Cambricon

Cambricon is a company from Beijing that designs computer chips. These chips are built to handle the massive computing tasks behind programs like ChatGPT. Such programs need to perform billions of computing steps per second, and ordinary processors from a laptop are not enough for that. The company was founded in 2016, spun out of a state research institute. It has been listed on the Shanghai stock exchange since 2020. The name alludes to the Cambrian, a geological era in which a huge number of new life forms suddenly emerged — the reference is to a similar explosion of diversity in computing machines.

China’s answer to Nvidia

The market for AI chips is dominated worldwide by a single company: the US corporation Nvidia. Its components are found in almost every major data center where AI models are trained. Since 2022, however, the US has banned the export of its most powerful chips to China. The justification is security policy: Washington wants to prevent China from building militarily usable AI on a large scale.

This created a gap for Chinese technology companies. They still need computing power, but can no longer purchase it without restriction. Cambricon is one of the providers meant to fill this gap. As a result, the company’s share price rose dramatically in 2024 and 2025, at times by several hundred percent.

That’s why the name regularly appears in business news. There, Cambricon stands less for a specific product than for a question: how independent can China become in AI technology? An important domestic competitor is Huawei with its Ascend chip series.

What makes an AI accelerator different

Cambricon does not manufacture its chips itself. The company only designs them and has them produced in specialized factories. This model is called fabless, meaning without a factory. Nvidia works the same way.

The chips themselves are called AI accelerators. A normal processor is an all-rounder: it works through tasks one after another, but very flexibly. An accelerator can do much less, but performs these few things a thousand times over, simultaneously. The comparison: a processor is a professor who can answer any question. An accelerator is a school class that knows only one type of task, but solves it in chorus.

AI needs exactly this type of task. Neural networks — the computational structures behind modern AI systems — consist almost entirely of multiplications of large number tables. Cambricon’s product lines carry names like Siyuan (in English, MLU). A chip always comes with software that allows developers to access it. This software is often the bigger hurdle: Nvidia’s programming environment CUDA has been the standard for years, and Cambricon has to convince customers to rewrite their programs.

When the name makes headlines

As a private individual, you will never buy a Cambricon chip. These components are found in data center servers, not in phones or game consoles. You will therefore mainly encounter the name in stock market and business news.

Typical occasions are quarterly results, new chip generations, or rumors about major orders from Chinese internet companies. Political news also leads to price movements, for instance when the US tightens export rules or Beijing mandates domestic purchases. A common misconception: Cambricon does not build AI models, only the hardware underneath them.

To put such news into perspective, it helps to take a sober look at the scale involved. Cambricon’s revenue long remained in the hundreds of millions of yuan range, while Nvidia generates tens of billions of dollars. The stock market valuation thus reflects expectations above all, not today’s business.