The PLA Seeks High-End AI Chips from Nvidia: Nearly 900,000 Approval Records Expose Shortcomings in Intelligent Warfare

Submitted by: Li MinghuiLi Minghui
Loading...Strong SignalCredibility: 25%
Report Content

The PLA Seeks High-End AI Chips from Nvidia; Nearly 900,000 Approval Records Expose Shortcomings in Intelligent Warfare

From November 18 to 20, U.S. media continuously disclosed a verifiable piece of news: the U.S.-China Economic and Security Review Commission, in its 2026 annual report, stated that China is attempting to acquire high-end artificial intelligence chips from Nvidia for military AI and the so-called "intelligent warfare" capability development. Reuters cited the report, stating that the U.S. government has approved nearly 900,000 licenses for AI chips intended for China in the past six months. This number does not mean that all these chips have entered the PLA, but it indicates that the boundaries of the dual-use supply chain are becoming very blurred.

What Does the Report Actually Say?

Public reports did not provide evidence that the PLA has already equipped a specific model of Nvidia chips in bulk, nor did they specify the end users of the chips. The core judgment of the report is that the Chinese military and relevant research institutions are still actively seeking Nvidia's advanced computing products to train large models, process battlefield data, and support autonomous systems, intelligence analysis, and command decision-making. Media outlets like CNBC emphasized the terms "seeking" and "acquisition channels," rather than confirming that a specific weapon has been upgraded.

This distinction is important. Writing "application, procurement, or transfer" directly as "the PLA is already using" amounts to treating supply chain clues as facts of equipment. Li Minghui believes that the more prudent conclusion at this stage is that the Chinese military-industrial system is striving for computing resources, rather than having already secured a stable supply of high-end AI chips.

Why Chips Have Become a Weakness in Military Industry

  • Training large models requires high-end GPU clusters, advanced interconnects, and long-term stable power and cooling conditions;
  • Battlefield applications also require compressing models to edge devices and addressing latency, anti-jamming, and data security issues;
  • The real difficulty lies in continuously obtaining chips, software ecosystems, and maintenance parts, rather than just buying a batch of samples.

From an engineering perspective, the value of Nvidia chips is not just in computing power. The CUDA software ecosystem, mature cluster scheduling tools, and a large pool of algorithm talent constitute a complete set of "computing infrastructure." Domestic alternatives may play a role in some inference tasks, but to simultaneously handle large model training, satellite image processing, and drone swarm coordination, system-level gaps still exist. In simple terms, missing a few chips may not halt weapon systems, but lacking a sustainable and scalable computing platform will slow down the entire combat system.

Why Export Controls Did Not Immediately Close Off Channels

The approval figure of nearly 900,000 also reveals another reality: export controls target products and licenses but find it difficult to identify end uses separately. Distributors, cloud service providers, research institutions, and private enterprises may all be part of the transaction chain. As long as chips enter the market under the guise of civilian computing, autonomous driving, or data center demands, subsequent flows require stricter tracking.

However, this does not mean that the PLA can easily obtain advanced computing power on the same scale. The packaging of high-end chips, server motherboards, network interconnects, and data center operations are also restricted; even obtaining a small number of chips through third parties makes it difficult to replicate the complete development environment available in the open market. The U.S. restrictions seem more like raising the time costs for China's military industry rather than causing all AI research to come to an immediate halt.

The Real Competition Lies Beyond Chips

My judgment is that in the next two to three years, the key to the U.S.-China military AI competition will not be the performance score of a specific GPU, but whether "computing power—data—algorithms—deployment" can form a closed loop. The PLA has advantages in application scenarios related to drones, satellite reconnaissance, and electronic warfare data, but the quality of data labeling, cross-service sharing mechanisms, and model reliability in complex environments still lack sufficient public information. Acquiring high-end chips is merely entering the race; it does not automatically generate reliable battlefield decisions.

Therefore, what is most noteworthy in this report is not the striking figure of "900,000," but that the U.S. has already treated China's military AI as a supply chain issue. What to observe next is whether domestic GPUs in China can achieve a balance between large-scale training and military edge deployment, and whether export controls will, in turn, push the PLA to establish a more closed but more controllable computing system.

Sources and Notes

Main sources: U.S.-China Economic and Security Review Commission "2026 Annual Report" (November 18, 2026); related reports from Reuters (November 18, 2026); related reports from CNBC (November 18, 2026). Public reports have not proven that all chips have entered the PLA, and this article strictly distinguishes between "seeking," "approval," and "actual equipment."

Classification
Region
East Asia & Pacific
Analytical Domain
Strategic
Primary Category / Secondary Categories
Political-Military / Weapons & Equipment
SALUTE Report
Size
900,000 AI chips
Activity
China is seeking advanced AI chips from Nvidia for military applications and capabilities development.
Location
China
Unit
People's Liberation Army
Time
November 18-20, 2026
Equipment
Nvidia AI chips
Summary

China is seeking advanced AI chips from Nvidia for military applications, as reported by U.S. sources between November 18-20, 2026. The U.S. government has approved nearly 900,000 licenses for AI chips intended for China, although there is no confirmation of their deployment by the People's Liberation Army. The report underscores the challenges China faces in securing a stable supply of high-end AI chips for military use.

Key Facts
  • China is attempting to acquire Nvidia's advanced AI chips for military applications.
  • The U.S. government approved nearly 900,000 AI chip licenses for China in the past six months.
  • There is no evidence that these chips have been deployed by the People's Liberation Army yet.
  • The distinction between seeking and actual use of chips is emphasized in the report.
  • The report highlights challenges in sustaining a supply of high-end AI chips for military use.