US AI Chip Rule for Overseas Data Centres: What It Means for India

The United States has issued a new rule that makes it much harder for Chinese companies to use American AI chips by renting them in data centres outside China. The Commerce Department’s Bureau of Industry and Security (BIS) announced the interim final rule on Monday, November 16, and it takes effect once it appears in the US Federal Register, which law firm Nixon Peabody expects on November 19. The move comes days after Nvidia’s leadership upheaval and a criminal probe into a Singapore data centre firm, and it changes the rules for chip buyers across Asia.
What the New US Rule Actually Does
Since October 2022, US companies have generally needed a licence to sell advanced AI chips to China, Hong Kong and Macau. But Chinese developers could still log in remotely to powerful servers in third countries and use the same chips. The new rule targets that gap.
According to the BIS announcement and a legal summary by Nixon Peabody, it works in two ways:
- Any data centre, anywhere: A licence is needed to export, re-export or transfer advanced AI chips and servers to a data centre that gives large amounts of computing power to companies headquartered in China, Macau or countries under a US arms embargo. The trigger is more than 12 million “aggregate TPP” (total processing performance), counted across one site or a group of linked sites.
- Five named countries: Shipping advanced chips to data centres in Indonesia, Malaysia, Singapore, Thailand and Brazil now needs a licence, because Washington says they carry an “unacceptable risk of diversion.” Commerce can add more countries later.
There is a carve-out for buyers that are, or are owned by, US-headquartered companies, provided the exporter files monthly reports. Recommendation engines used only to rank products or content are also excluded. The rule formally scraps the Biden-era AI Diffusion framework, including its country tiers.
Key Dates for Companies and Data Centres
The deadlines below are counted from the expected November 19 publication. If the Federal Register date shifts, each one moves with it.
| Step | Time allowed | Likely date |
|---|---|---|
| Rule takes effect | On publication | November 19, 2026 |
| Shipments already in transit must be completed | 30 days | Around December 19, 2026 |
| Public comments close | 60 days | Around January 18, 2027 |
| Existing data centres serving covered firms must get a licence | 90 days | Around February 17, 2027 |
Why Southeast Asia Is in the Spotlight
The rule lands in the middle of the storm around Nvidia. Former chief executive Jensen Huang stepped down last week during a board investigation, as we reported in our story on Huang’s resignation and the interim CEO. On Monday the US Commerce and Justice Departments confirmed a criminal investigation into Megaspeed, a Singapore-based data centre company with facilities in Malaysia. Megaspeed has said it “conducts legitimate and lawful business” and has hired outside counsel.
Singapore said on Tuesday it was studying the measure, contacting affected companies and exploring “trusted access” for its firms, Al Jazeera reported. China has rejected the espionage claims linked to the scandal as groundless.
Markets took the news badly. On Monday the Nasdaq fell 2.5% and Nvidia dropped 4.1%, extending a fall of more than 30% in a month. On Tuesday, Japan’s Nikkei slipped 1.3% and South Korea’s Kospi lost 2.6%.
What the Rule Means for India
India is not one of the five countries on the new licence list, and the sources we reviewed do not mention India at all. Here is the context Indian readers need:
- India’s export control status: In 2018 the US moved India into Country Group A:5, the group of closest partners that also includes Japan, Australia and NATO members. The rule’s separate “regional stability” licence requirement excludes A:5 destinations, according to Nixon Peabody.
- End of the old GPU caps: Under the January 2025 AI Diffusion framework, Indian buyers faced limits such as 50,000 H100-equivalent chips through 2027 under country-specific licences, per an analysis by the Takshashila Institution. The Trump administration said in May 2025 it would not enforce that framework, and this rule now formally rescinds it.
- The China test still applies: Location is not the only trigger. An Indian data centre that sells over 12 million TPP of computing to a China-headquartered company would need a US licence like any other site in the world. Exporters must also collect “know your customer” details on who owns and controls foreign data centres.
- Supply chain partner: India joined the US-led Pax Silica coalition on chips, minerals and AI infrastructure on February 20, 2026, according to the Press Information Bureau.
In simple terms, Indian companies building AI capacity do not face a new country-level barrier from this rule, but any business serving Chinese clients with large amounts of compute will need to check its exposure.
What Happens Next in Washington
The rule may not be the last step. Fortune reported that several bills are pending in Congress, including the Remote Access Security Act, which the House passed in January and which would give Commerce power to licence and penalise foreign firms using US chips abroad. Commerce Secretary Howard Lutnick is expected to testify before Congress, according to NPR. Georgetown researcher Hanna Dohmen told Fortune that rules alone will not be enough without enforcement.
The timing also matters because US President Donald Trump is due to meet Chinese President Xi Jinping and Russian President Vladimir Putin on November 19, as covered in our report on the planned three-way meeting. The rule was released just days before that meeting, and how Singapore, Malaysia and other hubs respond in the 90-day window will show how far the rule really reaches.



