[News] China Reportedly to Triple AI Chip Output Next Year With Fabs Serving Huawei, Cutting NVIDIA Reliance

According to Financial Times, China aims to triple its overall AI processor output next year. The report, citing sources, notes that one fab dedicated to manufacturing Huawei’s AI chips is slated to begin production by the end of this year, with two additional plants expected to come online in 2026.

As the report highlights, the new facilities are intended to supply Huawei, but their ownership remains uncertain, and the company denies plans to build its own fab.

Notably, the report, citing sources, points out that once fully ramped, the combined capacity of these three plants could surpass the current output of comparable lines at Semiconductor Manufacturing International Corporation (SMIC), China’s leading foundry.

The report, citing sources, also notes that SMIC plans to double its 7-nanometre production capacity next year — the most advanced process currently in mass production in China. While Huawei is at present its largest customer for these processor lines, the report further indicates that this move will allow smaller Chinese chipmakers such as Cambricon, MetaX, and Biren to secure larger shares of SMIC’s capacity, intensifying competition in China’s rapidly expanding market.

China’s AI Chip Drive Gains Momentum with DeepSeek

The report further notes that DeepSeek announced last week its models had adopted an FP8 data format tailored for the next generation of domestic chips, though it did not disclose the supplier. This move is expected to provide a lift to Chinese semiconductor companies such as Cambricon and SMIC, according to the report.

Cambricon reported record earnings in the first half of the year, according to CNBC. The company said revenue surged more than 4,000% year-on-year to 2.88 billion yuan ($402.7 million), while net profit reached an all-time high of 1.04 billion yuan, the report adds.

Huawei’s 910D and Cambricon’s 690 are regarded as leading products aligned with DeepSeek’s preferred standards, while several smaller Chinese chipmakers are racing to develop their own versions, the report notes. Still, as the Financial Times points out, although DeepSeek has conducted small-scale trials on Chinese chips to demonstrate technical feasibility, its current model training continues to rely on NVIDIA clusters.

Beijing is stepping up its drive to strengthen China’s AI chip development. As the report highlights, Shanghai-listed Cambricon won approval this year to raise about $600 million, while four smaller AI chipmakers — including Biren and MetaX — aim to go public as early as year-end after securing roughly $3 billion in pre-IPO funding.

Read more

• [News] China’s Cambricon to Raise RMB 4B via A-Share Issuance for AI Chip and Software Push
• [News] Is Cambricon the Next NVIDIA or Unsustainable Growth Story? Three Key Risks Behind the Epic Rally

(Photo credit: FREEPIK)