China’s AI chips: TrendForce expects the market share of foreign suppliers in China’s AI server market to decline, countered by the expansion of Huawei, Cambricon, and domestically designed chips.
Article index
- Expectations of a sharp decline in the Chinese market
- The rise of domestic suppliers and national alternatives
- Continued reliance on Nvidia for model training
- Supply uncertainty and the long-term shift
- Global expansion challenges for Chinese tech companies
- FAQs
Expectations of a sharp decline in the Chinese market
The combined market share of Nvidia and AMD in China’s AI server chip market is projected to drop to just 21% in 2026, down from 34% last year, according to new data released by Taipei-based research firm TrendForce during its industry conference in Shenzhen. This projection highlights an acceleration of a multi-year decline driven by stringent US export controls and Beijing’s persistent pursuit of self-sufficiency in AI infrastructure.
The rise of domestic suppliers and national alternatives
Local chip suppliers, led by Huawei and Cambricon, are expected to capture 56% of the market in 2026, up from 46% the previous year, according to the TrendForce report. By adding custom integrated circuits designed by Chinese internet giants such as Alibaba, Baidu, and Tencent—which are expected to account for 23% of the market—domestically manufactured silicon is on track to cover nearly 80% of China’s AI server hardware this year. Another supply chain report indicated that local solutions could capture 90% of the advanced chip segment.
Continued reliance on Nvidia for model training
Despite this purchasing shift, AI model developers in China continue to train on Nvidia hardware. The South China Morning Post reported that high migration costs keep China’s large language model developers reliant on Nvidia chips. Nvidia’s CUDA software platform remains the industry standard, while migrating to alternatives such as Huawei’s Ascend architecture requires extensive script rewriting, adding at least 50% in time and costs.
Supply uncertainty and the long-term shift
Although US authorities have permitted the sale of limited quantities of H200 chips to specific Chinese companies, actual shipments remain subject to approvals from both governments, leaving supply uncertain. This volatility pushes Chinese firms to design domestic chips into their projects from the outset rather than treating them as temporary alternatives. Observers expect capital expenditures by ByteDance, Tencent, Alibaba, and Baidu to grow by over 80% year-over-year in 2026, directing demand straight toward local supply chains.
Global expansion challenges for Chinese tech companies
Experts emphasize that the ability of Chinese firms like Huawei and Cambricon to bridge the technological gap domestically represents a major industrial achievement despite significant challenges in precision supply chains. Although Nvidia still controls about 64% of the global graphics server market, the near-complete loss of the Chinese market constitutes a geopolitical shift reshaping the global semiconductor map.
The figures require careful reading because they measure the AI server market according to specific definitions, and do not mean Nvidia and AMD will disappear from China entirely. Calculated shares also vary depending on whether they include in-house custom chips or processors imported via authorized channels, making annual comparisons more important than a single figure.
This development places China’s AI chips at the heart of the debate, while distinguishing between confirmed results and forecasts that still await further testing or data.
Sources: South China Morning Post
FAQs
Question: What will Nvidia and AMD’s market share in China be by 2026?
Answer: Their combined share is projected to drop to 21% compared to 34% recorded last year.
Question: Which Chinese companies are leading local chip alternatives?
Answer: Huawei and Cambricon are leading the domestic market alongside custom chips from Alibaba, Baidu, and Tencent.
Question: Why do Chinese companies continue to use Nvidia’s CUDA currently?
Answer: Due to the cost of rewriting code and the time required to migrate workloads away from the CUDA ecosystem.
