category
Dec 9, 2025
Chinese AI chip maker targets half a million units despite low yields
TechRadar
Chinese chipmaker Cambricon Technologies aims to triple AI chip production in 2026, seeking to fill the gap left by Nvidia's retreat from the Chinese market.
Bloomberg reports that the company intends to produce approximately 500,000 AI accelerator chips next year, with 300,000 units consisting of its main Siyuan 590 and 690 models.
This represents a sharp increase from the roughly 142,000 units expected in 2025, yet Cambricon still faces major fabrication challenges.
The reported yield rate for its 590 and 690 chips stands at just 20%, meaning that only one in five chips produced is usable.
Even with access to capacity at Semiconductor Manufacturing International, the effective output could fall far below projections.
By comparison, TSMC's 2nm technology, seven generations ahead of SMIC's capabilities, reaches a 60% yield, showing the efficiency gap.
Shortages of memory, including HBM and LPDDR components, further threaten the ability to meet production goals, potentially slowing delivery to data center clients.
Cambricon's move comes as Chinese companies like Alibaba and ByteDance increasingly favor local suppliers.
They are supported by the Chinese government's incentives to boost China's semiconductor independence.
Cambricon's reported revenue for the last quarter surged fourteenfold, showing strong domestic demand and investor confidence.
However, this plan will put Cambricon in direct competition with tech giant Huawei, which is planning to double its chip output, increasing pressure on Cambricon.
Both companies compete for similar wafers and fabrication resources, creating bottlenecks that could limit the speed and scale of production.
Cambricon's strategy relies heavily on the "N+2" 7nm process node at SMIC, but whether it can sustain large scale fabrication remains uncertain.
Trade restrictions and chip embargoes over the past year have restricted access to high end AI hardware, making domestic alternatives essential for national AI ambitions.
