China's $3 billion AI infrastructure pipeline relies on restricted Nvidia chips

The scale of the pipeline

A review by Reuters has uncovered a massive pipeline of Chinese computing projects that are being built around Nvidia's restricted processors. The investigation identified more than $3 billion in projects that specifically call for Nvidia hardware to power their AI ambitions. These developments are occurring even as the American chipmaker has stated that US export controls have effectively blocked its access to the Chinese AI data-center market.

The scope of these projects is geographically diverse, reaching from wealthy coastal technology hubs to remote western provinces. This distribution suggests that the demand for high-performance computing is a national priority across China. It underscores how deeply Nvidia's technology remains embedded in the country's broader AI expansion strategy despite growing geopolitical friction.

The findings come at a critical time for the industry following the resignation of Nvidia co-founder Jensen Huang as chief executive. Huang had been a vocal advocate for easing the export restrictions that currently limit the sale of high-end chips. His departure may leave Chinese companies more vulnerable to supply disruptions and delays as the United States continues to tighten its trade curbs.

Specific project details

One of the most significant projects identified is a proposed AI computing center in Xianyou county, located in the coastal Fujian province. This facility is seeking commercial partners and has a budget of 6.2 billion yuan, which is approximately $873 million. The project specifically requests Nvidia H100 chips to support large-scale AI development efforts.

In the northwestern Shaanxi province, a state-owned engineering group called Sinomach announced that a subsidiary won a contract for a 3 billion yuan computing center. This specific project also names Nvidia H100 and H200 processors as the required hardware. These high-end chips are the industry standard for training complex AI models but are subject to heavy oversight from the US government.

Other projects are in various stages of development across the country. In northeastern Jilin province, a government prospectus from April describes a 2.26 billion yuan project that intends to combine Nvidia H100 processors with chips developed within China. The developers of this specific facility expect to finance about 80% of the total investment through bank loans.

Infrastructure in progress

In the Inner Mongolia region, a development project proposes the installation of 625 Nvidia H200 servers. This project has a budget of 1.5 billion yuan, and industry reports from June indicated that construction had already begun on two buildings for its first phase. These plans highlight the ongoing reliance on US-made processors despite the risks associated with trade restrictions.

Further progress is visible in the Gansu province, where a local company was awarded a 1.16 billion yuan construction contract in July. This contract was for a computing center intended to house approximately 3,072 Nvidia processors, specifically the H100 and H200 models. The contract was awarded to a consortium led by a subsidiary of the state-owned Shaanxi Construction Engineering.

While the tender documents for the Gansu project specify the hardware, they do not confirm whether the processors have actually been delivered to the site yet. However, the fact that these projects are being planned with specific Nvidia models in mind shows a clear preference for American technology. Other Chinese facilities are already operational and currently using restricted Nvidia hardware to power their systems.

Risks of trade curbs

The heavy reliance on Nvidia hardware creates a significant point of failure for Chinese AI development. Because the US government is actively tightening export controls, these projects face a high risk of construction delays. If the hardware cannot be sourced, developers may be forced to halt construction or find alternative ways to build their data centers.

Beyond simple delays, these trade restrictions could lead to significantly inflated costs for Chinese developers. Companies might have to pay a premium for older, less capable chips or spend extra resources on complex workarounds. These hurdles could slow the pace of AI development in China compared to other regions with easier access to high-end hardware.

There is also the possibility of forced redesigns for these facilities. If Nvidia chips become completely unavailable, developers may have to pivot to domestic alternatives that might not be as efficient. This could lead to a fragmented AI landscape where Chinese infrastructure is built on less capable, home-grown technology. Such a shift would represent a major hurdle for companies aiming to compete on a global scale.

The path forward

The current situation highlights a paradox where China continues to plan for a future powered by the very technology it is trying to develop independently. The $3 billion pipeline shows that while the government wants to reduce reliance on the West, the current technical capabilities of domestic chips may not yet be sufficient to replace Nvidia. This creates a period of high uncertainty for investors and developers alike.

Moving forward, the stability of these projects will depend on how the US manages its export policies and how China manages its domestic production. If the US continues to tighten the screws, the projects in Shaanxi, Fujian, and Inner Mongolia may face major hurdles. These projects could become symbols of the friction between global trade and national security interests.

For the tech industry, the next few months will be critical as these construction projects move from the planning phase to reality. Observers will be watching to see if these facilities can actually secure the Nvidia hardware they specify. The outcome will determine whether China can maintain its current trajectory for AI expansion or if it will be forced to accelerate its transition to domestic alternatives.