Skip to main content
Reports

China’s Procurement of Restricted NVIDIA AI Chips

By February 3, 2025No Comments

By Sam Burgiss, Exovera, Director of Product Strategy

Access to the procurement records linked in this article is available via Exovera’s exoINSIGHT platform. If you don’t have a subscription, contact us for a free trial account.

The U.S. Department of Commerce’s Bureau of Industry and Security (BIS) has progressively tightened export controls on advanced semiconductor technologies, particularly impacting NVIDIA’s high-performance chips.

  • October 7, 2022: The BIS imposed new license requirements for items destined for supercomputer or semiconductor development or production end-uses in China, which applied to NVIDIA’s A100 and H100 chips.
  • October 17, 2023: The BIS announced a rule expanding existing controls to include additional advanced computing items and imposed further licensing requirements for exports to China, which affected the export of NVIDIA’s A800 and H800 chips.
  • December 2, 2024: The BIS introduced significant new controls on high-bandwidth memory used in the NVIDIA H200 and upcoming B200 chips.

Multiple Chinese entities, including government-affiliated institutions and military procurement bodies, have procured NVIDIA chips as if no rules existed. These chips were publicly tendered and acquired through various procurement channels. Notable companies involved in these procurements include Beijing Jinghong Anxin Technology Development Co., Ltd., Shenzhen Junzheng Technology Co., Ltd., Nanjing Gezhiwu Technology Co., Ltd., and Beijing Parallel Technology Co., Ltd.

Guangdong Institute of Intelligent Science and Technology Procurement of H100 GPUs

On April 9, 2024, the Guangdong Institute of Intelligent Science and Technology acquired 10 NVIDIA H100 GPUs through a contract awarded to Shenzhen Junzheng Technology Co., Ltd. The procurement was part of a broader effort to enhance the institute’s computing infrastructure, likely supporting AI research and development.

This acquisition illustrates a growing pattern of sanctioned NVIDIA hardware procurement by Chinese government-linked entities. Institutions such as Tsinghua University and the Chinese People’s Liberation Army (PLA) have also been documented obtaining similar high-performance GPUs, highlighting a coordinated effort to bypass restrictions and sustain China’s AI development. The involvement of Shenzhen Junzheng Technology Co., Ltd., a domestic supplier, suggests that Chinese entities are leveraging intermediary companies to facilitate access to restricted semiconductor technology. These ongoing procurements reinforce concerns that such hardware could be repurposed for military applications, AI-driven surveillance, or other strategic initiatives aligned with China’s national security objectives.

Procurement of NVIDIA H100 GPUs by Tsinghua University

On June 17, 2024, Tsinghua University announced the procurement of NVIDIA H100 accelerator cards as part of the Qingcai Bixuan No. 20240391 project. The winning bidder for this procurement was Beijing Jinghong Anxin Technology Development Co., Ltd. The openness of this procurement action indicates that the acquisition followed a structured bidding process.

The technical specifications of the acquired hardware confirm that the purchased NVIDIA H100 accelerator cards are built on the GH100 Hopper architecture. These GPUs are designed for large-scale AI training and inference tasks.

Procurement of NVIDIA H800 NVL GPUs by Tsinghua University

On July 17, 2024, Tsinghua University announced the procurement of NVIDIA H800 NVL high-performance computing cards as part of the Qingcai Bixuan No. 20240597 project. The winning bidder for this procurement was Nanjing Gezhiwu Technology Co., Ltd.

 

The technical specifications of the acquired hardware confirm that the purchased NVIDIA H800 NVL GPUs feature 94GB of HBM3 memory with a bandwidth of 3.9TB/s. The contract stipulates a service period of 36 months, with delivery expected within 10 working days after signing.

Chinese Military Announcement of Intent to Procure NVIDIA H100 GPUs

On July 25, 2024, the Logistics Support Department of the Central Military announced its intent to procureintelligent computing servers equipped with four NVIDIA HGX H100 NVLink GPUs as part of an Intelligent Computing Server Procurement Project. The project’s goal is to support artificial intelligence algorithm calculations and large-scale model training.

The procurement notice highlights the need for high reliability, fault tolerance, and concurrent access capabilities—essential for deep learning model development, particularly in applications requiring large-scale simulations, real-time decision-making, and automation.

Procurement of AI Large Model Computing Services by Huazhong University of Science and Technology

On September 30, 2024, Huazhong University of Science and Technology announced the procurement of AI large-model computing power services under the project HW20240372. The winning bidder for this procurement was Beijing Parallel Technology Co., Ltd.

The contract includes provisions for AI model training and inference computing power. The technical specifications indicate that the supplier will provide NVIDIA Tesla H800 GPUs for model training, with a usage allocation of 43 unit months. The procurement is structured as a 24-month service agreement.

Chinese Military Announcement of Intent to Purchase H100 GPUs and 4090 Services

On January 22, 2025, the Logistics Support Department of the Chinese Communist Party’s Central Military Commission announced an intent to procure notice detailing multiple server acquisitions aimed at enhancing artificial intelligence and computing capabilities. The acquisition included one 8-card H100 server (80GB per GPU). The announcement requires “genuine licensed products.”

Implications

The procurement of sanctioned NVIDIA components by Chinese government and government-linked institutions, including Tsinghua University and the PLA via the Logistics Support Department of the Central Military Commission, highlights China’s continued practice to openly secure advanced AI hardware despite U.S. export restrictions. These GPUs provide Chinese research and military entities with critical computing power necessary for pursuing artificial intelligence development. The evidence of acquisitions of such high-performance accelerators through companies providing restricted hardware and access to hardware through service-structured contracts suggests that providers are circumventing U.S. restrictions.

 

Given the Chinese government’s prioritization of AI for both economic and national security purposes, procurement of restricted NVIDIA hardware by any Chinese company represents a strategic challenge to U.S. technological controls. While recent U.S. restrictions aim to limit China’s access to advanced semiconductor technology, the persistence of these procurement efforts underscores the need for enhanced supply chain monitoring, improved ‘Know Your Customer’ research, and targeted countermeasures to prevent further acquisition of restricted AI hardware.