Cambricon is a fabless AI chip designer focused on “cloud, edge, terminal” intelligent computing. It generates revenue mainly from selling data center AI chips, accelerator cards, and server-class systems, supported by system software and tooling. In 2025, total operating revenue reached RMB 6.50 billion, with RMB 6.48 billion coming from the cloud product line at a 55.22% gross margin, showing a business mix dominated by data center deployments.
Core activities include:
- Cloud products (data center)
Cloud intelligent chips and accelerator cards, plus “smart” server systems built around Cambricon chips. Cambricon also sells an intelligent computing cluster system that bundles its cards or smart servers with partner servers, networking, and storage, plus Cambricon’s cluster management software for packaged data center deployments.
- Edge products
Edge chips and hardware aimed at workloads that run between terminals and centralized cloud infrastructure. In 2025, the edge product line contributed RMB 3.39 million in revenue, reflecting a smaller commercial footprint than cloud products.
- IP licensing and platform software
Licensing of Cambricon processor IP, plus a unified base system software platform and development toolchain intended to reduce porting work across Cambricon’s cloud and edge product families.
- Developer ecosystem
NeuWare and MagicMind form the core of Cambricon’s software offering for deployment and inference acceleration, supported by integrations and developer tools, including open-source enablement for PyTorch on Cambricon MLU hardware.
Market position
Cambricon ranks among China’s best-known domestic AI accelerator suppliers and benefited from policy support for localized compute. In December 2025, the Financial Times reported that Huawei and Cambricon processors were added to a government procurement catalogue, reinforcing public-sector demand signals. Cambricon’s 2025 swing to profitability, with net profit attributable to shareholders of RMB 2.06 billion, strengthened its standing with large customers.
Competition centers on Huawei (Ascend) and a widening group of domestic GPU and accelerator designers. The main adoption friction across the segment remains software migration away from Nvidia’s CUDA-centered stack, making tooling, framework support, and migration workflows a decisive battleground.