Big Tech

U.S. AI Chip Rules Tighten Overseas Access to Nvidia Computing

U.S. AI Chip Rules Tighten Overseas Access to Nvidia Computing
U.S. AI Chip Rules Tighten Overseas Access to Nvidia Computing

New licensing requirements follow the users of overseas computing power. They add another dependency to national AI infrastructure plans.

[For more news, click here]

National AI projects seek greater control over computing capacity, data and locally developed models. Those ambitions still depend on the conditions attached to imported technology. Nvidia itself describes national AI capabilities as a combination of domestic infrastructure, local datasets and homegrown expertise in its AI strategy guidance.

That relationship is central to new controls announced by the U.S. Bureau of Industry and Security on November 16. They target remote access to advanced American chips and servers through data centers in other countries. The aim is to prevent restricted users from obtaining computing capabilities through overseas facilities.

For Chinese AI companies, renting capacity abroad can provide processing power without importing the chips. For countries hosting that capacity, the change connects local decisions more closely to the identity and ownership of their customers.

Licensing follows the user

The interim final rule requires licenses for chip shipments and transfers intended for certain overseas computing arrangements. Its threshold is access to more than 12 million aggregate TPP, a regulatory measure of computing capability. It covers companies headquartered in specified restricted destinations, including China, or with an ultimate parent there. Linked data centers also count.

Customer checks must examine ownership and control. Applications are reviewed case by case. Existing covered services must obtain a license within 90 days of effectiveness to continue.

Additional destination licensing covers Indonesia, Malaysia, Singapore, Thailand and Brazil. A reporting-based exclusion applies to qualifying recipients or end users headquartered in the United States, or with a U.S. ultimate parent. A separate exception covers exclusively commercial recommendation systems, with reporting requirements. Generative systems fall outside that exception.

The changes sit alongside existing product-specific controls. In January, BIS introduced case-by-case review for Nvidia H200, AMD MI325X and similar chip exports to China. Approval depends on security safeguards, supply conditions and independent testing in the United States. Access remains conditional rather than uniformly prohibited.

Local infrastructure, overlapping rules

The distinction is already visible in Southeast Asia. Malaysia's Ministry of Investment, Trade and Industry said on November 13 that it was checking possible chip reexports involving Megaspeed affiliate Speedmatrix. Routine facility visits, it said, did not amount to clearing either company.

Malaysia permits AI chip imports without a specific chip license, according to the statement. However, exports, reexports, transshipment and transit of high-performance U.S. AI chips require a permit under Malaysian law.

“Malaysia welcomes investment in data centres and technology, but will not allow itself to be used to get around export controls.”

Megaspeed says it has cooperated with BIS inspections and retained a U.S. law firm to inventory its GPUs and review compliance.

For data-center operators, these checks reach beyond where servers are installed. A facility can operate under local law while the use of its imported chips remains subject to another country's licensing conditions. That leaves the host country with substantial control over infrastructure, but less freedom to decide which customers it can serve.

This is where the question of AI sovereignty becomes more practical. Keeping data within national borders addresses one part of it. Access to computing hardware, and the authority to decide how it is used, addresses another.

The same dependence is visible in Saudi Arabia. HUMAIN’s partnership with Nvidia links their national AI ambitions to foreign computing hardware. They illustrate how those dependencies coexist. Its partnership with Nvidia, announced in May 2025, envisaged Saudi AI factories alongside local skills development. The company's infrastructure expansion builds domestic capability using foreign-designed computing technology. National ownership and external hardware dependence therefore sit within the same sovereignty strategy.

For HUMAIN, the new rules could shape which international workloads its facilities can accept. Saudi Arabia is outside the five newly named destinations. However, the worldwide end-use controls can still reach Saudi facilities serving restricted customers when the rule's conditions are met. Those arrangements would require authorization.

The distinction also applies to enterprise deployments. DataRobot and Dell Technologies say their Nvidia-powered architecture can keep data and model activity within a customer's infrastructure. That offers operational control, while leaving the conditions attached to imported hardware as a separate dependency.

The new controls reinforce why that dependency is a critical component of sovereignty. As access rules divide technology markets, the question extends to the reliability of the infrastructure supporting public services and business operations. Local investment, skills and data governance remain essential. Their value also depends on dependable access to the computing systems behind them.

Related Articles

Meta’s €805 Million Fine Puts Facebook Photo Access Under Scrutiny
Big Tech

Meta’s €805 Million Fine Puts Facebook Photo Access Under Scrutiny

AI generated image
Big Tech

SpaceX Plans $74 Billion AI Data Center Push as Compute Moves Closer to Power

Nvidia
Big Tech

NVIDIA Faces Leadership Change as Jensen Huang Steps Down After 33 Years

Google warns Australian online safety bill could block under-18s from Gmail
Big Tech

Google warns Australian online safety bill could block under-18s from Gmail

Flags framing a skyline
Big Tech

America Is Leading the Middle East's AI Race, but China Is Not Giving Up

Protestors Oppose Data Center Construction
Big Tech

US Midterm Elections Put Data Center Expansion Under Growing Political Pressure