Cambricon Posts First AI Profit

Post LinkedIn

📊Read original on Bloomberg Technology

#ai-chips #china-boom #localizationcambricon-chips

💡China's AI chip profit milestone offers Nvidia alternatives for infra builders

⚡ 30-Second TL;DR

What Changed

First-ever annual profit for Cambricon

Why It Matters

Strengthens China's AI hardware independence, pressuring global chip leaders and opening alternatives for AI infrastructure.

What To Do Next

Assess Cambricon chips for cost-effective AI training in restricted markets.

Who should care:Enterprise & Security Teams

🧠 Deep Insight

Web-grounded analysis with 7 cited sources.

🔑 Enhanced Key Takeaways

•Cambricon projects RMB 6-7 billion in full-year 2025 revenue, a 411-496% surge, with net profit of CNY 1.9-2.2 billion.[7]
•ByteDance accounts for over 50% of Cambricon's orders, with expected expansion to Alibaba as a new major customer.[3]
•Cambricon plans to ship 500,000 AI accelerators in 2026, including 300,000 units of Siyuan 590 and 690, more than tripling 2025's estimated 142,000 units.[2]
•Current yields for top-line Siyuan 590 and 690 chips at SMIC are only 20%, constrained by N+2 node limitations and HBM supply shortages.[3]

🛠️ Technical Deep Dive

•Siyuan 290 (launched 2021): Fabricated on TSMC 7nm node, OAM form factor, 32GB HBM2 memory, uses in-house MLU-Link for chip-to-chip interconnects, cornerstone for single-node AI systems.[1]
•Siyuan 590: Flagship current product, upgraded from Siyuan 290, produced on SMIC N+2 node for high-end AI training and inference.[1]
•Siyuan 690: Next-generation chip in testing phase, mass production potentially delayed to H2 2026, also on SMIC N+2 node with current 20% yields.[1]

🔮 Future ImplicationsAI analysis grounded in cited sources

Cambricon's domestic market share in China AI chips will exceed 10% by end-2026

Aggressive 500K unit shipment target in 2026, backed by major customers like ByteDance and Alibaba, leverages Beijing's Nvidia restrictions amid SMIC production ramps.[2]

SMIC yield improvements to over 40% on N+2 node will enable Cambricon's production goals

Current 20% yields limit output, but China's focus on domestic manufacturing and demand pressure necessitate rapid advancements for 2026 tripling of volumes.[3]