Chinese Customers Order 6 Million NVIDIA RTX Chips, Only ~1% Deliverable Within Four Months
nashnova research
Alibaba, Tencent, and ByteDance together ordered 6 million Nvidia RTX Pro 5500 chips worth roughly $70 billion, yet expect to receive only tens of thousands — about 1% — within four months, exposing the structural AI-compute shortage gripping China.
Six million chips ordered — why only 1% delivered?
The RTX Pro 5500 is currently the most powerful Nvidia AI chip China can legally buy. Stronger models — B300, H200, H100 — are all blocked by U.S. export controls.
Alibaba, Tencent, ByteDance and others ordered roughly 6 million units at over $11,000 each, totaling about $70 billion.
This means → demand is real and enormous, but Nvidia simply cannot deliver at scale: only tens of thousands of chips — under 1% of the order — are expected within four months.
When will bulk shipments arrive?
Only two delivery batches are confirmed so far: a first tranche of several thousand chips by year-end, and tens of thousands more around February next year.
One key buyer's internal estimate projects cumulative deliveries of about 700,000 units by the end of 2027 — assuming Nvidia can ramp production on schedule.
In plain terms = even under the most optimistic scenario, China will have received roughly 12% of its order after two and a half years. The gap remains vast.
Where is the production bottleneck?
Nvidia relies on TSMC — the world's largest chip foundry — to manufacture its AI chips. TSMC is shifting more capacity toward Nvidia's next-generation Rubin architecture.
This means → the RTX Pro 5500 is being deprioritized on the production line. Nvidia isn't refusing to sell; the fabs simply can't keep up.
An Nvidia spokesperson declined to comment on specific delivery figures, saying only that supply is constrained by both U.S. export controls and China's own restrictions on U.S. imports.
Can domestic alternatives fill the gap?
In the near term, Chinese hyperscalers still depend on homegrown options: Huawei, Cambricon, and T-Head (Alibaba's chip unit).
But Huawei raised prices on its most advanced AI accelerator by roughly 60% around September this year; Cambricon's latest earnings also flagged capacity constraints.
This reflects a deeper bottleneck in China's domestic chip industry: U.S. controls on advanced chipmaking equipment and a persistent shortage of HBM — high-bandwidth memory, a critical component in AI chips — are capping how fast local fabs can scale.
What does this mean for the U.S.–China compute gap?
A $70 billion order book underscores the urgency of China's AI-compute demand. Beijing has signaled approval for Alibaba, ByteDance and others to buy the chip.
Yet a 1% near-term delivery rate means the Nvidia channel cannot close the gap anytime soon.
In plain terms = the trajectory of the U.S.–China AI-compute divide now hinges on whether domestic alternatives can match both the quality and the volume — the single most consequential variable over the next two years.
市场有风险，内容仅供研究参考，不构成投资建议。