Chinese AI companies cannot buy Nvidia’s best chips. So for years they did the next best thing: they rented them, logging in to data centers in Singapore, Malaysia and Thailand. That back door is now closing.
The US Commerce Department has moved to treat remote access to advanced AI chips the same way it treats a physical shipment. If you build, buy or rent AI compute anywhere in the world, this story affects you.
Update, 16 November 2026: When this story first broke, Commerce was reportedly finalizing the rule. It is now official. On 16 November 2026 the Bureau of Industry and Security (BIS) released an interim final rule, and this article has been updated with the confirmed details below.
What Exactly Happened?
Here is the timeline, using official statements and reporting from established outlets.
- August 2026: CNBC reported that ByteDance, Alibaba and Tencent had used Nvidia chips through data centers in Thailand, Malaysia and Japan.
- 11 November 2026: The Information reportedly said Commerce was finalizing a rule that would cover seven countries: Singapore, Malaysia, Thailand, Indonesia, Brazil, Norway and South Africa.
- 12 November 2026: Reuters reported, citing three people, that Chinese AI firms were rushing to rent overseas capacity before the rule landed. The same day, Jensen Huang resigned as Nvidia CEO. We covered that in our Jensen Huang resignation explainer.
- 13 November 2026: Treasury Secretary Scott Bessent reportedly criticized Commerce Secretary Howard Lutnick for not doing enough to restrict chip purchases by Megaspeed, a Singapore cloud company at the center of the Nvidia scandal.
- 16 November 2026: BIS released the interim final rule (RIN 0694-AK50). Commerce said it was written by and at the direction of the Treasury Department and the National Security Council.
The official BIS press release says the goal is to make sure existing restrictions cannot be bypassed through remote computing arrangements in third countries. The rule text is public.
What the rule does
- A US export license is required to send advanced chips and servers to data centers anywhere in the world that give significant computing power to companies headquartered in China or other restricted destinations.
- Reporting on the rule text says the threshold is more than 12 million aggregate TPP (total processing performance).
- License requirements also apply to advanced chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil. The earlier draft reportedly listed seven countries, so the final list is shorter.
- Data centers that already serve covered companies get 90 days from the effective date to apply for a license, or face export restrictions.
- Public comments are reportedly due within 60 days.
- The rule also rescinds the Biden-era AI Diffusion Rule regulations.
One caution: when BIS released the rule, it had not yet been published in the Federal Register, so the final effective date was not confirmed. Check the official text before you act on any deadline.
Why Chinese firms were renting chips abroad
US export controls stop Nvidia from shipping its top chips to China. But renting compute in another country was a legal gray area. A Chinese company did not import a single chip. It just paid for time on hardware sitting in a Johor or Singapore data center.
Megaspeed shows how big the problem became. The Singapore firm reportedly bought more than 4 billion dollars of Nvidia chips and equipment, and US officials suspect some of it reached Chinese buyers. Megaspeed has not been proven guilty of anything in the sources we reviewed, so treat that as an allegation.
Why It Matters
For Users
If you use AI apps built by Chinese companies, expect slower model upgrades and tighter usage limits over time. Less compute means fewer experiments and fewer free tiers. If you use US tools like ChatGPT or Claude, you are unlikely to notice any direct change.
For Businesses
Companies that rent GPU capacity in Southeast Asia now have homework. Data center operators must check who ultimately owns and controls each customer, who gets remote access, and whether separate facilities share compute with restricted firms. If your cloud provider cannot answer those questions, your supply could be at risk.
Nvidia has also reportedly told customers it will not sell chips to operators that rent more than half of their GPUs to Chinese companies. That is a private rule on top of the government rule.
For the AI Industry
This is the first big policy move after the Huang scandal, and it shows export policy is now being written with speed and politics in mind. China-based customers made up about 7.9 billion dollars, roughly 8 percent, of Nvidia’s 96.2 billion dollar July-quarter revenue, according to Forbes. Nvidia’s current-quarter forecast reportedly assumed no China data-center computing revenue, so the direct hit may be limited.
The bigger effect is on neoclouds and Southeast Asian data center builders, who have been betting on huge demand. For the wider politics of AI infrastructure, see our post on the data center moratorium debate.
Social Media Reaction
We could not verify individual posts well enough to quote them, so here is what is documented and what the public argument looks like.
Supporters say the rule closes an obvious loophole. Senator Maria Cantwell wrote to Commerce Secretary Lutnick the day before the rule came out, saying companies with China ties were exploiting a loophole in US export controls. Hawks on chip policy argue that renting a chip is no different from owning it.
Critics raise three points. First, some trade lawyers reportedly doubt Commerce can enforce remote-access limits without new legislation. The House passed the Remote Access Security Act 369 to 22 in January, but the Senate has not acted. Second, interim final rules skip much of the normal rulemaking process, which worries industry groups. Third, honest customers in Southeast Asia may get caught in the paperwork.
Many tech watchers on X and Reddit also point out the timing, since the rule came days after Huang resigned. Treat that as commentary, not proof of motive.
What Should You Do Next?
- Audit your compute supply. If you rent GPUs abroad, ask your provider in writing whether it serves China-headquartered customers and how it screens them.
- Read the primary sources. Start with the BIS press release and rule text, and watch for Federal Register publication, which sets the real deadlines.
- Plan for a backup. Do not depend on a single overseas GPU vendor. Line up a second provider and keep your workloads portable.
FAQ
Can Chinese companies still rent Nvidia GPUs?
Not in large volumes through data centers covered by the rule, unless the operator gets a US license. Existing operators have 90 days from the effective date to apply.
Which countries does the rule cover?
It requires licenses for advanced chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil, and it covers data centers worldwide that serve China-headquartered companies.
Is the rule final?
It is an interim final rule, so it takes effect without the usual wait, but BIS is expected to take public comments. Details can change.
Will this affect Nvidia stock and AI prices?
Nvidia’s forecast reportedly assumed no China data-center revenue, so the direct impact looks limited. Effects on GPU rental prices in Southeast Asia are harder to predict.
Conclusion
The US has stopped treating cloud access as a loophole. Chip policy now covers who can log in, not just who can buy. For builders, the lesson is simple: know where your compute comes from.
Want daily, plain-English AI news like this? Follow AI Trends Setter and check back every day for the next story.
