Cambricon
Overview
Cambricon (Chinese name: 寒武纪; Korean: 한무기) is a fabless AI semiconductor design company founded in Beijing, China, in 2016. It develops neural processing units (NPUs) specialized for deep learning computation, AI accelerators, and related software stacks, and is regarded as a leading player in China's 'AI chip self-reliance' efforts to reduce dependence on Nvidia. It listed on the Science and Technology Innovation Board (STAR Market) of the Shanghai Stock Exchange in 2020, and amid the generative AI boom of 2024–2025, its stock price surged, making it a symbolic stock in the discourse of China's semiconductor self-reliance.
Main Content
Background of founding and origin of name
Cambricon was founded by brothers Chen Tianshi (陳天石) and Chen Yunji (陳雲霽), who came from the Institute of Computing Technology of the Chinese Academy of Sciences. In the early 2010s, the two attracted international academic attention for research on a deep learning-specific accelerator architecture called 'DianNao', and they founded the company to commercialize the results of this research. The company name is derived from the 'Cambrian Explosion,' about 540 million years ago, when living species underwent explosive diversification, symbolizing explosive evolution in the age of artificial intelligence.
Major product lineup
- Cambricon 1A/1H: Early NPU IP (design assets); it was adopted in Huawei's Kirin 970 chip and recorded as the world's first case of integrating an independent NPU into a smartphone.
- MLU series: AI accelerator lineup for data centers and edge. It includes MLU270, MLU290, MLU370, MLU590, etc., and targets cloud inference and training workloads.
- Cambricon software stack: A development tool system provided together with hardware, including the BANG C language, Neuware SDK, MagicMind inference engine, etc.
Technical characteristics
Cambricon's chips adopt a domain-specific architecture (DSA) specialized for deep learning computation, not general-purpose GPUs. They focus on maximizing throughput of low-precision (INT8, FP16, etc.) matrix operations and increasing power efficiency, and aim for ecosystem lock-in with an integrated software stack spanning cloud-edge-device. However, the assessment that its developer ecosystem is inferior to Nvidia's CUDA ecosystem has long followed it.
Business structure and customers
Initially, the model of licensing IP to smartphone and device makers such as Huawei accounted for a large share of revenue. Later, as Huawei internalized its own NPU (Da Vinci architecture), the share of the IP business declined, and the center of gravity shifted to sales of AI servers and accelerator cards and cloud service businesses targeting governments, public institutions, telecom carriers, and internet companies. The high revenue concentration and heavy dependence on certain large customers have long been pointed out as risks.
Listing and capital markets
At the time of its STAR Market listing in July 2020, its market capitalization was around 100 billion yuan, and the stock price rose sharply on the first day of listing. Afterwards, it suffered a long-term slump due to continued losses and poor performance, but from late 2024, as expanded AI infrastructure investment and strengthened U.S. semiconductor export controls on China coincided, it rapidly emerged as a beneficiary of the 'de-Nvidia' trend. In 2025, its stock price soared to the hundreds of yuan range, surpassing Kweichow Moutai to become the highest-priced stock on China's A-share market, becoming the market's biggest topic stock.
Sanctions and geopolitical issues
In December 2022, the U.S. Department of Commerce added Cambricon to the Entity List of companies subject to export controls. This was due to concerns that advanced AI chips could be used for China's military and surveillance capabilities, and afterward restrictions arose on access to U.S.-made design tools (EDA) and foundries. Cambricon had used overseas foundries such as Taiwan's TSMC, but with tightened regulations it focused more on cooperation with domestic Chinese foundries (such as SMIC) and building its own software ecosystem.
Latest Trends
In 2024–2025, as demand for generative AI and large language model (LLM) training and inference exploded, Cambricon came to the fore as a key alternative for supplying AI accelerators to data centers in China. As U.S. export controls on advanced semiconductors to China continued and it became difficult for Chinese big tech companies to secure Nvidia's latest chips (H100, B200, etc.), they turned their eyes to domestic chips, and Cambricon received expectations of cooperation with Alibaba, Tencent, Baidu, etc., and benefits from government-led localization projects. In 2025, with expectations for improved performance and policy support overlapping, whether it would turn to a quarterly profit became the market's greatest interest, and at the same time debate over its high valuation intensified. Meanwhile, the Chinese government is pushing semiconductor self-reliance as a national strategy and injecting large-scale funds into equipment and design fields through a large fund (Big Fund Phase III), so the financing environment for domestic AI chip companies including Cambricon is developing favorably.
Related Topics
- [[AI semiconductor]]
- [[Nvidia]]
- [[Semiconductor export controls]]
- [[STAR Market]]
- [[Huawei]]
- [[Neural processing unit]]