Podcast Episode
The model employs a sophisticated hybrid architecture combining a 9 billion parameter autoregressive model with a 7 billion parameter diffusion decoder. Training was conducted on Huawei's Ascend Atlas 800T A2 servers using the MindSpore framework, demonstrating that advanced AI development can proceed without access to American semiconductors.
Zhipu, which was placed on a US blacklist last year, stated that the achievement proves the feasibility of developing high performance multimodal generative models on a domestically developed full stack computing platform. The announcement sent the company's shares surging approximately 17 percent to 222 Hong Kong dollars and 60 cents, whilst other domestic chip manufacturers also saw significant gains.
The rally reflects growing investor confidence in China's semiconductor ecosystem and its ability to support advanced AI workloads. Since its listing, Zhipu's shares have climbed more than 80 percent as enthusiasm builds around China's AI industry and domestic chip ambitions.
The restrictions were implemented just one day after the US Commerce Department changed its licensing policy for these chips from presumption of denial to case by case review, fulfilling an agreement between political leaders. The US approval included requirements such as proof of sufficient domestic supply and strict security conditions.
According to sources, the wording from Chinese officials is so severe that it essentially constitutes a ban for now, though the situation might evolve in the future. The government indicated it would only approve purchases under special circumstances, such as for university research and development laboratories.
Zhipu's demonstration with Huawei chips represents notable progress in this direction. The company has been working with domestic chipmakers including Cambricon to adapt and develop AI models since being blacklisted. The successful training of GLM-Image suggests China's semiconductor ecosystem is advancing toward supporting sophisticated AI workloads without foreign hardware.
Chinese AI startups are also reaching public markets faster than their American counterparts, benefiting from supportive domestic policies and clearer paths to enterprise revenue. Shanghai Biren Technology's shares surged nearly 120 percent on the Hong Kong Stock Exchange following a 717 million dollar initial public offering that saw its retail portion subscribed more than 2,300 times.
The developments challenge assumptions about the necessity of the most advanced foreign chips for training sophisticated AI models. Whilst domestic alternatives may require more engineering effort and innovation, Chinese companies are demonstrating that viable pathways exist to develop competitive AI capabilities using homegrown semiconductor technology.
The situation highlights the complex dynamics shaping the global AI and semiconductor landscape, where technological capability, industrial policy, and geopolitical strategy intersect. As both the United States and China continue developing their respective approaches to AI governance and chip production, the coming months will likely see further tests of each nation's technological resilience and strategic positioning.
China Demonstrates AI Self-Reliance as Semiconductor Tensions Escalate
January 15, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Chinese semiconductor stocks rallied sharply this week following a dual demonstration of Beijing's technological independence strategy: the successful training of an advanced AI model on domestic chips, paired with strict limitations on imports of cutting edge American processors. The developments signal China's determination to build self sufficient AI capabilities even as Washington and Beijing navigate complex technology trade relations.
Breakthrough in Domestic AI Development
Chinese AI startup Zhipu announced Wednesday that it has successfully trained GLM-Image, a state of the art open source image generation model, using exclusively Huawei Technologies Ascend chips. The company described this as the first multimodal model of its calibre to complete training entirely on domestically produced hardware, marking a significant milestone in China's push for technological autonomy.The model employs a sophisticated hybrid architecture combining a 9 billion parameter autoregressive model with a 7 billion parameter diffusion decoder. Training was conducted on Huawei's Ascend Atlas 800T A2 servers using the MindSpore framework, demonstrating that advanced AI development can proceed without access to American semiconductors.
Zhipu, which was placed on a US blacklist last year, stated that the achievement proves the feasibility of developing high performance multimodal generative models on a domestically developed full stack computing platform. The announcement sent the company's shares surging approximately 17 percent to 222 Hong Kong dollars and 60 cents, whilst other domestic chip manufacturers also saw significant gains.
Market Response and Stock Rally
Semiconductor Manufacturing International Corporation, China's largest contract chipmaker, rose 2.1 percent in Hong Kong trading, whilst Hua Hong Semiconductor jumped 5.6 percent. Cambricon Technologies and Moore Threads Technology, both viewed as potential domestic alternatives to traditional chip suppliers, also advanced on the Shanghai STAR Market.The rally reflects growing investor confidence in China's semiconductor ecosystem and its ability to support advanced AI workloads. Since its listing, Zhipu's shares have climbed more than 80 percent as enthusiasm builds around China's AI industry and domestic chip ambitions.
Beijing Blocks Advanced Chip Imports
The stock gains came against a backdrop of escalating semiconductor tensions. Chinese customs authorities instructed agents this week that certain advanced AI chips are not permitted to enter the country, according to reports. Technology companies were summoned to meetings where officials told them not to purchase the processors unless necessary.The restrictions were implemented just one day after the US Commerce Department changed its licensing policy for these chips from presumption of denial to case by case review, fulfilling an agreement between political leaders. The US approval included requirements such as proof of sufficient domestic supply and strict security conditions.
According to sources, the wording from Chinese officials is so severe that it essentially constitutes a ban for now, though the situation might evolve in the future. The government indicated it would only approve purchases under special circumstances, such as for university research and development laboratories.
Strategic Self Reliance Push
Beijing's cautious approach reflects its broader strategy of reducing dependence on American technology whilst building domestic capabilities. Major Chinese technology companies including Alibaba and ByteDance have reportedly expressed interest in ordering more than 200,000 advanced chip units each, but officials have signalled they want firms to prioritise locally made processors.Zhipu's demonstration with Huawei chips represents notable progress in this direction. The company has been working with domestic chipmakers including Cambricon to adapt and develop AI models since being blacklisted. The successful training of GLM-Image suggests China's semiconductor ecosystem is advancing toward supporting sophisticated AI workloads without foreign hardware.
Industrial Capacity Expansion
The policy push is backed by significant industrial investment. Huawei, widely regarded as China's leading AI chipmaker, is preparing to sharply increase production of its most advanced semiconductors this year. Cambricon is planning to more than triple its production of AI chips in 2026, whilst Semiconductor Manufacturing International Corporation's 7 nanometre manufacturing capacity is set to double, with Huawei as its primary client.Chinese AI startups are also reaching public markets faster than their American counterparts, benefiting from supportive domestic policies and clearer paths to enterprise revenue. Shanghai Biren Technology's shares surged nearly 120 percent on the Hong Kong Stock Exchange following a 717 million dollar initial public offering that saw its retail portion subscribed more than 2,300 times.
Geopolitical Implications
Analysts suggest Beijing's restrictions on advanced chip imports could be aimed at building leverage ahead of high level diplomatic meetings as both sides navigate an uneasy truce on trade and technology. The approach demonstrates China's willingness to forego access to cutting edge foreign semiconductors in favour of accelerating domestic development.The developments challenge assumptions about the necessity of the most advanced foreign chips for training sophisticated AI models. Whilst domestic alternatives may require more engineering effort and innovation, Chinese companies are demonstrating that viable pathways exist to develop competitive AI capabilities using homegrown semiconductor technology.
The situation highlights the complex dynamics shaping the global AI and semiconductor landscape, where technological capability, industrial policy, and geopolitical strategy intersect. As both the United States and China continue developing their respective approaches to AI governance and chip production, the coming months will likely see further tests of each nation's technological resilience and strategic positioning.
Published January 15, 2026 at 5:56am
