What happened
According to IDC data cited by Growth Dragons, Nvidia's market share in China's AI accelerator shipments dropped from 55% in 2025 to 34% in the first half of 2026. Huawei's share increased to 28%, narrowing the gap to six percentage points. This shift coincides with a surge in non-GPU accelerator shipments, which overtook GPU shipments for the first time, reflecting a market pivot toward tasks where Chinese chips compete on cost and efficiency.
Growth Dragons reports that IDC data indicates a significant shift in China's AI accelerator market during the first half of 2026. Nvidia's shipment volume decreased by 44% year-on-year to 678,000 units, reducing its market share from 55% in 2025 to 34%. In contrast, Huawei nearly doubled its shipments to 572,000 units, capturing 28% of the market. Other domestic players, including Alibaba's T-Head (7%), Cambricon (5%), Baidu's Kunlunxin (4%), and Hygon (3%), collectively held about 47% of the market share.
The source highlights a change in the type of chips being purchased. Shipments of non-GPU accelerators, designed for specific AI tasks rather than general-purpose graphics processing, rose nearly 80% to 1.016 million units. This category overtook GPU shipments, which fell 26% to 991,000 units. Chinese suppliers hold approximately 95% of this non-GPU segment, leveraging advantages in energy efficiency and system-level performance for workloads.
To compensate for manufacturing limitations, Chinese firms are adopting cluster-based architectures. Huawei's Atlas 950 SuperCluster, featuring 256,000 chips, is currently being deployed, with designs scalable to one million chips. This approach relies on high-speed networking and software optimization rather than individual chip superiority. Additionally, SMIC, China's largest contract chipmaker, reported a 36% year-on-year revenue increase to US$3.0 billion in Q2, with factories operating at 93.7% capacity.
Source details: growthdragons.substack.com ↗
Why it matters
The decline in Nvidia's dominance signals a structural shift in China's AI infrastructure toward domestic supply chains. As becomes the primary workload, Chinese manufacturers like Huawei and Alibaba can compete effectively without matching Nvidia's raw training power. This trend reduces reliance on foreign hardware and accelerates the development of a self-sufficient Chinese AI ecosystem, impacting global chip market dynamics and export control strategies.
The data suggests that China's AI industry is transitioning from a dependency on imported high-end GPUs to a domestic ecosystem optimized for . Since inference accounts for a growing portion of AI workloads, Chinese chipmakers can compete on total cost of ownership and energy efficiency, areas where they have made significant strides.
This shift has implications for global AI hardware markets, as it reduces the potential revenue for foreign suppliers like Nvidia in the Chinese market. It also demonstrates the effectiveness of domestic substitution strategies, potentially influencing other nations' approaches to AI infrastructure security and supply chain resilience.
The rise of non-GPU accelerators indicates a maturing AI market where specialized hardware is preferred for specific tasks. This trend could lead to a more fragmented but efficient global AI hardware landscape, with different regions optimizing for different workload types.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
Monitor the deployment scale of Huawei's Atlas 950 SuperCluster and the performance of non-GPU accelerators in real-world tasks. Also track SMIC's capacity utilization and whether Beijing expands limited Nvidia H200 chip purchases, which could further influence the balance between domestic and imported AI hardware.
The actual performance and adoption rate of Huawei's Atlas 950 SuperCluster in large-scale data centers will be a key indicator of the viability of cluster-based solutions for AI .
Further developments in SMIC's manufacturing capacity and yield rates for advanced nodes will determine the long-term sustainability of China's domestic chip supply chain.
Policy changes regarding the import of Nvidia H200 chips could alter the competitive dynamics, as limited access to foreign hardware may continue to support domestic alternatives or create a hybrid market structure.