Why China is racing to triple AI chip output in response to US export bans

Why China is racing to triple AI chip output in response to US export bans

China's semiconductor industry is accelerating efforts to expand self-sufficiency after US curbs starved it of Nvidia's best processors. Chinese chipmakers plan to nearly triple the nation's overall output of artificial intelligence chips next year, with at least three new production facilities in the works to supply Huawei, executives with knowledge of the plans said. The plants will match, and in some instances surpass, existing production capacity at Semiconductor Manufacturing International Corporation (SMIC), China's biggest foundry, the Financial Times said.

Huawei and DeepSeek's position in the new ecosystem

Huawei has become the anchor for China’s AI hardware push, with its latest 910D processor designed to meet the technical requirements of DeepSeek, the country’s leading AI start-up. DeepSeek has promoted a new FP8 data format that prioritizes efficiency over precision, allowing domestic chips to train large models despite lagging behind Nvidia’s most advanced GPUs. Huawei’s alignment with DeepSeek is seen as a turning point, creating a unified hardware-software ecosystem to compete against US dominance.

Adding capacity at SMIC and beyond

SMIC is doubling capacity for 7-nanometre processors, the most advanced nodes China is currently capable of mass-producing. The expansion will allow Cambricon, Biren and MetaX and other smaller Chinese chip firms to secure more allocations and compete more aggressively in the domestic AI market. In combination with Huawei-linked fabs, the increased output might render China less reliant on imported processors and more influential in its negotiations with the US.

Memory and enabling technologies

China is also spending big on supporting technologies like memory and connectivity. Industry leader CXMT will come out with HBM3 high-bandwidth memory chips in a year's time, closing the gap with global leaders Samsung, SK Hynix and Micron. Advances in memory and interconnects will be key to making China's AI chips cost-effective at scale since AI models need huge data throughput in addition to raw compute power.

Market reaction and political support

Shares of Chinese chipmakers like Cambricon and SMIC surged following DeepSeek's guarantee that its models began using the FP8 standard. Beijing has bet big on the sector, with the State Council calling for the accelerated adoption of AI across industries and clearing billions of dollars for investing in chipmakers. Cambricon has already raised $600 million, while Biren and MetaX are gearing up for IPOs after having collectively raised $3 billion in pre-listing rounds.

The greater geopolitical interests

The US has blocked the export of high-speed processors to impede China's AI drive, but situation is igniting innovation in China. As analysts point out, the integrated approach to chip design, manufacturing, memory, and software recalls the integrated ecosystem which propelled Nvidia to market dominance. Liang Wenfeng, founder of DeepSeek, has framed the challenge as national, arguing that China needs an equivalent "whole-of-industry" response to bridge the gap with the West in AI.

What's next

Although China is still a few generations behind in chipmaking technology, its fast growth suggests that domestic production will no longer be the limiting factor for long. If coordination between Huawei, SMIC, DeepSeek, and other players goes as planned, Beijing will be in a position to accelerate the advancement of AI independent of US sanctions. The outcome will decide not only the fate of AI competition but also the direction of semiconductor supply chains worldwide.