Cambricon Technologies

Cambricon Technologies (Chinese: 寒武纪) is a Chinese fabless artificial-intelligence accelerator design company founded in March 2016 as a spinout from the Chinese Academy of Sciences Institute of Computing Technology and headquartered in Beijing. The company designs the Siyuan (思元) Machine Learning Unit (MLU) line of AI accelerator chips, including the MLU 290, MLU 370, MLU 590, and the next-generation MLU 690, manufactured at SMIC's 7-nanometer-class N+2 process node. As of May 2026, Cambricon is the largest Chinese pure-play AI accelerator startup by revenue and one of the principal alternatives to Huawei's Ascend platform for Chinese cloud and enterprise customers transitioning from NVIDIA-based AI compute.

At a glance

• Founded: March 15, 2016 in Beijing by Chen Tianshi and Chen Yunji, spun out from the Chinese Academy of Sciences Institute of Computing Technology.
• Status: Public. Listed on the Shanghai Stock Exchange STAR Market (688256.SS) in July 2020. Market capitalization above $200 billion USD as of late 2025 following share-price appreciation.
• Funding: Public-market financing on Shanghai STAR Market. Reported full-year 2025 revenue of approximately RMB 6.50 billion (approximately $900 million USD) with net profit of approximately RMB 2.06 billion. Q1 2026 revenue reached approximately $423 million USD.
• CEO: Chen Tianshi, Chairman and Chief Executive Officer.
• Other notable leadership: Chen Yunji, Chief Scientist and Co-Founder. Brother of Chen Tianshi. Both founders trace their research backgrounds to the Chinese Academy of Sciences Institute of Computing Technology elite computer-science youth class.
• Open weights: Not applicable. Cambricon is a chip designer, not a model producer. The Cambricon Neuware software stack provides the principal developer interface to MLU hardware.
• Flagship products: Siyuan MLU 590 (current flagship, volume production from 2023), Siyuan MLU 690 (next-generation, mass production targeting 2026), Siyuan MLU 370 and MLU 290 (earlier generations), and the Cambricon Neuware development software stack.

Origins

Cambricon Technologies was founded in March 2016 by Chen Tianshi and Chen Yunji, brothers who developed neural-network accelerator architectures at the Chinese Academy of Sciences Institute of Computing Technology beginning in 2008. The brothers were both alumni of the elite computer-science youth class at the University of Science and Technology of China, with Chen Yunji specializing in semiconductor architecture and Chen Tianshi specializing in artificial intelligence. Their research at ICT in the 2008 to 2015 period produced the DianNao series of neural-network accelerator architectures, an influential academic line on dedicated AI hardware. The Cambricon project was initially funded by approximately RMB 10 million from CAS and named for the Cambrian explosion of biological diversity.

The 2018 to 2020 period saw Cambricon's transition from a chip-IP licensing business (with Huawei HiSilicon licensing Cambricon neural processor IP for the Kirin 970 and 980 mobile system-on-chip lines) to a standalone chip design and product company. The Huawei licensing relationship was the principal early commercial revenue contributor and ended in 2019, with Huawei subsequently developing in-house neural-processor IP through the Da Vinci architecture used in Ascend and Kirin successor chips.

The July 2020 initial public offering on the Shanghai Stock Exchange STAR Market raised approximately RMB 2.58 billion ($367.7 million USD) at a $12 billion USD valuation, with the share price closing 230 percent above the listing price on the first trading day. The IPO established Cambricon as the principal Chinese pure-play AI accelerator company on public-market terms.

The 2021 to 2024 period was characterized by Cambricon's continued investment in the Siyuan MLU line, with the MLU 290 (2021), MLU 370 (2022), and MLU 590 (2023) generations succeeding each other. Revenue growth was modest through 2024 with the company reporting consistent operating losses, with principal commercial customers being Chinese cloud and internet companies (Alibaba, Tencent, ByteDance) transitioning AI compute under US export-control pressure on NVIDIA H100 and H200 access.

