AI chip export rules used to stop at the shipping dock. As of November 16, 2026, they follow the login too. The US Commerce Department’s Bureau of Industry and Security (BIS) issued an interim final rule that treats remote access to advanced AI chips in overseas data centers as something that can need an export license. The goal: stop Chinese companies from renting top-end hardware they can’t legally buy.
This piece covers what the rule says, the numbers that matter, who gets hit first, and where enforcement looks weakest.
Table of Contents
ToggleWhat do the new AI chip export rules change?
The new AI chip export rules require a US license to ship advanced chips to any data center worldwide that gives companies headquartered in China, Macau or other D:5 countries more than 12 million aggregate TPP of computing power. The rule also adds license requirements for chip exports to five countries: Indonesia, Malaysia, Singapore, Thailand and Brazil.
Key facts at a glance
- Issued: November 16, 2026, as a BIS interim final rule that the agency says was written by and at the direction of the Treasury Department and the National Security Council.
- Core trigger: more than 12 million aggregate TPP supplied to companies headquartered in China, Macau or other D:5 countries.
- Grace period: data centers already serving such customers get 90 days from the effective date to obtain a license.
- Also rescinded: the Biden-era AI Diffusion Rule, which BIS says was never enforced.
Why remote access became the loophole
Since October 2022, US companies have generally needed a BIS license to sell advanced computing chips to China, including Hong Kong and Macau, according to NPR’s reporting. That covered hardware crossing a border. It didn’t cover a Chinese lab logging into someone else’s servers.
Picture a data center in Southeast Asia that legally buys a few thousand top-tier GPUs. The chips never reach China. But a Chinese AI firm rents the cluster for a training run, pays by the hour, and gets compute it couldn’t have imported. On paper, nothing was exported. In practice, the capability crossed the border anyway. Cassia King of the Institute for AI Policy and Strategy made the same point last week: export controls covered the physical transfer of chips but not remote access, as Quartz reported.
BIS now treats that gap as something it can regulate. Its press release says the rule exists so existing restrictions can’t be sidestepped through remote computing arrangements involving data centers in third countries.
What the new rule requires
| Requirement | What it means |
| License for covered data centers | Exporting, reexporting or transferring advanced chips to a data center needs a license when it will supply more than 12 million aggregate TPP to firms headquartered in China, Macau or other D:5 countries, including firms whose ultimate parent is based there |
| Existing providers | 90 days from the effective date to get a license |
| Customer due diligence | Data centers, foreign foundries and chip exporters must check who directly or indirectly owns or controls a customer |
| Five-country license | Advanced chip exports to Indonesia, Malaysia, Singapore, Thailand and Brazil now need a license; US-headquartered companies are exempt if they file monthly reports |
| Exemption | Commercial recommender systems, such as those used by online retailers and social media companies |
| AI Diffusion Rule | Formally rescinded |
The ownership checks matter more than they look. They’re aimed at the oldest trick in export control: restructure the company or spin off a subsidiary so the real customer no longer looks Chinese.
The 12 million TPP threshold in plain numbers
TPP stands for Total Processing Performance, BIS’s yardstick for chip computing power. It’s a chip’s peak dense throughput multiplied by the bit length of the operation. An H100 rates about 15,800, according to TechPowerUp. Twelve million works out to roughly 750 H100s or H200s, or around 330 B200s.
The count is aggregate. It adds up every accelerator in a facility and across linked data centers, and it reaches indirect access too, like a cloud reseller training models for a Chinese customer.
I think 750 GPUs is a low bar. Large training clusters are commonly counted in the tens of thousands of GPUs, so a serious training run blows past the line quickly. What the threshold really leaves alone is small inference and experimentation work.
Which countries and companies are affected
The rule hits two groups.
Customers. Companies headquartered in China, Macau or other Country Group D:5 destinations, including firms whose ultimate parent sits there. The D:5 group covers China, Russia, Iran, North Korea and 18 other countries, according to TechCrunch’s reporting as summarized by The Vanguard.
Transit hubs. Indonesia, Malaysia, Singapore, Thailand and Brazil now need licenses for advanced chip exports because the rule describes them as carrying a higher-than-normal risk of diversion, per The Next Web.
