Nvidia's China AI Chip Market Share Drops by One-Third as Huawei Nearly Doubles Shipments
nashnova research
Nvidia's share of China's data-center AI chip market fell 34% in H1 2026, per IDC data cited by Reuters, while Huawei nearly doubled shipments to 572,000 units — export controls and domestic substitution are reshaping China's AI compute landscape in parallel.
How much ground has Nvidia actually lost?
In H1 2026, Nvidia shipped 678,000 accelerators into China's data-center AI chip market — down 44% year-on-year, with market share shrinking 34%.
This means → U.S. export controls and Beijing's push for domestic alternatives are squeezing Nvidia from both sides simultaneously.
Industry estimates put Nvidia at roughly 8% of China's AI chip market today; in 2022 the figure was 95%. In plain terms = in three years, near-monopoly has collapsed to single digits.
What is driving Huawei's surge?
Huawei shipped 572,000 AI accelerators in the same period, capturing 28% of China's market — nearly double the roughly 300,000 units shipped a year earlier.
Huawei said this month that its Ascend accelerators have overtaken Nvidia in Chinese AI chip sales. Some industry estimates now place Ascend at about 50% market share.
This reflects a shift from fallback option to primary supplier — the Ascend 910C chip sold out by end of August, signaling active downstream migration.
Non-GPU accelerators outsold GPUs for the first time — why does that matter?
China's total AI accelerator shipments topped 2 million units in H1 2026. GPU shipments fell 26% to 991,000; non-GPU accelerators — chips designed specifically for AI without traditional GPU architecture — surged nearly 80% past 1 million, overtaking GPUs for the first time.
Domestic vendors drove nearly all of it: Huawei, T-Head (Alibaba), Cambricon, and Kunlun (Baidu) together accounted for 95% of non-GPU shipments. Of the 11 Chinese chip firms IDC tracked, 9 posted year-on-year growth.
This means → China's AI compute stack is shifting from "buy Nvidia GPUs" to "deploy domestic non-GPU accelerators." This is not a single-vendor swap — it is an entire technology-path migration.
Is there still demand for Nvidia chips in China?
Yes — and they remain in short supply. Nvidia's most advanced B200 and B300 chips are explicitly banned from export to China, yet smuggling and routing through Southeast Asian data centers persist.
The U.S. later allowed small-batch H200 exports, but Nvidia failed to meaningfully recover share. In plain terms = the permitted volume is far too small to offset the ban's impact.
This reflects a contradictory reality: China's market still values Nvidia's performance edge, but the supply channel has been sharply narrowed by controls.
Can domestic chips fully replace Nvidia?
Not yet. Chinese chips still lag in yield rates, advanced packaging, and high-bandwidth memory (HBM) — they cannot match Nvidia across the board on performance.
But the migration trajectory is clear: DeepSeek's CEO said last month the company plans to move model training onto domestic chips.
This means → export controls are pushing China's AI compute from reliance on single high-performance chips toward system-level integration. Whether this structural shift holds depends on how fast domestic chipmakers can close gaps in yield and ecosystem maturity.
市场有风险，内容仅供研究参考，不构成投资建议。