The 2025 to 2026 period has been transformative. The company reported its first full-year operating profit in 2025, with revenue growth of approximately 14 times year on year in Q3 2025 and full-year 2025 revenue reaching approximately RMB 6.50 billion. The trajectory reflects substantial Chinese-domestic AI compute demand under continued US export restrictions on NVIDIA chip access, with Cambricon positioned as the principal pure-play alternative to Huawei Ascend. Bloomberg, Reuters, and Tom's Hardware reporting from late 2025 has characterized Cambricon as targeting approximately 500,000 AI chip shipments in 2026, with corresponding revenue projections approaching RMB 23 billion per Chinese investment-analyst estimates.

Mission and strategy

Cambricon's stated mission is to be the principal Chinese AI compute hardware company and to provide Chinese cloud, internet, and enterprise customers with high-performance alternatives to NVIDIA-based AI compute. The strategic premise reflects structural Chinese-domestic demand for AI compute under US export-control restrictions.

The strategy combines three threads. First, the Siyuan MLU product line with continued architectural and process-node iteration through MLU 590, MLU 690, and successor generations. Second, the Cambricon Neuware software stack as the principal developer interface to MLU hardware, with PyTorch and other framework compatibility layers that aim to ease customer migration from NVIDIA CUDA. Third, capital-market access through the Shanghai STAR Market listing.

The competitive premise is that Chinese-domestic AI compute demand will continue to grow under US export-control restrictions, that Chinese cloud customers will diversify across multiple Chinese accelerator suppliers (Cambricon, Huawei Ascend, Biren, Moore Threads, MetaX, T-Head, Hygon) to manage supply concentration risk, and that Cambricon's pure-play focus and academic research credentials provide a structural positioning advantage relative to vertically integrated competitors.

Products

• Siyuan MLU 590. Current flagship. Volume production from 2023 on SMIC's N+2 process. Industry coverage places performance at approximately 80 percent of NVIDIA A100 on standard AI training benchmarks.
• Siyuan MLU 690. Next-generation accelerator targeting NVIDIA H100-class performance. Mass production targeting 2026.
• Siyuan MLU 370 and MLU 290. Earlier generations (2021 and 2022). The MLU 370 was manufactured at TSMC's 7-nanometer node before the October 2022 US export-control restrictions on TSMC's service to Chinese AI accelerator customers.
• Cambricon Neuware software stack. Developer software including the Compiler Suite, runtime, and framework compatibility layers for PyTorch and TensorFlow.
• Edge and embedded products. MLU220 and Siyuan 100 lines targeting smart-camera, automotive, and industrial-edge applications.

Distribution channels include direct enterprise sales to Chinese cloud and internet customers, server-OEM relationships with Chinese system vendors, and the Cambricon developer platform. International distribution is restricted by US export-control inclusion on the Entity List from December 2022.

Performance and positioning

Cambricon's performance positioning is anchored by direct comparison with NVIDIA accelerator products and with Huawei Ascend. Industry coverage in 2025 and 2026 has consistently characterized the Siyuan MLU 590 as roughly 80 percent of NVIDIA A100 performance on standard AI training benchmarks, and the upcoming Siyuan MLU 690 as targeting NVIDIA H100-class performance. The performance gap relative to NVIDIA's leading-edge accelerators (B200, GB200, and successor SKUs) remains substantial.

Cambricon's 2024 China AI chip market share was reported at approximately 1 percent, with NVIDIA at 66 percent, Huawei Ascend at 23 percent, and AMD at 5 percent per industry coverage. The 2025 revenue acceleration has continued through Q1 2026. The principal commercial constraint is SMIC's 7-nanometer-class capacity, on which Cambricon's targeted 2026 shipments of approximately 500,000 AI chips depend substantially. Industry coverage from Tom's Hardware and Trendforce in late 2025 has characterized SMIC's capacity as the binding constraint on Cambricon's commercial trajectory.

Leadership

As of May 2026, Cambricon Technologies' senior leadership includes:

• Chen Tianshi, Chairman and Chief Executive Officer, Co-Founder. Mathematics undergraduate (2005) and computer-science PhD (2010) at the University of Science and Technology of China and the Chinese Academy of Sciences. Net worth reported at approximately $22.5 billion in late 2025 per Bloomberg's billionaire index.
• Chen Yunji, Chief Scientist, Co-Founder and brother of Chen Tianshi. Specialist in semiconductor architecture and the principal architect of the DianNao neural-network accelerator architectures developed at ICT.