One naming quirk is worth knowing if you’re searching the rule text. The government now says super intelligence instead of artificial intelligence, following a September 2026 executive order. BIS calls the hardware SI chips and servers.
Why now: the Nvidia and Megaspeed backdrop
The timing is hard to miss. The rule arrived days after Jensen Huang stepped down as Nvidia’s CEO following reports about a relationship with Alice Huang (no relation), a former executive at Megaspeed International. Megaspeed is a Singapore data center company with facilities in Malaysia, and US officials suspect it of selling remote access to US AI chips to Chinese companies. NPR says it hasn’t independently verified the claims about the relationship. Megaspeed says its business is lawful and that outside counsel is running an internal investigation.
Before the scandal, NPR notes, the Trump administration had resisted adding rules to limit foreign access to American AI technology. In May 2025 it said it would not enforce the Biden-era AI Diffusion Rule and planned to rescind it. This rule finishes that job on paper.
Who feels it first
Data center operators in Southeast Asia: Analysts told Fortune the rule could hit countries that want to host data centers for Chinese firms. Joshua Kurlantzick of the Council on Foreign Relations warned the controls could be especially damaging for Malaysia.
Nvidia and chip stocks: On Monday, Nvidia shares fell as much as 2.9% and the Philadelphia Semiconductor Index dropped as much as 5.5%, according to Bloomberg as relayed by The Next Web.
Cloud and chip suppliers: Some foreign data centers may now think twice about hosting American AI hardware at all, which could push demand toward non-American providers over time. Demand for AI hardware is exactly why this matters: AMD’s data center revenue jumped 97% to $8.5 billion in Q3 2026, as we covered in AMD Q3 2026 earnings. The build-out is also being funded with debt, which we looked at in Oracle’s $20 billion AI raise.
Can the US actually enforce remote access controls?
My read: this is the weak point. Cloud traffic is global, encrypted and often routed through several jurisdictions. A remote login doesn’t look like a shipping container, and it’s still unclear how Commerce will verify who is using what.
Chris McGuire of the Council on Foreign Relations called the rule a major step forward but said loopholes remain, including some covering activity inside the US. He also pointed out that he had never seen a BIS rule written by someone other than BIS.
So the rule leans on pressure instead of inspection: license conditions, customer checks, and a warning that Commerce will watch countries that tolerate evasion. Possible responses include import restrictions, new license rules and Entity List designations. For a data center, losing access to future chip shipments is a bigger risk than losing any single customer. That’s the real enforcement tool.
What to watch next
- Federal Register publication: The rule takes effect once it’s published, and the public then has 60 days to comment.
- The 90-day clock: Data centers already serving covered customers need licenses within 90 days of the effective date.
- Trump and Xi: The two leaders are due to meet Thursday in Shenzhen, per The Next Web.
- Congress: NPR reports Huang and Commerce Secretary Howard Lutnick are expected to testify this month and next, with more investigations likely once Democrats take control of Congress in January 2027.
Details can shift once the final text appears in the Federal Register, so check the rule text from BIS before acting on anything here. For more on the chip and cloud race, browse our Artificial Intelligence coverage.
FAQ
What is the remote access rule for AI chips?
It’s a BIS interim final rule, issued November 16, 2026, that treats remote access to advanced AI chips in third-country data centers as something that can require a US export license. It targets arrangements where companies in China and other restricted countries rent computing power instead of buying chips.
Does the rule ban Chinese companies from using cloud computing?
Not outright. A license is required for chip exports to data centers that will supply more than 12 million aggregate TPP to firms headquartered in China, Macau or other D:5 countries. Smaller arrangements fall below that trigger, but the rule also requires ownership checks on customers.
Which countries now need licenses for advanced AI chips?
Indonesia, Malaysia, Singapore, Thailand and Brazil. The rule says they carry a higher-than-normal risk of diversion. US-headquartered companies are exempt as long as they file monthly reports.
When does the rule take effect?
It takes effect when it’s published in the Federal Register, and comments are due 60 days later. Data centers already providing covered computing capacity then have 90 days from the effective date to get a license.
Is the Biden AI Diffusion Rule still in force?
No. The new rule formally rescinds it. BIS says the earlier rule was never enforced, and its country tiers, Data Center VEU program and model-weight controls go away with it.








