US tightens AI chip export controls to deter Chinese companies
The US Commerce Department rolled out export controls requiring global data centers above 12 million TPP to obtain licenses before serving Chinese AI firms.
By Dillip Chowdary • Nov 16, 2026 • Source: TechCrunch

At a glance
- Regulation
- Commerce Department interim export rule
- Target
- Remote compute access by D:5 countries
- Performance limit
- 12 million total processing performance
- Equivalent hardware
- Approximately 750 Nvidia H100 or H200 chips
- Compliance window
- 90 days for existing data centers
- Prior policy
- January 2025 AI Diffusion Rule rescinded
In this post
The Trump administration is introducing sweeping export controls designed to prevent overseas data centers from renting advanced artificial intelligence processing power to restricted foreign entities. Under an interim rule prepared by the Commerce Department, technology providers must secure formal export licenses before transferring high-end semiconductors, servers, or computing infrastructure that foreign adversaries could access remotely. According to TechCrunch's report, the policy directly targets firms seeking to bypass existing hardware embargoes by leasing server capacity abroad.
This analysis details the operational performance thresholds, geographic diversion targets, and compliance responsibilities established by the Commerce Department's updated export framework. The regulatory changes directly affect global data center operators, enterprise cloud infrastructure providers, semiconductor supply chain managers, and multinational technology organizations navigating international compliance to understand how remote access licensing alters global AI compute deployment.
01Global licensing for remote compute access
The Commerce Department now requires technology firms to obtain specific export, reexport, or transfer licenses before sending advanced artificial intelligence chips and servers to data centers worldwide. Regulators established this requirement to ensure foreign infrastructure providers cannot sell computing access to enterprises headquartered in China or other designated nations. Existing regulations had already barred direct chip deliveries to businesses owned by entities in these jurisdictions.
The interim rule focuses specifically on enterprises based in the American D:5 list of embargoed jurisdictions, which encompasses China, Russia, Iran, North Korea, and 18 additional countries. Regulators described the measure as a logical extension of existing controls intended to stop companies from countries of concern from utilizing American Super Intelligence semiconductors to train sophisticated models.
Federal officials stated that restricting remote access closes an enforcement gap that allowed overseas facilities to host frontier compute for restricted entities. By regulating the infrastructure pipeline internationally, the Commerce Department seeks to undermine the capacity of foreign developers to train advanced Super Intelligence systems on American technology.
02Megaspeed scrutiny and broker oversight
The regulatory update arrives shortly after Nvidia co-founder Jensen Huang resigned following a controversy involving an alleged Chinese spy associated with a customer called Megaspeed. United States authorities have considered adding Megaspeed to a special entity list over allegations that the firm acquired thousands of Nvidia graphics processors to route remote compute resources directly to Chinese organizations.
Reporting from Reuters recently revealed that Chinese artificial intelligence enterprises have accelerated efforts to lease computing capacity in foreign server facilities before tighter restrictions took effect. To prevent evasive contracting arrangements, the new rules extend accountability to transactions conducted through third-party brokers and intermediaries.
Exporting vendors and international data center hosts must now continuously monitor customer ownership structures and corporate actions to verify that no entities linked to the D:5 embargoed list operate the leased infrastructure. This compliance burden forces hosting providers to audit end-user identities rather than relying on unverified third-party declarations.
03Twelve million total processing performance threshold
Advertisement
Tech Pulse Daily
Get tomorrow's pulse first
Join engineers who read Tech Pulse before stand-up. Free, weekday mornings.
The federal licensing requirement applies to any facility or interconnected network capable of providing more than 12 million total processing performance, a regulatory computing benchmark known as TPP. In practice, that performance ceiling equates to the collective maximum compute delivered by roughly 750 Nvidia H100 or H200 accelerators operating together.
Regulators emphasized that clustering compute beyond 12 million TPP enables foreign organizations to train and operate advanced Super Intelligence models capable of aiding military and intelligence programs. By pegging restrictions to raw performance density, the administration prevents foreign engineers from aggregating smaller server slices into unified clusters that support frontier model development.
Data centers maintaining high-density clusters must now determine whether their aggregated infrastructure exceeds the 12 million TPP ceiling before executing client agreements. Because the metric applies to any network that can supply this capacity, distributed server configurations that pool compute across facilities also fall under federal oversight.
04Diversion risk territories and commercial exemptions
The Commerce Department also imposed expanded licensing requirements on semiconductor exports, reexports, and transfers to Malaysia, Singapore, Indonesia, Thailand, and Brazil. Officials cited an elevated risk of illicit diversion through these hubs, though the licensing obligations will not apply when exported hardware is destined for American-headquartered businesses operating locally.
Note: Existing data center operators providing services that fall under this new rule are required to get licenses within 90 days of the rule coming into effect. This transition timeline gives international facilities three months to file regulatory documentation with the Commerce Department before mandatory compliance takes full effect.
The interim framework explicitly narrows its scope to high-end machine learning workloads, leaving standard consumer applications untouched. Commercial recommendation systems, such as algorithms deployed across e-commerce storefronts and social media networks, are exempted from the export licensing restrictions. This carveout ensures daily retail and content personalization feeds continue operating without federal intervention.
05Rescission of the previous diffusion framework
Alongside the remote compute mandates, the interim final rule formally rescinds the Biden administration's AI Diffusion Rule introduced in January 2025. That prior framework had established country-level quotas on advanced chip exports and imposed statutory restrictions on the transfer of powerful artificial intelligence models across borders.
The Trump administration announced plans to discard that framework in May, arguing that its broad limits placed unnecessary burdens on American businesses and international allies. By replacing the diffusion model with targeted compute thresholds, the Commerce Department focuses enforcement directly on data centers capable of supporting military and intelligence activities that threaten national security.
Officials noted that the updated interim final rule establishes a focused barrier against unauthorized military model development without maintaining sweeping geographic restrictions on friendly nations. The policy shifts regulatory scrutiny directly to the physical computing density and ownership networks powering frontier artificial intelligence infrastructure.
Frequently asked questions
Which countries face remote compute restrictions under the new interim rule?
The restrictions target companies based in the United States D:5 group of embargoed nations, which includes China, Russia, Iran, North Korea, and 18 other jurisdictions.
What compute performance threshold triggers the export license mandate?
Any data center or network capable of delivering over 12 million total processing performance, equivalent to roughly 750 Nvidia H100 or H200 chips, requires an export license.
Are there specific international diversion risk countries identified by the Commerce Department?
Yes, advanced semiconductor transfers to Malaysia, Singapore, Indonesia, Thailand, and Brazil face licensing rules unless the hardware is destined for American companies.
Which workloads are exempted from the remote access export licensing rule?
Commercial recommendation systems used across e-commerce and social media platforms are explicitly exempted from the licensing mandate.
Sources
Author
Dillip Chowdary
Writes Tech Bytes coverage of AI, engineering, and the tools that actually ship. Editor of Tech Pulse Daily.
Related on Tech Bytes
Advertisement