← All companies

Chips / Company profile

Cambricon

What Cambricon does

Cambricon (中科寒武纪科技股份有限公司; SSE STAR Market: 688256) is a fabless AI semiconductor and intelligent computing systems company headquartered in Beijing. The company’s public-facing strategy emphasizes an “End + Cloud” AI ecosystem and the development of AI core processor chips plus system software platforms intended to cover cloud training, cloud inference, and edge/terminal inference use cases. On its website, Cambricon describes its product portfolio across: (1) cloud accelerator cards built around its “思元370” family (including MLU370-S4/S8 and other variants), (2) edge AI acceleration modules in its “思元220” family (including MLU220-based modules/cards), (3) terminal intelligent processor IP (Cambricon-1M and Cambricon-1H), and (4) software stacks such as Cambricon NeuWare and MagicMind intended to help customers deploy models on Cambricon hardware.

Business model (as described through public disclosures): Cambricon designs chips and system software and sells chip products and associated platforms to customers/partners, while delegating manufacturing/packaging/testing to external foundry/assembly-test partners (a fabless model).

Technical differentiation (based on company-published product and platform descriptions): - Hardware: Example product pages position the MLU370-S4/S8 as a high-density cloud inference-focused accelerator card using the company’s “MLUarch03” architecture on a 7nm process, with PCIe Gen4 support and listed peak performance/feature specs. - Edge computing architecture: The MLU220 family page highlights integration of an ARM Cortex-A55 (4-core) plus AI acceleration to support edge use cases, and describes the MLUv02 architecture as using on-chip networking (NoC) for improved compression and cache/bandwidth characteristics. - Software–model integration: Cambricon’s developer-community posts describe “Day 0” model support efforts using its NeuWare software ecosystem, with references to integration with PyTorch/vLLM/Diffusers and optimization via libraries and kernel/code-generation approaches (as claimed in the post).

Current strategic position (publicly observable): Cambricon is an established public company on China’s STAR Market and continues to fund R&D and operations through capital markets activity, including a 2025 targeted equity issuance. It also continues to emphasize ongoing software ecosystem work and model adaptation to open-source model releases.

News

Company record

Apr 30, 2026 · Reporting · 上海证券报·中国证券网(roadshow.cnstock.com)Cambricon investor Q&A: company leadership and R&D workforce structure (2025 annual results roadshow)

During the 2025 annual results roadshow on cnstock.com, Cambricon’s Chairman/General Manager answers questions and states that as of year-end 2025, 80.13% of employees are R&D personnel and 80.95% of R&D employees hold master’s degrees or above (per the roadshow transcript).

Apr 24, 2026 · Official · 寒武纪开发者社区Cambricon Day 0 adaptation for DeepSeek-V4 via vLLM

In a developer-community post, Cambricon states it completed “Day 0” adaptation (code open-sourced) for DeepSeek-V4-flash (285B) and DeepSeek-V4-pro (1.6T), attributing the result to its NeuWare ecosystem and chip+algorithm co-innovation efforts; the post also describes optimizations across software and hardware paths (as claimed by the company).

Oct 09, 2024 · Official · 寒武纪(公司新闻)Cambricon warns about fraudulent hiring using the company name

Cambricon issued an official statement warning job seekers about fake recruiting activities impersonating Cambricon, including guidance that official channels and the HR email reccruit@cambricon.com are used for verification.

Date not disclosed · Official · 寒武纪(产品技术页面)Cambricon product and performance overview: MLU370-S4/S8 cloud inference accelerator card

Cambricon’s product technical page for the MLU370-S4/S8 includes claimed hardware specs such as listed peak performance figures, memory capacity/bandwidth, PCIe Gen4 interface, and passive cooling/board power.

Date not disclosed · Official · 寒武纪(产品技术页面)Cambricon product and architecture overview: MLU220 edge computing module series

Cambricon’s MLU220 series page describes edge compute integration (ARM Cortex-A55 + AI acceleration) and discusses the MLUv02 NoC-based architecture plus software deployment flexibility.

Source map · 2 recurring channels · 23 references

Still resolving: Newsroom · Official X

Funding

Latest disclosed valuation$25.7B
Tracked capital$7B7 sourced rounds
DateRoundRaisedValuationLead / investorsEvidence
Jul 14, 2020IPO (STAR Market) — offering economics disclosed (gross proceeds)$2.6B
Aug 01, 2016Pre-series A (announced as “tens of millions” funding)
UndatedIPO / STAR Market listing (A-share IPO listing)
Not attributed

What it builds

AIDC MLU370-S4/S8 intelligent accelerator cards

Cambricon positions the MLU370-S4/S8 as a cloud high-density inference-focused accelerator card based on its “MLUarch03” architecture (7nm process per the product page) with PCIe Gen4, and lists peak performance and board-level specs including memory type/capacity/bandwidth and 75W passive cooling design.

source ↗
MLU220 edge computing modules (Si Yuan 220 series)

Cambricon describes the MLU220 as an AI acceleration product for edge computing scenarios, integrating a quad-core ARM Cortex-A55, LPDDR4x memory, and rich peripheral interfaces; the page also describes the MLUv02 architecture as NoC-based for improved on-chip compression and cache/bandwidth behavior.

source ↗
Cambricon NeuWare

NeuWare is described as Cambricon’s software stack/ecosystem that supports model deployment on Cambricon hardware; Cambricon’s developer-community materials discuss it as enabling Day-0 adaptation workflows for newly released models (per the post).

source ↗
MagicMind

MagicMind is listed on Cambricon’s product navigation as the company’s software development platform product category alongside NeuWare, indicating it is part of Cambricon’s end-to-end deployment tooling.

source ↗

Milestones & partnerships