The founder team has remained intact since 2016, with no public departures of senior leadership through May 2026.

Financials

Public-market listing on the Shanghai Stock Exchange STAR Market (688256.SS). Market capitalization above $200 billion USD as of late 2025 following substantial share-price appreciation. Reported full-year 2025 revenue of approximately RMB 6.50 billion, with net profit of approximately RMB 2.06 billion (the first full-year operating profit in the company's history). Q1 2026 revenue reached approximately $423 million USD. Chinese investment-analyst projections for full-year 2026 revenue have been reported at approximately RMB 23 billion.

Cambricon was added to the US Department of Commerce Entity List in December 2022, restricting access to US semiconductor manufacturing technology and software.

Industry position

Cambricon Technologies occupies a structurally distinctive position as the largest Chinese pure-play AI accelerator startup by revenue and the principal pure-play alternative to Huawei Ascend for Chinese cloud and enterprise customers. Industry coverage from Bloomberg, Reuters, and other outlets in late 2025 and early 2026 has consistently characterized Cambricon as the principal Chinese AI chip startup benefiting from US export-control restrictions on NVIDIA chip access in China.

The structural position is constrained by SMIC's 7-nanometer-class capacity, by the absence of high-bandwidth memory supply (HBM3 and HBM3e access remains restricted), and by the continued performance gap relative to NVIDIA's leading-edge accelerators. The market-capitalization expansion through 2025 to roughly $200 billion USD on revenue of approximately $900 million USD has produced extended discussions of valuation in industry coverage, with implied price-to-earnings and price-to-sales multiples cited as substantially above NVIDIA's at the equivalent revenue scale.

Competitive landscape

• Huawei Ascend. The largest Chinese AI chip producer by units shipped. Vertically integrated with Huawei's CANN software stack, MindSpore framework, and the broader Huawei product line. Approximately 23 percent of the Chinese AI chip market in 2024.
• NVIDIA. The displaced incumbent in the Chinese AI chip market under US export-control restrictions. Continues to ship export-compliant SKUs (H20 and successor variants) with substantially reduced capability.
• Biren Technology. Chinese pure-play GPU startup. Hong Kong IPO in January 2026. BR100 and BR104 GPGPU products.
• Moore Threads, MetaX. Chinese pure-play GPU startups with IPO filings in late 2025.
• Alibaba T-Head. Alibaba subsidiary with the Hanguang and PPU AI accelerator lines, vertically integrated with Alibaba Cloud distribution.
• Hygon Information Technology. Chinese AMD-licensed CPU and accelerator designer.

Outlook

• The Siyuan MLU 690 mass-production timing through 2026 and the corresponding revenue contribution.
• The 2026 shipment trajectory toward the reported 500,000-unit target and the corresponding SMIC N+2 capacity allocation.
• The HBM supply environment for Chinese AI accelerator producers and the implications for Cambricon's MLU 690 specification.
• The competitive dynamic against Huawei Ascend and against the Chinese pure-play GPU competitors (Biren, Moore Threads, MetaX) following the late-2025 and early-2026 IPO wave.
• The valuation trajectory at the current Shanghai STAR Market market capitalization and any commentary from Chinese investment analysts on price-to-sales sustainability.
• Continued US export-control trajectory and any extension or relaxation through 2026 and 2027.

Sources

• Cambricon official site. Company reference.
• Wikipedia: Cambricon Technologies. Company history reference.
• Trendforce: Cambricon remains China's top AI chip startup; rumored 2026 triple output faces SMIC limits. 2026 production targets reference.
• Bloomberg: Cambricon aims to triple output to replace NVIDIA in China. 2026 production strategy reference.
• Tom's Hardware: Cambricon targets 500,000 AI chips in 2026. Production targets and constraints reference.
• Fortune: 'The NVIDIA of China' says revenue spiked 14X last quarter. Revenue trajectory and founder net worth reference.
• Trendforce: China GPU race intensifies: Cambricon turns profitable in 2025. 2025 profitability reference.
• VnExpress: From youth geniuses to tech founders. Founder background reference.