The Commerce Department will require a license to ship advanced AI chips to Singapore, Malaysia, Indonesia, Thailand and Brazil, and to any data center that gives companies from China and other US-embargoed countries large-scale remote access to them, under an interim final rule filed for publication on Monday, November 16, 2026. The rule takes effect when it appears in the Federal Register, which is scheduled for November 19.
It arrives a day before Nvidia reports quarterly results, and Nvidia does not disclose what it ships to the five countries. Its latest quarterly report sorts revenue by where the direct customer is headquartered. Every country other than the United States, Taiwan and China together produced $1.28 billion, or 1.3% of revenue, in the quarter ended July 26. Customers based in the five named countries sit inside that figure. It is smaller than the $2.16 billion margin Nvidia placed around its own revenue outlook for the quarter it reports on Tuesday. The filing cannot show how many chips reach those countries through buyers based elsewhere.
Article Brief
Key Takeaways
4 Points24s Read
- The ruleCommerce will require a license to ship advanced AI chips to Singapore, Malaysia, Indonesia, Thailand and Brazil. It takes effect on publication, scheduled for November 19.
- Remote accessData centers that give companies from China and other embargoed countries more than 12 million aggregate TPP of computing power need a license. Existing ones have 90 days.
- Nvidia’s disclosureCustomers based outside the US, Taiwan and China produced $1.28 billion, or 1.3% of revenue, last quarter. That is less than the $2.16 billion margin on Nvidia’s outlook.
- The unknownThe filing does not show where chips sold to Taiwan-based customers, 28% of revenue, are installed. Nvidia reports on November 17.
What the rule requires
The first change is a license requirement by destination. The Bureau of Industry and Security, the Commerce unit that writes export rules, says the five countries present an unacceptable risk of diversion, and that it is acting on concerns that controlled chips “are being diverted to entities and countries of concern without authorization.” Applications will be reviewed case by case, with each country judged on its own export controls and its cooperation with the bureau. The requirement does not apply when the end user is headquartered in the United States or has a US parent, provided the exporter files reports under a new section of the regulations.
The second change follows the chips after they are installed. An exporter needs a license if it knows the chips are bound for a data center, anywhere in the world, that provides more than 12,000,000 “aggregate TPP” of computing power to or for companies headquartered in Macau or in Country Group D:5, the regulations’ list of countries under US arms embargoes, which includes China. Linked data centers count together. The rule also reaches a customer outside China that trains models for a Chinese company.
Data centers already serving such companies have 90 days from the effective date to obtain a license “or face future export restrictions.” A later change in how the chips are used requires authorization. Data center operators, foreign foundries and chip exporters must check customers for direct or indirect ownership or control from those countries. The data center control does not apply to computing power used only for commercial recommender systems, of the kind online retailers and social media companies run, as long as reports are filed. Systems that generate text, images or code do not qualify.
The 117-page document uses “Super Intelligence” in place of AI, citing an executive order. It also formally rescinds the 2025 AI Diffusion Rule, which it says was never enforced, and warns that countries that “tolerate evasion” could face import restrictions, new license requirements and a downgrade of their export-control status. The bureau expects about 100 more license applications a year. The rule does not name Nvidia or any other chipmaker.
| Provision | What the rule says |
|---|---|
| Countries needing a license | Singapore, Malaysia, Indonesia, Thailand and Brazil |
| Exempt buyers | End users headquartered in the US or with a US parent, if reports are filed |
| Remote-access threshold | More than 12,000,000 aggregate TPP for companies from Macau or Country Group D:5 |
| Existing data centers | License within 90 days of the effective date |
| Carve-out | Commercial recommender systems, with reporting |
| Review policy | Case by case |
| Effective date | On publication, scheduled for November 19, 2026 |
| Comment period | 60 days after publication |
| Earlier framework | AI Diffusion Rule formally rescinded |
How large 12 million TPP is
TPP stands for total processing performance, the bureau’s score for a chip’s computing power, and aggregate TPP is the sum across a group of chips. The rescinded framework gives a reference point. The AI Diffusion Rule let any single buyer receive up to 26,900,000 TPP a year under a low-volume exception and capped most countries at 790,000,000 TPP through 2027. The new threshold is 45% of that single-buyer allowance and 1.5% of the country cap. The two rules work differently, since one counted shipments and the other counts access, but the comparison shows the bar for “significant” has been set low.
