Cambricon targets 500K AI accelerator shipments in 2026; 300K units Siyuan 590/690 via SMIC N+2

Cambricon Technologies is ramping AI accelerator production to 500,000 units in 2026, up from approximately 116,000–142,000 in 2025, including as many as 300,000 units of its Siyuan 590 and next-generation Siyuan 690 chips. The production ramp targets roughly 3x growth and represents Cambricon's push to fill the gap left by NVIDIA's curtailed market access in China following U.S. export controls. Cambricon is now one of only two companies on China's government-approved AI hardware procurement list, alongside Huawei; NVIDIA, despite some easing of restrictions, is still excluded.

The Siyuan line uses SMIC's N+2 (7nm-class) process and accounts for the bulk of Cambricon's revenue. Cambricon reported $423 million in Q1 2026 revenue (160% YoY) and $1 billion in net profit, a sharp turnaround from losses in early 2024. ByteDance is the dominant customer, pre-ordering roughly 200,000 Siyuan 590 units. However, manufacturing yield challenges loom: reported yield rates for Siyuan 590/690 stand at just 20%, meaning most wafers remain unusable. Additionally, Cambricon depends on SK Hynix and Samsung for high-bandwidth memory, creating a critical supply bottleneck.

The 500K production target hinges on SMIC's ability to allocate 7nm capacity amid competing demands from Huawei (planning to double its own output) and other chipmakers. JPMorgan analysts estimate Cambricon may receive only 300–350K usable dies in 2026 vs. the 500K aim, accounting for yield rates. Success depends not just on fab capacity but also on HBM availability and customer absorption. Architects should track SMIC capacity allocation and HBM supply as the binding constraints for Chinese domestic chip scaling.