Huawei Erodes Nvidia's China AI Chip Share, Gap Narrows to 6 Percentage Points
nashnova research
Huawei shipped 572,000 Ascend AI chips in H1 2026, narrowing its market-share gap with Nvidia from 31 percentage points to just 6 — as export controls and domestic substitution reshape China's $50 billion AI chip market.
How did Nvidia's lead shrink by 80% in a single year?
IDC data shows Huawei shipped 572,000 Ascend chips in H1 2026, capturing 28% of China's AI accelerator market. Nvidia shipped roughly 678,000 units for a 34% share.
A year ago the gap was 31 percentage points. Now it is 6. This means → Nvidia's position in China has gone from dominant to barely ahead.
AMD shipped just 46,000 units — effectively a non-factor. The two U.S. chipmakers hold a combined 36%, while domestic suppliers — Huawei, Cambricon, Kunlunxin, Hygon and others — collectively account for roughly 60%.
Non-GPU chips outsold GPUs for the first time — what does that signal?
In H1 2026, non-GPU accelerators — ASICs, FPGAs, and NPUs (chips designed for specific tasks, unlike the do-everything GPU) — shipped 1.016 million units, up nearly 80% year-on-year, overtaking GPUs at 991,000 for the first time.
In plain terms = the market is shifting from general-purpose GPUs used for AI training to specialized chips built for AI inference — training teaches the model; inference puts it to work, and that is where large-scale commercial deployment happens.
Huawei, Alibaba's T-Head, Kunlunxin, and Cambricon together control 95% of China's non-GPU accelerator market. This reflects an overwhelming domestic advantage in the inference chip lane.
Where is Nvidia blocked?
U.S. export controls bar Nvidia's latest Blackwell chips from China entirely. The H200 — the most advanced chip cleared for export — contributes less than 1% of Nvidia's data-center revenue.
CEO Jensen Huang has repeatedly criticized the controls, warning they risk handing China's $50 billion AI chip market to Huawei.
China now accounts for roughly 9% of Nvidia's revenue. Three years ago the figure was at least one-fifth of data-center revenue alone. This means → China's revenue contribution to Nvidia has already shrunk sharply and is still falling.
Does Huawei face bottlenecks of its own?
The Ascend 910C is constrained by yield issues at SMIC's 7 nm process node and a shortage of HBM — high-bandwidth memory, the component that moves data at speed inside AI chips.
ByteDance and Alibaba have already exhausted their Ascend inventory. Tencent and Baidu are still waiting for deliveries. Huawei's rotating chairman Xu Zhijun has acknowledged that supply capacity is the main bottleneck limiting Ascend shipment growth.
Put simply = Huawei's problem is not demand — it is manufacturing. Orders are running ahead of production.
How does Nvidia itself view the China outlook?
CFO Colette Kress, reporting FY2027 Q2 results, guided next-quarter revenue to $108 billion — above the Street consensus of roughly $104.2 billion.
But the guidance explicitly excluded China data-center revenue. This means → China has effectively dropped out of Nvidia's near-term earnings outlook — not shelved temporarily, but written off from the forecast altogether.
Morgan Stanley projects China's AI chip self-sufficiency rate will rise from 35% in 2025 to 45% in 2026, then 68% in 2027 and 77% in 2028. If that trajectory holds, Nvidia's window to re-enter the China market narrows further with each passing quarter.
市场有风险，内容仅供研究参考，不构成投资建议。