cambricon technologies
on this page
cambricon technologies corporation limited is a chinese semiconductor company specializing in neural processing units (npus) for artificial intelligence applications.
company overview
founded in march 2016 by brothers chen tianshi and chen yunji, both alumni of the university of science and technology of china and the chinese academy of sciences’ loongson cpu project.1
company facts
- founders: chen tianshi and chen yunji (ustc prodigies)
- headquarters: beijing, china
- status: partially state-owned, publicly listed (688256.ss)
- market cap: $60.4 billion with 418 million shares
- employees: thousands across r&d and manufacturing
- backing: alibaba group, tencent holdings, state investors
products
mlu series neural processing units
cambricon’s core offering is the machine learning unit (mlu) series, manufactured on 7nm processes via smic.2
tl;dr
model | release | performance | memory | target |
---|---|---|---|---|
mlu590 (siyuan 590) | 2024 | 1 petaflops cnns, 64 cores | 512 gb/s hbm3 | large-model training |
mlu370-x8 | 2022 | 256 int8 tops | 48gb lpddr5 | cloud inference |
mlu290-m5 | 2021 | 512 int8 tops | 32gb hbm2, 1,228 gb/s | edge ai |
mlu270-s4 | 2019 | foundational specs | variable | early cloud/edge |
the mlu series supports fp32, fp16, bf16, int16, int8, and int4 formats for deep learning frameworks.3
software ecosystem
cambricon develops platform software for end-to-end ai ecosystems, emphasizing “end-cloud integration” across servers, edge devices, and terminals.1
financial transformation
early struggles and huawei partnership loss
cambricon initially made 98% of its revenue from intellectual property licensing to huawei for kirin 970 and 980 smartphone chips (2017-2018).4 the partnership ended when huawei developed its proprietary davinci/ascend architecture in 2018, causing revenue to drop from ¥117 million to ¥16-18 million.5
recent turnaround
tl;dr
period | metric | value | significance |
---|---|---|---|
q4 2024 | profit | ¥240-328 million | first profitable quarter |
q1 2025 | revenue | ¥1.111 billion | 4,230% year-over-year growth |
july 2025 | funding | $560 million (¥4 billion) | private placement for ai development |
august 2025 | stock surge | 20% daily limit | ue8m0/deepseek market response |
current | order book | ¥30 billion ($4.1 billion) | reported orders in pipeline |
major funding rounds
china’s big fund phases have been key investors alongside private capital.6
major backers include:
- alibaba group (early investor, exited september 2023)
- tencent holdings
- china merchants bank (exited 2023)
- sdic venture capital (sold 99.98% stake for ¥1.48 billion in 2023)
- government strategic funds
manufacturing and production
transition to domestic supply chain
cambricon shifted from tsmc to smic for 7nm production following us sanctions:7
- current capacity: 30,000+ wafers per month (2025)
- target capacity: 100,000 wafers per month by 2026
- process technology: smic 7nm/5nm despite yield challenges
- equipment sourcing: global suppliers where possible
performance positioning
cambricon chips lag nvidia equivalents by approximately 4-5 years in performance but focus on cost-effectiveness and domestic availability.8 the mlu590 delivers competitive inference performance for chinese ai models.
strategic partnerships
bytedance collaboration
close relationship with bytedance for ai infrastructure deployment, positioning cambricon as china’s “pure-play” ai chip company.9
enterprise customers
- chinese telecommunications companies
- cloud service providers
- government institutions requiring domestic hardware
- ai companies developing with ue8m0 data format
market position
competitive landscape
cambricon competes with moore threads (gpu focus), huawei ascend (enterprise), and other domestic chip companies in china’s ai hardware market.10
the company benefits from government mandates requiring local chips in data centers and the broader china ai hardware decoupling strategy.
stock performance
cambricon shares rose 383% during 2024, making it china’s top-performing stock. august 2025 saw additional 20% gains when deepseek announced ai models optimized for “next-generation domestic chips.”11
technical challenges
manufacturing constraints limit cambricon to 7nm processes through 2026, driving focus on architectural efficiency rather than raw performance. the company’s chips excel in inference workloads where power efficiency matters more than peak compute.
error rates remain higher than western equivalents but acceptable for many ai applications, especially when combined with algorithmic optimizations.
future outlook
cambricon positions itself as a critical component of china’s ai independence strategy. success depends on:
- smic yield improvements for advanced nodes
- continued government support and procurement
- development of software ecosystems competitive with cuda
- partnerships with chinese ai companies for hardware-software co-optimization
the company’s transformation from huawei supplier to independent ai chip leader reflects china’s broader semiconductor strategy evolution.
references
[1] cambricon technologies - wikipedia.
[2] machine learning unit (mlu) - cambricon - wikichip.
[3] cambricon technologies, potential challenger of nvidia, still has a long way to go | equalocean.
[4] key stat | huawei’s previous partner cambricon to become china’s first ai chip stock | kr asia.
[5] cambricon: staggering growth, ipo and post-huawei uncertainty | equalocean.
[6] china chip industry gets $47.5 billion in new funding | cnn business.
[7] us-china ai chip race: cambricon’s first profit lands | artificial intelligence news.
[9] chinese cambricon’s first profit signals beijing’s tech independence | techhq.
[10] chinese players in ai chip design | 1421 by acclime.
[11] china’s ai chip leader cambricon jumps 20% on self-reliance bets | bloomberg.