Chinese chipmaker Cambricon Technologies aims to triple its artificial intelligence (AI) chip production in 2026 to half a million units, positioning itself to fill the gap left by Nvidia Corp.'s forced exit from China, according to Bloomberg on Thursday. The Beijing-based company plans to produce up to 300,000 units of its flagship Siyuan 590 and 690 chips, leaning primarily on Semiconductor Manufacturing International Corp. (SMIC)'s 7-nanometer (nm) production process.
AI Generated Image
The expansion reflects China's accelerating push toward technological self-sufficiency following U.S. export restrictions. Cambricon's shares rose 2.8% in Shanghai on Thursday, while SMIC gained 3.9% in Hong Kong. The company reported a 14-fold revenue surge in the September quarter and has seen its market value increase ninefold since 2021.
ByteDance currently accounts for more than half of Cambricon's orders, and the company is positioned to win business from Alibaba Group in coming years, Bloomberg cited people familiar with the matter. However, Cambricon later issued a WeChat statement calling media reports about its products, customers and production forecasts inaccurate, without specifying outlets.
Huawei Technologies Co. is simultaneously preparing to double its advanced AI chip output, while Moore Threads Technology debuts Friday in Shanghai, underscoring growing domestic competition to replace American hardware in China's AI sector.
Production Targets and Manufacturing Challenges
Cambricon's 2026 target represents more than triple the 142,000 AI chips Goldman Sachs estimates it will produce in 2025. The ramp-up depends heavily on securing manufacturing capacity at SMIC, where Huawei and other domestic chipmakers are also competing for production slots.
Manufacturing quality poses significant obstacles. Cambricon's top-tier 590 and 690 chips currently achieve yields of just 20%, meaning four out of five silicon dies are unusable. By comparison, Taiwan Semiconductor Manufacturing Co.(TSMC) achieves estimated yields of at least 60% with its 2-nanometer process, which analysts consider three generations or seven years ahead of SMIC's technology.
Memory supply represents another potential bottleneck. High-bandwidth memory (HBM) chips required for AI accelerators remain challenging for Chinese companies. Huawei's latest 910C AI accelerators still rely on memory chips from SK Hynix and Samsung Electronics, highlighting persistent technological gaps.
Market Dynamics and Investor Appetite
The broader Chinese chip sector shows strong momentum. Moore Threads' Shanghai listing drew retail subscriptions oversubscribed by 4,000 times, with shares priced at 114.28 yuan—the highest level for an A-share IPO this year. MetaX, another fabless GPU designer, set its IPO price at 104.66 yuan, targeting 4.2 billion yuan in proceeds.
Bloomberg Intelligence analyst Robert Lea noted that while ramping production has increased domestic AI accelerator supply, availability remains insufficient to meet surging demand. "This imbalance is exacerbated by low production yields on advanced 7-nm chips from SMIC, which likely will remain a key bottleneck for the foreseeable future," Lea said in a note.
Bernstein analysis estimates China's AI chip supply could surpass domestic demand by 2028, suggesting plans for global market expansion. Huawei is expected to capture 50% market share by 2026, significantly eroding Nvidia's position in China.
Nvidia's China Dilemma
Nvidia CEO Jensen Huang said in November the company is effectively blocked from selling into China, creating openings for domestic competitors. While the Trump administration is considering permitting sales of Nvidia's H200 chips, Beijing's stance on broad adoption remains unclear.
Speaking at the Center for Strategic and International Studies in Washington on Wednesday, Huang warned that Huawei has an "AI Belt and Road" plan, similar to its 5G strategy.
"We shouldn't concede the entire market to them... we ought to go compete for it," Huang said. "They'll definitely diffuse the Chinese technology as quickly as possible, because they understand that the sooner you get there, the sooner you build the ecosystem on top, the sooner you become an essential part of that ecosystem."
Chinese President Xi Jinping and the Communist Party have criticized U.S. efforts to restrict American AI chip sales to China and explicitly stated their goal of building domestic capabilities to reduce vulnerability to Washington's political priorities. Beijing's push for self-reliance could draw $98 billion in combined government and corporate spending this year alone, with Cambricon—founded in 2016 by computer scientist Chen Tianshi—regarded as a primary beneficiary.
