Plans to disclose Nvidia’s China chip export licenses as requests hit 3.4 million units
Congressional materials show regulators approved licenses for nearly 900,000 AI chips, while Nvidia says it has stopped shipping licensed products to China
Rep. Brian Mast plans to disclose the export licenses tied to Nvidia’s sales of AI chips to China. The disclosure comes as requests to ship Nvidia and equivalent AI chips to the country have reached 3.4 million units.
Washington approved licenses covering nearly 900,000 of them. That is roughly one approval for every four chips requested.
The numbers behind the license pile
The figures surfaced in congressional materials associated with Mast. They show that the Bureau of Industry and Security (BIS), the Commerce Department arm that polices sensitive exports, received requests totaling 3.4 million units. It signed off on licenses for nearly 900,000 chips.
Those approvals were not merely theoretical. Nvidia booked $2.4 billion in sales of licensed Hopper-generation data center products to China in the quarter ending October 25, 2026.
Then came the reversal. On its November 17, 2026 call with investors, Nvidia said it had stopped shipping those licensed products to China.
Even before the halt, the China business was a strange creature. Licensed H200 shipments to China made up a small slice of Nvidia’s data center revenue, under 1% in some prior quarters. Meanwhile, hundreds of thousands of chips sat delayed or on hold in Hong Kong while waiting for further approvals.
A new rule widens the net
One day before Nvidia’s earnings call, BIS tightened the screws again. On November 16, 2026, the agency issued an interim final rule extending licensing requirements to advanced AI chips bound for overseas data centers.
The target is facilities that provide substantial computing capacity to entities in China and its restricted affiliates.
The crypto and markets news you actually need—every morning.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Under the rule, data centers with aggregate processing power above 12 million units will face particular scrutiny. The rule also adds requirements for certain destinations in Southeast Asia and Brazil.
Existing operators must obtain the necessary licenses within 90 days of the rule’s publication.
How we got here
US policy on advanced chip sales to China has swung back and forth. Earlier rules carried a presumption of denial, meaning applications were expected to be rejected by default.
That changed with a policy that took effect on January 15, 2026. It moved the H200 and equivalent chips from presumption of denial to case-by-case review.
The loosening came with strings attached. Conditions included independent testing in the US and assurances that American customers would still have adequate supply.
The 3.4 million unit request figure suggests Chinese demand responded to that opening quickly. The nearly 900,000 approvals suggest regulators were far more cautious than buyers hoped.
What this means for Nvidia and the chip trade
For Nvidia shareholders, the immediate question is how much the halt changes the outlook. The China licensed business produced $2.4 billion in a single quarter.
In some prior quarters, licensed H200 sales to China were under 1% of data center revenue.
Cloud operators and data center builders in Southeast Asia and Brazil now face new licensing requirements. Any of them serving restricted entities will need to sort out compliance within a 90-day window.
For now, the gap between 3.4 million requested units and nearly 900,000 approvals tells the story. Demand is plentiful. Permission is scarce, and getting scarcer.
Disclosure: This article was edited by John Chen. For more information on how we create and review content, see our Editorial Policy.

