Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseInsight · On RadarTechCrunch→ original

US Introduces Licensing for AI Chips in Data Centers Above 750 H100s

Chinese companies got around the ban by simply renting AI compute in data centers in other countries. Now the US Commerce Department requires a license for chips and servers in any data center that can serve customers from China and other D:5 countries (the embargo list). The threshold is 12 million TPP (a measure of computing power), or roughly 750 Nvidia H100 chips. Providers will have to verify who actually stands behind the tenant.

AI-processed from TechCrunch; edited by Hamidun News
US Introduces Licensing for AI Chips in Data Centers Above 750 H100s
Source: TechCrunch. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTechCrunch ↗
Editor-in-Chief
Key metric #1New chip licenses for Malaysia, Singapore, Indonesia, Thailand, and Brazil
Key metric #2Existing data centers must obtain a license within 90 days
Key metric #3The Biden-era rule on country-level limits from January 2025 has been rescinded

Imagine a shop that is not allowed to sell you a product, but its neighbor at the market sells the same product to your acquaintance, who then happily lets you use it. That is roughly how the loophole in US restrictions on AI chips worked. Selling top Nvidia accelerators to Chinese companies was prohibited, but nothing stopped them from renting compute in a data center in Malaysia or another country. According to TechCrunch, the Trump administration is closing this loophole with a temporary rule from the Commerce Department.

What Changes: Control Shifts From Chips to Who Receives the Compute

Previously, a license was required to supply AI chips to companies in D:5 or to businesses owned by entities from those countries. D:5 is the list of embargoed countries: China, Russia, Iran, North Korea, and 18 other states. Now the controls go further. A license must be obtained for exporting, re-exporting, or transferring chips and servers to data centers anywhere in the world if those centers can provide compute to companies from China or other restricted countries.

The rule applies to data centers and networks capable of delivering more than 12 million TPP. TPP, or total processing performance, is a metric the US uses to measure computing power in export controls. According to TechCrunch's estimate, 12 million TPP is roughly equal to the maximum capacity of 750 Nvidia H100 or H200 chips. In other words, this is not about a couple of server racks but about substantial clusters.

The rule also covers indirect access through intermediaries and brokers. Exporters and data center operators must monitor the ownership structure of their customers and their activities to ensure that no one from D:5 stands behind a tenant. Existing operators whose services fall under the rule must obtain licenses within 90 days after it takes effect.

New Countries at Risk and Background

Separately, licenses are being introduced for shipments of advanced semiconductors to Malaysia, Singapore, Indonesia, Thailand, and Brazil. The Commerce Department explained the choice by saying these countries "present an elevated risk of diversion" of chips. There is an exception: no license is needed if the chips are destined for companies based in the United States. In addition, the rule does not cover commercial recommendation systems, for example in e-commerce or social networks.

TechCrunch links the move to the story around Megaspeed, an Nvidia customer. US authorities reportedly want to add the company to a special restrictions list, believing it bought thousands of Nvidia chips in order to sell compute to Chinese businesses. A few days earlier, Nvidia co-founder Jensen Huang stepped down amid a scandal over an alleged Chinese spy linked to one of the customers. Reuters reported last week that Chinese AI companies were hurriedly reserving capacity abroad, fearing tighter rules.

At the same time, the Biden administration's "AI diffusion rule" from January 2025 is being repealed. It introduced country-by-country limits on chip exports and restricted the transfer of powerful models. The Trump administration had said back in May that it would drop this framework as an unnecessary burden on American companies and allies. Instead of country-level limits, the emphasis is now on verifying who exactly the data center sells compute to.

What This Means in Practice

For data center operators and cloud providers outside the US, compliance becomes part of the product. They will need to check not only the direct customer but also those who own it or resell its capacity. Startups renting GPUs from regional providers in Southeast Asia or Brazil may face delays: suppliers must obtain licenses, and some capacity may become more expensive or unavailable.

For Chinese companies, the path to training frontier models becomes harder: renting abroad is no longer a simple workaround. The rule's authors themselves write that the goal is to undermine the ability of companies from "countries of concern" to train advanced models on American chips.

Uncertainty remains. The rule is temporary, and the speed of license issuance, the details of implementation, and the industry's reaction are still unknown. But if your product depends on rented GPUs outside the US, it is worth rereading your contracts now, checking the ownership chain of your customers, and building in a buffer for timelines.

Story

Key Questions Answered

What Changes: Control Shifts From Chips to Who Receives the Compute?

Previously, a license was required to supply AI chips to companies in D:5 or to businesses owned by entities from those countries. D:5 is the list of embargoed countries: China, Russia, Iran, North Korea, and 18 other states. Now the controls go further. A license must be obtained for exporting, re-exporting, or transferring chips and servers to data centers anywhere in the world if those centers can provide compute to companies from China or other restricted countries.

What are the key details on New Countries at Risk and Background?

Separately, licenses are being introduced for shipments of advanced semiconductors to Malaysia, Singapore, Indonesia, Thailand, and Brazil. The Commerce Department explained the choice by saying these countries "present an elevated risk of diversion" of chips. There is an exception: no license is needed if the chips are destined for companies based in the United States. In addition, the rule does not cover commercial recommendation systems, for example in e-commerce or social networks.

What This Means in Practice?

For data center operators and cloud providers outside the US, compliance becomes part of the product. They will need to check not only the direct customer but also those who own it or resell its capacity. Startups renting GPUs from regional providers in Southeast Asia or Brazil may face delays: suppliers must obtain licenses, and some capacity may become more expensive or unavailable. For Chinese companies, the path to training frontier models becomes harder: renting abroad is no longer a simple workaround.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…