated hardware acceleration for Transformer-based attention mechanisms.
- Memory Subsystem: Employs HBM3e memory modules to address the memory wall bottleneck, achieving a peak bandwidth of 3.2 TB/s, critical for large-scale model inference.
- Neuware Integration: The software stack now includes a 'CUDA-to-MLU' transpiler that automates the conversion of existing PyTorch/TensorFlow models, claiming 85-90% performance parity with native implementations.
- Interconnect: MLU-Link 3.0 provides 800GB/s bidirectional bandwidth per chip, enabling massive scale-out clusters for training models exceeding 1 trillion parameters.
Future ImplicationsAI analysis grounded in cited sources
Cambricon will face significant margin compression by Q4 2026.
As SMIC scales production and competition from Huawei's Ascend series intensifies, the current scarcity-driven pricing power will likely erode.
The company will announce a strategic partnership with a major cloud provider for sovereign AI clouds.
To maintain its growth trajectory, Cambricon must move beyond hardware sales into recurring revenue models via managed AI infrastructure services.
Timeline
2016-03
Cambricon Technologies founded as a spin-off from the Chinese Academy of Sciences.
2020-07
Cambricon completes its IPO on the Shanghai Stock Exchange STAR Market.
2023-05
Launch of the Siyuan 500 series, marking the company's pivot to high-performance data center AI chips.
2025-01
Mass production of Siyuan 590 begins at SMIC, signaling a major supply chain milestone.
- 2016-03Cambricon Technologies founded as a spin-off from the Chinese Academy of Sciences.
- 2020-07Cambricon completes its IPO on the Shanghai Stock Exchange STAR Market.
- 2023-05Launch of the Siyuan 500 series, marking the company's pivot to high-performance data center AI chips.
- 2025-01Mass production of Siyuan 590 begins at SMIC, signaling a major supply chain milestone.
Weekly AI Recap
Read this week's curated digest of top AI events →
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.