As U.S. artificial intelligence (AI) chip corporations are struggling in the Chinese market—either scaling back operations one after another or accepting a "0% market share" scorecard—China's homegrown fabless (semiconductor design) corporations are swiftly filling the gap and reaping windfall gains. The warning by Nvidia Chief Executive Officer Jensen Huang that "U.S. semiconductor sanctions against China will ultimately hand over the entire Chinese chip market to local competitors" is becoming reality.
◇ Opportunity created by U.S. sanctions… Cambricon and Moore Threads on a tear
According to the industry on the 20th, Chinese fabless corporations are showing steep growth in earnings under the U.S. sanctions environment. Cambricon, dubbed the "Chinese Nvidia," disclosed on the 17th (local time) that its third-quarter (July–September) revenue this year reached 1.7268 billion yuan (about 344.0 billion won). That is a surge of nearly 14 times from a year earlier. Net profit was 567.0 million yuan (about 113.0 billion won), turning to a profit from a loss of 194.0 million yuan (about 39.0 billion won) in the third quarter last year.
Founded in 2016 with support from the Chinese Academy of Sciences (CAS), the country's largest national research institute, Cambricon is a first-generation local AI chip startup. Early on, it focused on designing AI chips for Huawei smartphones, but when Huawei moved to develop its own chips, Cambricon shifted its business toward high-performance AI chips for data centers. It now competes in the domestic AI chip market not only with Nvidia but also with its former client Huawei. After turning profitable for the first time in the fourth quarter last year, when U.S. sanctions against China ramped up, Cambricon has posted profits for three consecutive quarters. Riding the earnings uptrend, Cambricon's share price has risen more than 91% this year.
Moore Threads, a GPU (graphics processing unit) specialist fabless founded in 2020 by a former Nvidia China country manager, has also drawn market attention recently by passing the Shanghai Stock Exchange STAR Market IPO (listing) review. According to the company's prospectus, Moore Threads' revenue in the first half of this year was 701.0 million yuan (about 140.0 billion won), surpassing its total revenue over the past three years (about 608.0 million yuan; about 120.0 billion won). About 95% of its first-half revenue came from AI-related products such as AI computing clusters.
◇ As U.S. semiconductors retreat, "chip self-reliance" accelerates… Production bottlenecks remain
Meanwhile, the poor performance of leading U.S. semiconductor corporations in China is deepening across sectors, including AI chips and memory. CEO Jensen Huang recently said at a Citadel Securities event in New York that due to U.S. export controls, Nvidia's market share for advanced chips in China fell from 95% to 0%, adding, "We are currently 100% excluded from the (advanced chip) market in China." The situation is no different for AMD and Intel, which are chasing Nvidia. With their exports of advanced AI chips to China blocked, they are trying to find a way forward with lower-spec chips tailored for the Chinese market, but they are losing out to local corporations.
It is not only the AI chip sector. U.S. memory semiconductor corporation Micron received a Chinese government ban two years ago on purchases for "critical infrastructure" amid U.S.-China tensions. It is reported to have recently decided to withdraw after concluding that it would be difficult to revive its business for data center server chips in China. China, which accounted for 16% of total revenue just three years ago, has now fallen to around 7%.
As China's big tech companies begin deploying domestic chips in real-world operations, Chinese fabless corporations have entered a virtuous cycle that allows them to validate and advance their technology in earnest. In particular, in this "domestic substitution" race, Cambricon is winning favor from big techs by leveraging "software compatibility." According to the Financial Times (FT), engineers at China's ByteDance said, "Thanks to software compatibility, Cambricon's products are easier to use than Huawei's Ascend chips." This stems from Cambricon's focus over the past four years on improving software so that AI models trained in Nvidia's CUDA ecosystem can run easily on its own "SiYuan" chips, pouring 5.6 billion yuan (about 1.064 trillion won) into R&D. On the back of this competitiveness, Goldman Sachs projected Cambricon's market share in China's AI chip market to grow from 3% this year to 11% by 2028.
However, industry watchers say the long-term growth of Chinese fabless corporations depends on whether SMIC, China's largest foundry (semiconductor contract manufacturing) and their manufacturing partner, can expand its advanced chip production capacity. With bottlenecks intensifying as China's AI chip supply fails to keep up with demand, the key variable for these corporations' growth will be how quickly SMIC—the virtually only company in China capable of mass-producing advanced chips—can ramp up output for process nodes at 7 nanometers (nm; 1 nanometer is one-billionth of a meter) and below.
Some speculate that the Chinese government is getting involved in SMIC's production allocation to alleviate bottlenecks. The claim is that the Chinese government instructed SMIC to allocate separate capacity for Cambricon, rather than assigning all advanced nodes to Huawei. A semiconductor industry official said, "Production allocation directives to SMIC signal that the Chinese government intends to deliberately grow Cambricon alongside Huawei," adding, "This appears to be a strategy to strengthen the entire ecosystem by fostering internal competition, instead of relying on a single champion, Huawei."