Company at a glance
Scale
Revenue up ≈9× since FY2023
Growth prospects
Revenue trajectory FY2023–2025
China competition
Competes with Huawei Ascend, Biren, Moore Threads, Enflame in AI accelerators — peers listed at sector level.
SubstituteGlobal comparables
Global reference set: Nvidia, AMD, Google TPU.
Nvidia is the unavoidable benchmark, but AMD Instinct, Google TPU, AWS Trainium/Inferentia, and other accelerator programs are often more precise com…Technical gap
≈3× behind the global best on compute-per-dollar (MLU370-X8 vs AMD Instinct MI300X).
Obstacles
Sector bottleneck applies
CUDA-like software ecosystem, HBM supply, advanced packaging, leading-node access, and large-scale deployment proof.Official filings: Cambricon investor filings, STAR Market 688256 ↗
All sourced facts (5)
- Founded: Founded in 2016.
- Headquarters: Beijing.
- Listing: Shanghai STAR Market 688256.
- Core segment: AI accelerator chips, boards/cards, systems, and software stack.
- Public proxy: Standalone listed proxy for Chinese AI accelerator commercialization outside Huawei.
Tracked chips: MLU370-X8 · MLU590 (思元590)
About
Designs AI accelerator chips, cards, and related software for cloud, edge, and inference workloads.
Founded 2016
Why it matters
Cambricon is China's flagship listed AI-accelerator designer and the onshore market's primary proxy for domestic AI-compute independence. The test that matters is whether its MLU accelerator family can capture volume training and inference workloads at Chinese cloud and AI labs, rather than serving only as a policy-driven substitute label. Sustained revenue scale and named design wins would confirm real adoption; persistent losses and dependence on a narrow set of government or state-owned buyers would break the thesis.
Where it sits in the stack
Key semiconductor products
- MLU-series AI accelerator chips and accelerator cards for cloud, datacenter, and edge workloads.
- Cambricon Neuware software platform and toolchain for model deployment and optimization.
- AI inference/training hardware systems and board-level products sold into domestic customer channels.
- MLU590
Chip pricing
- MLU370-X8~$2.1K · 2026-08
- MLU590 (思元590)~$9K · 2026-08
Financials
688256.SS · Shanghai STAR Market 688256 · via Yahoo Finance ↗
Financial trajectory via annual filings via Yahoo Finance
RMB billionsMarket data refreshed daily; annual figures from exchange filings. Figures in the reporting currency shown on the chart.
Ecosystem
Customers / end markets
- Domestic cloud, internet, government, finance, telecom, and enterprise AI infrastructure customers
Global peer set
- Nvidia is the unavoidable benchmark, but AMD Instinct, Google TPU, AWS Trainium/Inferentia, and other accelerator programs are often more precise comparables by workload
- Chinese comparables include Huawei Ascend, Biren, Enflame, Moore Threads, and other domestic accelerator startups
Competitive position
- Cambricon is important because it is publicly listed and easier to track than many private AI-chip startups.
- Its challenge is the classic accelerator problem: chip capability is only one layer; software, developer support, model coverage, and reliable supply matter just as much.
- China's flagship listed AI-accelerator designer.
- Dependent on external foundry and advanced packaging, with export-control exposure as the key risk.
- Global comparables: Nvidia is the unavoidable benchmark, but AMD Instinct, Google TPU, AWS Trainium/Inferentia, and other accelerator programs are often more precise comparables by workload.
- Global comparables: Chinese comparables include Huawei Ascend, Biren, Enflame, Moore Threads, and other domestic accelerator startups.
- Advanced foundry access, high-bandwidth memory, packaging, board integration, and software compatibility are the main constraints.
- The product risk is not just TOPS/FLOPS; it is whether customers can run real workloads without excessive porting pain.
Milestones and signals
- 2016—Cambricon Technologies was founded.
- 2026-10-04—Market cap: RMB 635B
- FY2025—Revenue: RMB 6.5B
- FY2025—Net Profit: RMB 2.1B
- Listing—Shanghai STAR Market 688256.
What to watch
- Annual and quarterly reports for revenue concentration, gross margin, and R&D intensity.
- Named deployments in cloud, government, finance, telecom, and internet companies.
- Support for mainstream model frameworks and inference/training workloads.
- Packaging and memory access for higher-end accelerator products.
Questions
Last reviewed 2026-06-28Answered questions
Is revenue growth driven by broad customer adoption or concentrated procurement cycles?
Named exposure so far: Domestic cloud, internet, government, finance, telecom, and enterprise AI infrastructure customers.
Open questions
- Which products are selling into recurring deployments rather than pilots?
- Can Cambricon's software stack reduce switching friction from CUDA/Nvidia ecosystems?