600,000 Nvidia H200 Chips Held in Hong Kong as Beijing Withholds Import Approval

November 5 — Around 600,000 Nvidia H200 AI chips are sitting in Hong Kong awaiting shipment to the Chinese mainland, Reuters reported citing two sources, as Beijing has not approved large-scale imports despite strong demand from Chinese tech firms. In recent months only small batches have been cleared, with some internet companies receiving quotas of about 10,000 chips each.

The cost of the stockpile is already visible. A server with eight H200 cards was quoted above 6 million yuan in the mainland as of late October — roughly triple its price a year earlier, per the sources — and gray-market prices have surged in tandem. Every day the chips sit in Hong Kong warehouses, the scarcity premium climbs; the cadence of import approvals now directly sets the mainland’s compute price curve.

The pace of approvals reflects a two-sided calculation. H200’s clearance for China sales is itself recent and Washington’s posture keeps shifting; Beijing, for its part, wants the compute but hesitates to re-bind its industry chain to a single supplier at the critical layer — while domestic accelerators are ramping, and quota rationing objectively buys them an integration window.

Market effects are immediate. Quota winners bank compute headroom; those without approval pivot to three substitutes — leasing existing capacity, buying domestic-chip clusters, or moving demand offshore to data centers abroad, as Moonshot AI is reportedly doing. The three paths price differently but all carry an extra charge for policy uncertainty.

600,000 chips is not a rounding error — at eight per server, that’s the silicon for roughly 75,000 high-end AI servers. When and how fast the batch clears will shape mainland data-center build-outs and model-training schedules; for Nvidia, the timetable for cashing in the China market is no longer in its own hands.

Watch two variables: approval cadence and domestic chips’ order books — either loosening rewrites the price curve.