Cambricon Technologies

Cambricon Technologies (Chinese, 寒武纪) is a Chinese technology company that designs neural processing units and general-purpose graphics processors for artificial intelligence workloads. Spun out of the Chinese Academy of Sciences in 2016 and listed on the Shanghai Stock Exchange STAR Market in 2020, the company is often described in financial media as "the Nvidia of China" because it is the country's most prominent pure-play designer of AI accelerators.^[1][3]

History

Cambricon grew out of research at the Institute of Computing Technology of the Chinese Academy of Sciences, which set aside roughly 10 million yuan to fund the underlying chip project. Brothers Chen Tianshi and Chen Yunji co-founded the commercial company to bring that research to market, and early chips found their way into smartphones and servers built by customers including Huawei. When Huawei moved AI chip development in-house through its subsidiary HiSilicon, Cambricon lost its anchor customer — a blow that shaped its subsequent pivot toward data-centre accelerators.^[1]

The company completed its initial public offering on 20 July 2020 on the Shanghai Stock Exchange STAR Market, raising about US$369 million. Cambricon was added to the United States Department of Commerce Entity List in December 2022 alongside more than three dozen other Chinese technology firms, restricting its access to American technology and putting it at the centre of US-China chip export controls.^[1][2]

After years of heavy research spending, Cambricon recorded its first quarterly profit in late 2024 and turned profitable for the full year 2025, paying a maiden dividend in March 2026.^[4] Growth then accelerated sharply as Chinese buyers stockpiled domestic accelerators. First-half 2026 revenue reached about RMB 6 billion, a 108 per cent year-on-year increase, and first-quarter 2026 revenue of RMB 2.88 billion was up roughly 160 per cent with net profit near RMB 1 billion.^[2][3]

Key Technologies

Cambricon's product line has moved from small embedded neural network processors to full data-centre accelerators. The original Cambricon 1A and related 1-series chips handled inference tasks on edge and mobile devices, while the MLU (Machine Learning Unit) family — including the MLU370 and higher-end MLU590 — targets training and inference in server racks, competing directly with Nvidia's data-centre GPUs.^[1]

The company designs its own instruction set and compiler stack rather than relying on Nvidia's CUDA ecosystem, which is both its main constraint and its strategic rationale: Chinese cloud providers and state procurement programmes that cannot buy Nvidia hardware need a domestic alternative with a complete software layer. Partly state-owned and headquartered in Beijing, Cambricon employs under a thousand people, a small headcount relative to the revenue growth of recent quarters.^[1][3]

Applications and Impact

Cambricon accelerators are used for large-model training and inference, smart-city video analytics, and government and enterprise AI procurement inside China. Its commercial fortunes are tightly coupled to Chinese industrial policy: export restrictions on Nvidia chips, including the reduced-specification variants previously sold into China, have pushed domestic buyers toward Cambricon, Huawei's Ascend line, and other local designers.^[2][3]

The company's 2025–2026 turnaround made it one of the most closely watched listings in China's technology sector, with its shares attracting retail investor frenzies typical of the country's AI sector.^[3]

>See Also

• Nvidia
• Huawei AI
• Tensor Processing Unit
• AI Chip Export Controls
• AMD