What Nvidia’s filing shows
Nvidia’s revenue table has four lines. In the quarter ended July 26, customers headquartered in the United States accounted for $60.1 billion, Taiwan $27.0 billion, China including Hong Kong $7.9 billion and all other countries $1.28 billion.
| Customer headquarters | Quarter ended July 26, 2026 ($ millions) | Share of revenue | Year earlier ($ millions) | Change |
|---|---|---|---|---|
| United States | 60,074 | 62.4% | 32,897 | +83% |
| Taiwan | 26,985 | 28.0% | 8,902 | +203% |
| China, including Hong Kong | 7,880 | 8.2% | 3,985 | +98% |
| Other | 1,282 | 1.3% | 959 | +34% |
| Total | 96,221 | 100% | 46,743 | +106% |
The “Other” line grew 34% from a year earlier while total revenue more than doubled, so its share fell from 2.1% to 1.3%. That caps what customers based in Singapore, Malaysia, Indonesia, Thailand and Brazil could have bought directly.
The same page carries the limit of that reading. “The end customer and shipping location may be different from our direct customers’ headquarters location,” the filing says. Nvidia lists contract manufacturers, distributors and system integrators among its direct customers. Revenue from Taiwan-based customers tripled to 28% of the total, and the filing does not say where the systems those customers assemble are installed.
Two other lines matter for the rule. The 62% of revenue from US-based customers falls under the exemption when those companies place chips in the five countries, subject to the new reports. Sales of advanced chips to the China-based customers behind the 8.2% line already require licenses. The remote-access control is aimed at computing power such companies rent abroad.
The forecast: what could reach Nvidia’s outlook
Nvidia’s August outlook called for revenue of $108.0 billion, plus or minus 2%, in the quarter that ended October 25. That quarter closed before the rule was filed, so the reported numbers are not affected. The rule takes effect in the current quarter, a 14-week period ending January 31, 2027, which makes the outlook Nvidia gives on Tuesday the first place an effect could appear.
These are sensitivities on second-quarter revenue, not estimates of what the rule will cost. Nvidia has not said what share of its Taiwan-based customers’ purchases ends up in the five countries. The first figure shows that even a complete stop in sales to customers based outside its three main markets would fall inside the range Nvidia already allows for.
The strongest argument against a large effect is that the rule licenses shipments, it does not ban them. Applications are reviewed case by case, US companies building in Singapore or Malaysia are exempt, and buyers can install capacity in countries that are not on the list.
The argument on the other side is about scope and timing. The rule lets the Commerce secretary add countries, and it puts a 90-day clock on data centers that already serve Chinese companies. Chips already in those facilities were sold in earlier quarters. What is at stake for Nvidia is further orders from them. The diligence duties also fall on Nvidia’s customers and their customers, which can slow orders that would ultimately be approved.
The stock was already under pressure. Nvidia said on November 12 that Jensen Huang had resigned as chief executive and that finance chief Colette Kress would serve as interim CEO. The shares closed at $156.14 on Monday, down 4.1%, according to Nasdaq data. TECHi’s review of Nvidia’s stock holdings on Monday found a separate drag on this quarter’s reported earnings.
What to watch
Nvidia reports on Tuesday, November 17, at about 1:20 p.m. Pacific time. The points to listen for are whether it puts a number on shipments to the five countries, and whether its outlook range widens.
The rule is scheduled for publication on November 19. If that date holds, comments are due by January 18, 2027, and data centers serving Chinese companies have until February 17, 2027 to obtain a license. Because it is an interim final rule, the bureau can revise it after reading the comments. Any addition to the country list would be the clearest sign that the controls are still widening. TECHi’s Nvidia quote page tracks the share price.
