Around 600,000 Nvidia H200 AI chips are stalled in Hong Kong awaiting mainland China approval, creating acute supply constraints that have sent eight-card server prices to 6 million yuan—roughly $900,000—by late October.
The H200, Nvidia's latest GPU for large language model training and inference, remains in short supply despite strong demand from Chinese AI firms. Several major cloud providers including Baidu and Alibaba have each secured approval for roughly 10,000 units—a fraction of the backlog—suggesting a controlled, gradual release by Chinese regulators.
The price premium reflects the market's hunger for compute capacity. Companies with approved access to H200s gain a competitive edge in offering AI services and development tools across China's cloud sector.
For Nvidia, delays in a market representing billions in potential annual revenue can ripple through its data center segment—the company's fastest-growing business. Export restrictions have fragmented the global AI chip market, forcing Chinese enterprises to bid up prices for whatever supply reaches them while Nvidia faces a bifurcated sales pipeline.
