Technology WWC APAC Desk

US AI Chip Curbs Target Overseas Compute—Can Nvidia and AMD Navigate the New Rules?

US AI Chip Curbs Target Overseas Compute—Can Nvidia and AMD Navigate the New Rules?

WASHINGTON — The US Commerce Department announced fresh controls on advanced computing technology on Nov. 16, 2026, targeting overseas data centres that supply substantial computing power to certain restricted customers. The US AI chip curbs also add licensing requirements for specified semiconductor shipments to five countries, raising new questions about compliance costs and access to international markets for chip suppliers.

The rules also scrutinise overseas computers serving customers remotely, beyond direct chip shipments.

US AI chip curbs: What the proposed rule actually covers

The Bureau of Industry and Security released an interim final rule titled Addressing Remote Access to Advanced Semiconductors. Its public-inspection document establishes a licensing requirement for certain advanced computing items used by overseas data centres when they provide more than 12,000,000 aggregate total processing performance (TPP) to entities headquartered in Macau or destinations in Country Group D:5, including China.

The restriction can cover directly linked computing facilities and certain indirect arrangements. The rule describes an exception for commercial recommendation systems, including functions used by online retailers and social media platforms.

A second element identifies Indonesia, Malaysia, Singapore, Thailand and Brazil for additional licensing of specified advanced semiconductor exports, re-exports and transfers. Qualifying shipments to US-headquartered recipients or their US-headquartered parent companies have an exception subject to reporting requirements.

The rule is scheduled for publication in the Federal Register on Nov. 19 and takes effect on publication. It provides a public-comment process; companies must distinguish that future effective date from the Nov. 16 announcement.

Why cloud computing complicates chip export controls

Traditional controls focus on where restricted hardware is shipped. Cloud computing creates another route: an overseas operator can retain the chips while a customer in another jurisdiction uses their computing power remotely.

The new rule addresses that difference through end-use restrictions and due diligence on customers, including ultimate-parent ownership. It does not say every cloud service is prohibited, nor does it establish wrongdoing by any named technology company.

Analysis: Data-centre operators may need more detailed customer screening and records of how computing capacity is allocated. Their obligations could depend on chip classification, the level of computing supplied and the identity of the ultimate customer. A licence requirement is not the same thing as an automatic ban.

The regional issue connects with WWC’s prior examination of Malaysia’s data-centre oversight, where equipment movement and declared end use have already raised separate regulatory questions.

What Nvidia’s own financial filing reveals

Nvidia’s quarterly filing for the period ended July 26, 2026 describes the effects of earlier restrictions. The company reported receiving US licences permitting small quantities of H200 shipments to selected China-based customers, while Chinese government restrictions limited sales.

It disclosed a US$400 million charge in the first half of fiscal 2027 related to H200 excess inventory and purchase obligations. Licensed H200 shipments accounted for less than 1% of its most recent quarter’s Data Center revenue.

Nvidia reported US$89.0 billion in second-quarter Data Center revenue. That is historical revenue, not a quantified effect of the November rule.

Analysis: A manufacturer can obtain a US export licence without completing a sale. Customer demand, destination-country restrictions and compliance costs can still affect the outcome. WWC previously explored that tension in its report on Nvidia chip-export licensing.

AMD’s filing shows a similar commercial dilemma

Advanced Micro Devices’ November quarterly filing says controls have affected its Instinct MI308 products. It received some licences to ship MI308 products to certain Chinese customers and began shipments at the end of fiscal 2025.

AMD also disclosed that US authorities granted some licences for MI325 products during the first half of 2026. Those products must undergo inspection in the United States before shipment, and the company describes a 25% tariff applying on their importation into the United States for inspection.

The company warned that future controls may restrict its ability to serve customers outside China as well. That risk statement is not evidence that the Nov. 16 rule has already generated a specific AMD write-down or cancelled order.

Analysis: Investors should separate documented historical charges, ongoing licensing conditions and possible future effects. They are three different categories of risk and should not be combined into a headline estimate of damage.

What the rule means for Southeast Asia and the Philippines

Singapore’s government stated on Nov. 16 that it was examining the measure, consulting affected companies and discussing implementation with US authorities. It also said it was exploring ways to preserve trusted access for legitimate Singapore businesses.

The Philippines is not one of the five specifically named destinations for the additional location-based licensing requirement. But overseas data centres anywhere may fall within the remote-access rule when the specified customers and computing thresholds are involved.

For Philippine operators, this creates a practical distinction between the location of a server and the ultimate user of its computing capacity. No general exemption can be inferred from the Philippines’ absence from the five-country list.

A separate example of scrutiny over cross-border technology transfers appeared in WWC’s coverage of Taiwan’s AI-server export case. That separate case does not establish violations by businesses covered by the new rule.

What happens next

The first milestone is the scheduled Nov. 19 Federal Register publication. The rule provides a 60-day comment period after publication and requires certain existing data centres that wish to continue covered remote-access services to obtain authorisation within 90 days of the effective date.

Investors should watch the final published text, agency guidance, future company filings and any verified licensing decisions. Original exchange data would also be needed to establish precise Nov. 17 stock movements and avoid attributing them entirely to one policy announcement.

The verified conclusion is that Washington is extending controls beyond the physical shipment of advanced processors. Whether the measures materially change individual chipmakers’ profits remains a question for future disclosures, not a result established by the rule itself.

FAQ

What are the new US AI chip curbs?

They add licensing requirements for specified advanced semiconductor transactions in five countries and for certain overseas computing arrangements serving restricted customers above a defined threshold.

Which countries face additional destination-based licensing?

Indonesia, Malaysia, Singapore, Thailand and Brazil are named, with a qualified exception for US-headquartered recipients subject to reporting.

Is selling advanced AI chips to China completely banned?

Not universally. Earlier rules permit some case-by-case licence reviews; the applicable product, recipient and transaction determine the legal requirements.

Does the new rule cover Philippine data centres?

Potentially. The Philippines is not on the five-country list, but the remote-access provisions may apply to facilities anywhere if the specified conditions are met.

Did the new rule cause a confirmed loss at Nvidia or AMD?

No company filing reviewed establishes a financial loss caused specifically by the Nov. 16 rule. Their disclosures describe earlier restrictions and risks.

Get our stories first on Google

More in Usa