Chinese semiconductor firm Cambricon Technologies is preparing a sharp increase in AI chip production as it seeks to benefit from the reduced presence of Nvidia in the Chinese market. According to Bloomberg, Cambricon plans to produce close to 500,000 AI accelerator chips in 2026, nearly tripling expected output from 2025.
Roughly 300,000 of these units are set to be the company’s flagship Siyuan 590 and 690 models. By comparison, Cambricon is expected to ship around 142,000 units in 2025, highlighting the scale of the planned expansion. The strategy reflects growing pressure on Chinese cloud providers and data center operators to source domestic hardware.
The move aligns with broader national efforts to reduce reliance on foreign chip suppliers amid ongoing trade restrictions and export controls.
Table of Contents
Low yields remain a serious constraint
Despite ambitious targets, Cambricon faces significant manufacturing challenges. Reports suggest the yield rate for its 590 and 690 chips is close to 20%, meaning four out of five chips produced fail to meet usable standards. This sharply limits effective output even if wafer capacity is secured.
The company relies heavily on SMIC for fabrication, using its N+2 7nm process. In contrast, TSMC achieves yields near 60% on far more advanced nodes, underscoring the technology gap.
Memory availability is another bottleneck. Shortages of high bandwidth memory and LPDDR components could slow deliveries and constrain system integration for data center customers.
Domestic demand and state backing
Chinese technology giants such as Alibaba and ByteDance are increasingly favoring domestic AI hardware, encouraged by government incentives aimed at semiconductor independence. This shift has driven a sharp rise in Cambricon’s revenue, with reports indicating a fourteenfold increase in quarterly earnings.
Investor confidence has followed, positioning Cambricon as one of the primary beneficiaries of Nvidia’s constrained sales in China. Domestic AI workloads, especially in inference and training at scale, are creating sustained demand for local accelerators even if performance lags behind global leaders.
Huawei complicates the picture
Cambricon’s expansion places it in direct competition with Huawei, which is also accelerating AI chip production. Huawei is reportedly planning to double its own output, intensifying competition for wafers, memory, and fabrication capacity.
Both companies depend on similar manufacturing resources, raising the risk of internal bottlenecks that could limit overall scaling. While Cambricon focuses on GPU-style accelerators, Huawei’s broader hardware ecosystem may give it an advantage in end-to-end deployment.
This rivalry could shape how China’s AI infrastructure develops over the next several years.
A strategic push with limits
Cambricon’s production ramp highlights the strategic importance of domestic AI chips in China’s technology roadmap. Government backing and rising demand provide strong momentum, but low yields, resource competition, and performance gaps remain unresolved.
While the expansion may help close supply gaps left by Nvidia’s exit, it does not yet challenge global leaders on efficiency or scale. The success of Cambricon’s plan will depend on improvements in fabrication reliability and sustained access to critical components.
