Chinese chip makers are looking to triple the country’s total production of artificial intelligence processors next year.
One manufacturing plant specializing in the production of Huawei AI processors is expected to start production soon at the end of this year, but two more will be released next year, with the two people who know the plan.
The new plant is designed to specifically support Huawei, but it is not clear who owns it exactly. Huawei refused to plan to launch its own fab and did not provide further details.
Chinese companies are also competing to develop next-generation AI chips that can adapt to the standards proposed by Deepseek, which has emerged as a major AI startup in the country.
Huawei’s latest products are considered to meet Deepseek’s requirements.
The total capacity from these three new plants could be exceeding the current total production of a similar line at China’s leading fab, Semiconductor Manufacturing International Corporation (SMIC).
People added that Smic is planning to double its capabilities next year to make chips at 7 nanometers. This is China’s most advanced mass production type. Huawei is currently the largest customer of Smic on such a processor line.
As a result, Chinese chip designers such as Cambricon, Metax and Biren can get a much larger allocation of Smic’s capabilities, spurring competition in the fast-growing Chinese market left behind by Nvidia after the US export ban.
Semiconductors are at the heart of trade tensions between Beijing and Washington. The US is restricting access to top AI processors created by global leader Nvidia to curb China’s efforts to develop AI.
“Domestic production will not be a problem for a long time. In particular, all capabilities will be online next year,” said the executive of the Chinese chipmaker.
Last week, Deepseek announced which type of FP8 data format its model uses. It is designed to adapt to the next generation domestic chips without specifying a supplier.
After the announcement, stock prices of listed Chinese semiconductor companies such as Cambricon and Smic have skyrocketed.
Huawei’s 910D and Cambricon 690 are considered major products tailored to support Deepseek’s preference standards, but several Chinese ship manufacturers are also accelerating the development of versions.
Deepseek has chosen this data format. This is because it could increase hardware efficiency at the expense of accuracy and create ways to help Chinese AI players compete with international rivals even as they become the generation of Nvidia’s leading products.
“If we succeed in developing and optimizing these Chinese chips and successfully train and implementing Chinese models in a continuous evolving Chinese ecosystem, one day we will look back at this shift as an even more important moment of deep seek,” said the chipmaker executive. “This unprecedented alignment could potentially compensate for less advanced hardware capabilities.”
However, such efforts will require years of collaboration between those who are creating hardware for memory and connections and the manufacturer of computing chips with the software tools needed to support them.
Recommended
China is also pushing in other key areas, including memory chips currently dominated by Samsung, SK Hynix and Micron. Everything is subject to US export controls.
The leading Chinese player CXMT is testing a sample with a target that will launch a sample of the “HBM3” high-bandwidth memory product next year, according to two people with knowledge of the issue. This product is just one of the most advanced memory products used by Nvidia chips.
Deepseek has done early work on Chinese chips with small samples, but to prove that it is technically feasible, the group’s current model training is being done on Nvidia chip clusters, reports Financial Times. This could allow China’s domestic chips to adapt to standards and improve performance.
“The need creates innovation,” said investors at multiple Chinese semiconductors and large-scale language models. “Companies in Silicon Valley don’t even care about doing what Deepseek is doing.
Beijing provides important political and financial support behind such pushes. The Council of State this week called for “integrated and coordinated development of AI-driven technology research and development, engineering implementation, and product commercialization” to promote the adoption of AI in the country.
Shanghai-listed Cambricon has been approved to raise approximately $600 million this year. Four small AI chip makers, including Biren and Metax, are about to make an announcement soon at the end of the year, raising around $3 billion in the pre-IPO round.
If successful, push to create an alignment between Chinese AI hardware and software companies will meet with Deepseek founder Liang Wenfeng’s call to challenge Nvidia’s global lead.
“Nvidia’s leads are not created by a single company, but a joint effort across the Western community and industry,” Liang told Chinese media last year. “China’s AI development requires the same ecosystem. China needs someone to be at the forefront of such development.”
Smic, Cambricon, Deepseek, CXMT, Biren, and Metax did not respond to FT’s request for comment.
