Nvidia’s 25% Permission Tax Is Small Until the Rules Change
Washington permits H200 shipments. Beijing restricts sales. Nvidia keeps the tariff, and Congress is pressing for wider controls.
Nvidia held U.S. licenses to ship H200 chips to China and still took a $0.4 billion H200 charge during the first half of fiscal 2027. Washington permitted shipments; Beijing restricted sales. The chips required U.S. inspection before shipment, with a 25% tariff upon importation into the United States. Nvidia says customers have absorbed none of it. You can hold a license and lose pricing power.
The Permission Tax is the cost a supplier carries when export authorization preserves a legal route to customers without preserving the economics of a sale. Nvidia’s 26 August 2026 Form 10-Q says, "We have been unable to pass along any of the tariff to our customers." That sentence matters more than a license announcement. The cost is manageable today. Governments retain the power to enlarge it. Permission changes what Nvidia is allowed to sell while leaving the company holding the expense.
An approval is only half a sale. Starting in February 2026, the U.S. government granted licenses for small H200 shipments to specific China-based customers. Nvidia then disclosed that Chinese authorities restricted those sales, leaving some licensed products unsold. One government dictates the conditions for export. The other decides whether demand becomes a transaction. Nvidia booked the cost as a charge for excess inventory and purchase obligations as H200 demand diminished. You might think a license fixes the sales funnel. Beijing still gets a veto.
The money is small by Nvidia standards, which is precisely why it is easy to dismiss. The company booked $0.4 billion, which is $400 million, for H200 excess inventory and purchase obligations during the first half ended 26 July 2026. Revenue for that half was $177,837 million, making the charge 0.225% of sales ($400 million ÷ $177,837 million). Operating income for that half was $117,270 million, making the charge 0.34% ($400 million ÷ $117,270 million). As a scenario only, a $4,500 million charge like the earlier H20 charge would be 3.84% of that half’s operating income ($4,500 million ÷ $117,270 million). The products and periods differ; the scenario sizes a possible cost shock.
The licensed lane is tiny. Nvidia’s Data Center division earned $89,023 million in the quarter ended 26 July 2026, while licensed H200 shipments represented less than 1% of that division’s revenue. In the same quarter, direct customers headquartered in China and Hong Kong generated $7,880 million of Nvidia’s $96,221 million total revenue, or 8.19% ($7,880 million ÷ $96,221 million). Those are different exposures. You are looking at the location of a customer’s headquarters in one column and a particular licensed product in another.
The geographic column invites a mistake. China and Hong Kong customer-headquarters revenue rose from $3,985 million in the quarter ended 27 July 2025 to $7,880 million in the quarter ended 26 July 2026. Yet first-half revenue on that measure fell from $13,644 million in the six months ended 27 July 2025 to $12,430 million in the six months ended 26 July 2026. Nvidia bases the geography on direct-customer headquarters, even when the final customer or shipping destination is elsewhere. If you take the latest quarter’s jump as H200 pricing power, you are confusing the address on an invoice with the profit on a chip.
The case for shrugging off H200 has numbers. Nvidia’s 26 August 2026 earnings release put next-quarter revenue in its outlook at $108.0 billion, plus or minus 2%, with gross margin at 74.0%, plus or minus 50 basis points. It added, "NVIDIA is not assuming any Data Center compute revenue from China in its outlook." Nvidia’s earnings outlook already assumes zero China Data Center compute. Why dwell on the H200 charge?
ASML offers the stronger historical objection. Its 2023 Form 20-F says, "Customers in China represented 26.3% of our 2023 total net sales." That was EUR 7,251.8 million of EUR 27,558.5 million in 2023, against EUR 2,916.0 million of EUR 21,173.4 million in 2022, or 13.8% then (EUR 2,916.0 million ÷ EUR 21,173.4 million). China sales grew despite announced controls. The analogy breaks at the product and the measurement: ASML sells lithography tools, while Nvidia’s geographic figure tracks direct-customer headquarters and its H200s face a specified tariff.
The resilience argument is right. Neither prices an H200 contract. Nvidia’s 25% levy applies to H200 imports into the United States, rather than to all its corporate revenue, and the taxable shipment value is absent from the public numbers. The aggregate tariff bill remains unquantified. The booked $400 million charge is separate, and Nvidia’s statement settles who bears the tariff now. The Permission Tax is a bargaining fact before it becomes a valuation shock. You own the consolidated income statement, but the leakage begins one customer at a time.
Nvidia booked a far larger charge under earlier restrictions. An April 2025 U.S. license requirement for H20 shipments to China produced a $4.5 billion inventory and purchase-obligation charge in the first quarter of fiscal 2026. As its 27 August 2025 Form 10-Q records, that was 10.2% of the quarter’s $44,062 million revenue ($4,500 million ÷ $44,062 million). In the next quarter, Nvidia sold approximately $650 million of H20 outside China and released $180 million in previously reserved inventory. Some value came back. Revenue and a reserve release are separate items. H20 faced a different restriction. The episode demonstrates both a sudden write-down and a possible salvage route.
Congress is turning the screw. On 12 November 2026, three House committee chairmen requested appearances from Nvidia's former chief executive, its general counsel and a Commerce under secretary, pressing for AI OVERWATCH, Chip Security, MATCH and Remote Access Security Act provisions in the year-end defense bill. Chairman John Moolenaar said, "In Congress we must act quickly to pass bipartisan export control legislation in the year-end national defense bill." Those are legislative demands, rather than enacted restrictions. The pressure now reaches beyond the H200 shipment.
Nvidia itself identified the larger perimeter in its August 2026 filing. GAIN AI would constrain the administration’s flexibility to alter export rules. The Remote Access Security Act, still under consideration, could reach cloud services supplied to companies with a China-headquartered ultimate parent. Nvidia warns the latter "could have a material impact on our business." The Permission Tax has a visible ledger entry. A rule governing access to compute can reach beyond chips physically shipped into China. Nvidia’s current outlook assumes zero China compute while lawmakers debate a broader definition of access.
I would buy Nvidia only at a valuation justified by its $108.0 billion next-quarter outlook issued on 26 August 2026, including its explicit assumption of zero China Data Center compute. The licenses deserve no additional premium. The 25% tariff remains Nvidia’s expense under the last disclosed terms. I would be wrong if Nvidia’s Q3 FY2027 release or call on 17 November 2026 showed that customers absorb the whole 25% H200 U.S.-import tariff. Until then, permission is a cost-bearing privilege. Pay for the earnings Nvidia expects from its existing business, rather than a China reopening whose costs it still carries.