Cambricon Technologies: Keystone of China’s AI-Chip Autonomy ?

Administrateur
Jun 1
3 min read

Founded in March 2016, Cambricon Technologies Corporation Limited grew out of a Chinese Academy of Sciences (CAS) project launched in 2008 to build brain-inspired deep-learning processors. Led by brothers Chen Tianshi and Chen Yunji, the company embodies China’s quest to create home-grown alternatives to Western GPUs. Its name—echoing the Cambrian explosion—signals an ambition to usher in a new era of AI chips.

Product portfolio: from cloud to device

Cambricon first supplied NPU IP cores—the Cambricon-1A powered Huawei’s Kirin 970 in 2017, making the Mate 10 and P20 among the first “AI-ready” phones and lifting shipments past 100 million in 2018. After Huawei internalised NPUs (“DaVinci”) in 2019, Cambricon diversified:

Siyuan cloud accelerators (100, 270, 290, 370) – PCIe cards up to 512 INT8 TOPS with 32 GB HBM for data-centre training or inference, available in turnkey AI clusters.
MLU220 edge chips and modules – Arm CPUs plus NPUs delivering 16 TOPS in a 16 W envelope for cameras, IoT gateways and autonomous-driving units.
Embedded IP and SoCs – licensable cores for robots, cameras and mobile devices, extending its original IP business.
Neuware software stack – compilers, drivers and the CNML library; indispensable yet still far from Nvidia’s CUDA breadth.

This cloud–edge–terminal matrix underpins Cambricon’s goal of a complete AI-computing ecosystem.

Markets: a mainly domestic clientele

Cloud providers (Alibaba, Baidu, ByteDance): cut off from Nvidia’s highest GPUs, they are testing Cambricon for large-model training.
Public sector and defence: supercomputers, safe-city projects and state labs offer a steady order flow.
Edge-device makers: industry and automotive partners integrate vision modules and ADAS boards.
Smartphones: since Huawei, adoption is modest; most OEMs still rely on Qualcomm or MediaTek solutions.

Ownership: founder and state in lockstep

Early 2025 figures show CEO Chen Tianshi owning ~29 %. CAS, through China Sciences Suanyuan, holds ~16 %, giving the pair almost 45 %. Early private backers such as Alibaba exited in 2023; domestic mutual funds have moved in.

Competitive field: internal duel with Huawei, global duel with Nvidia

Nvidia still sets the performance and software bar, but export controls bar A/H100 sales to China, creating the opening Cambricon and Huawei exploit.
AMD is minor in AI and subject to the same restrictions.
Huawei Ascend (edge 310, cloud 910/910C) leads the domestic market thanks to vertical integration (servers, cloud, telecom). Cambricon’s neutrality appeals to cloud giants who view Huawei as a competitor.

Financial performance

Valuation: strategic premium

Listed on the STAR Market in July 2020, shares rose 230 % on day one, valuing the loss-making firm at 12 B $. In 2024 the prospect of Nvidia substitution sent the stock up 380 %, above 40 B $—roughly 200 × sales—driven by:

Rarity: a pure AI-chip play on Chinese exchanges.
Policy tailwinds: subsidies, public procurement, inclusion in the 14th Five-Year Plan.
Hyper-growth hopes: forecasts above 50 % a year and improving margins.

Role in China’s tech strategy

Cambricon illustrates the pattern: public lab → subsidised start-up → pre-profit IPO → state-backed champion. Beijing also manufactures demand: ministries and municipalities are guided to buy local chips, cushioning the firm’s early losses. In parallel, China builds a parallel stack—chips (Cambricon, Ascend), interconnects, frameworks (MindSpore, PaddlePaddle, Neuware) and fabs (SMIC)—to reduce Western dependence.

Challenges

Tech gap – Siyuan 370 still trails Nvidia H100.
Software ecosystem – Neuware must win over CUDA developers.
Domestic rivalry – Huawei already claims half the local accelerator market.
Sustained profitability – R&D outlay exceeded revenue through 2023; balancing innovation and earnings is crucial.
Manufacturing risk – advanced 7 nm nodes depend on TSMC, sensitive to geopolitical pressure.

Outlook

If China stays committed to localisation, Cambricon could:

quadruple revenue by 2027 on generative-AI rollouts, sovereign cloud and smart-city surveillance;
lift net margin above 15 % through scale and software income;
export to Belt-and-Road partners needing affordable AI hardware.

The bullish case envisions a “Chinese Nvidia” earning multi-billion-dollar sales; the cautious case flags overcapacity, fragmented standards and cut-throat rivalry.

Conclusion

Cambricon personifies China’s tech-sovereignty drive: born in a public lab, fuelled by national VC, blessed by capital markets, it became an industry pivot despite years of red ink. Its 2024 inflection—sparked by US sanctions—shows the power of industrial policy plus market finance. The next decade will reveal whether Cambricon matures into a world-class rival or serves as a transitional link in an ever-broader “made in China” semiconductor ecosystem.