China is seeking triple production of artificial intelligence chips in the race with the United States

Chinese makers in China are seeking three times the country’s total output of artificial intelligence treatments next year, as Beijing is racing to develop the most advanced artificial intelligence.

One of the manufacturers for manufacturing for the production of AI processors in Huawei is scheduled to start production as soon as the end of this year, while two others are scheduled to be launched next year.

While the new plants are designed to support Huawei in particular, it is not clear exactly who has them. Huawei denied the existence of plans to launch his Fabs and did not provide more details.

Chinese companies are also racing to develop the next generation of artificial intelligence chips that can be adaptable to the standard DibsicWho emerged as the leadership of Amnesty International in the country.

The latest Huawei products are seen as among those that will meet Deepseek requirements.

The joint capacity of these three new factories, which was once fully condensed, can exceed the current product of similar lines in the SMIC, a leader in China, according to those people who have effort.

People added that SMIC has also planned to double its ability next year to make chips at 7 the most productive type in China. Huawei is the largest SMIC customer for the processor.

As a result, smaller Chinese chip designers such as Cambricon, Metax and Biren will be able to get much larger allocations for SMIC capacity, stimulating competition for the fast -growing China market left by NVIDIA after the American export ban.

Semiconductor At the heart of commercial tensions between Beijing and Washington. The United States is restricting access to the best artificial intelligence treatments by the world leader NVIDIA, in an attempt to curb Chinese efforts to develop artificial intelligence.

“The local product will not be a problem for a long time, especially with all the next capacity on the Internet next year,” said one of the Chinese Chinese executives.

Deepseek announced last week that its models are now using a kind of FP8 data format, designed to adapt to the next generation of local chips, without specifying the resource.

Stock prices It rose after the announcement.

Huawei’s 910D in addition to 690 Cambricon is seen as leading products designed to support Deepseek’s favorite standards, while smaller Chinese conflict makers also accelerate the development of their releases.

Deepseek chose this data format because it increases the efficiency of the devices, at the expense of accuracy, which may create a way to help the Chinese artificial intelligence players compete with international competitors even if their chips are generations behind the leading NVIDIA offers.

“If we succeed in developing these Chinese chips and improving them to train and operate Chinese models in a constantly advanced Chinese ecosystem, we will one day look at this transformation as a more important moment.” “This unprecedented alignment can be compensated for our ability for the least advanced devices.”

However, such an effort requires a year cooperation between computing chips with those who make devices for memory and communication, as well as software tools needed to support them.

China is also pushing in other critical areas, such as memory chips, which are currently dominated by Samsung, SK Hynix and Micron – are all subject to American export controls.

The pioneering Chinese player CXMT tests samples of high -frequency memory products “HBM3” with a goal to launch next year, according to two people familiar with the disease. The product is only one generation behind the most advanced memory products used by NVIDIA chips.

While Deepseek did an early work on Chinese chips in small samples, to demonstrate that they are technically viable, the current model training of the group is done on NVIDIA chips collections, according to The Financial Times. This can change again for local Chinese chips that become adaptable to its standard and improving its performance.

“The necessity generates innovation,” said an investor in many semiconductors and startups. “Companies in Silicon Valley will not even care about doing what Deepseek does because they can get the largest number of advanced NVIDIA chips as they want.”

Beijing puts important political and financial support behind such payments. State Council It is called this week to adopt greater artificial intelligence in the country, as well as “integrated and coordinated development and development of technology driven by artificial intelligence, implementation of engineering, and products marketing.”

Cambricon, the runway in Shanghai, obtained approval to raise about $ 600 million this year. Four of the artificial intelligence makers, including Biren and Metax, seek to put it to the public as soon as the end of this year, after collecting about 3 billion dollars combined in pre -subscription rounds.

If it succeeds, the batch needed to create an alignment between AI companies in China and software companies can meet the Deepseek Liang Wenfeng founder to challenge the global lead in NVIDIA.

Liang told the Chinese newspaper last year: “NVIDIA is not created by one company, but it is a cooperative effort by the entire Western community and industry,” Liang told the Chinese newspaper last year. “China’s artificial intelligence development needs the same ecosystem. China needs someone to stand at the forefront of this development.”

SMIC, Cambricon, Deepseek, Cxmt, Biren and Metax for FT order for comment.