Nvidia’s AI chip sales to China delayed by US security review

Nvidia’s AI chip sales to China remain delayed as the US conducts a national security review, nearly two months after Trump greenlit exports in December.

Chinese customers are hesitant to place orders until licensing conditions are clarified.

In December, Nvidia CEO Jensen Huang brokered a deal with the US government to export H200 chips to China, potentially opening a US$50 billion market.

Production has slowed as US agencies assess security risks, with some suppliers pausing component manufacturing.

The US Commerce Department completed its review, but the State Department is pushing for stricter restrictions, complicating approval.

The deal requires the US to take a 25% cut of sales revenue and mandates half of all shipments go to US customers, with third-party testing.

Concerns remain about the chips’ use in Chinese military and intelligence efforts. Chinese firms are exploring alternative sourcing amid delays.

🔗 Source: Financial Times

🧠 Food for thought

Implications, context, and why it matters.

The US approval for Nvidia’s China chip exports comes with extensive strings attached

The US shifted export reviews for chips like the H200 to China and Macau from a “presumption of denial” to “case-by-case,” which still falls short of a blanket approval 1.
Exporters must certify that sales will not divert global foundry capacity (semiconductor factory production slots) away from making similar-node or more advanced chips for US end users, and that supply remains sufficient for US customers 1.
The rule limits the aggregate “total processing performance” of advanced-node ICs (integrated circuits, or chips) shipped to China or Macau to no more than 50% of the aggregate shipped to US customers for end use in the US for the same commodities over a defined period, rather than using a simple “50% of shipments” cap 1.
Regulations also require IaaS (Infrastructure-as-a-Service) cloud providers and their end users to commit that they will not transfer model weights (the learned parameters of an AI model) trained on the AI commodities to any end user not previously disclosed on the license or without authorization from BIS (the Commerce Department’s Bureau of Industry and Security, which enforces US export controls) 1.

This chip standoff creates winners and losers far beyond Nvidia

Large Chinese tech firms may follow a dual-track strategy that uses H200s for large-scale model training while shifting domestic chips to inference (running AI models after they’ve been trained) plus smaller training jobs 2.
One analyst says higher H200 volumes could improve Nvidia’s production scale and make supply more predictable, which may open room for better pricing or availability for customers outside China 2.
Expanded H200 access might speed progress in Chinese open-source AI models, which can help enterprises experimenting with them 2.
Nvidia management argues that selling into China supports America’s ability to compete in AI and helps Nvidia stay the platform of choice for commercial business, including businesses in China 3.