BEIJING, November 13, 2026 — Artificial-intelligence computing projects worth more than 23 billion yuan, or about $3.2 billion, across China have been built around or planned for advanced Nvidia processors subject to U.S. export controls, according to a Reuters review of more than 150 procurement notices, investment plans, contracts and company disclosures.

The projects involve Nvidia’s H100, H200 and H800 processors and span major technology centres as well as less prominent regions. The review identified activity involving local government-backed ventures, universities and regional data centres, with some projects already operating or under construction and others facing delays, supply disruptions or costly redesigns as export restrictions tighten.

The $3.2 billion figure is an estimate of the value of the wider AI-computing projects and related infrastructure. It is not a tally of Nvidia chip purchases, nor does it mean that all the equipment has been delivered or installed. Reuters said the total includes projects that are planned or signed, completed installations and supporting infrastructure.

NVIDIA H100 PCIe artificial intelligence accelerator card
NVIDIA H100 PCIe GPU shown in a 2023 product-video screenshot. Creator: 极客湾Geekerwan / Wikimedia Commons, CC BY 3.0. This is an illustrative product image, not evidence of a specific chip shipment. Image source · Licence.

What The Reuters Review Identified

Reuters examined public procurement records, investment plans, contracts and company disclosures to identify projects designed around advanced Nvidia processors. The review found projects in a range of Chinese locations, including large technology centres and provincial areas where regional governments, universities and local enterprises are investing in high-performance computing capacity.

The projects include data-centre and AI-computing infrastructure intended to support model development, research and commercial applications. Their scale reflects the demand for powerful processors as Chinese institutions and companies expand AI training and deployment, even while access to some high-end foreign chips is subject to changing export rules.

The report does not establish how every restricted processor reached a particular project. It also does not prove that any specific institution or company violated U.S. export-control law. A procurement plan, a signed contract, a completed installation and an operating system are different stages of a project, and they should not be treated as interchangeable evidence of chip delivery.

Why H100, H200 And H800 Chips Matter

Nvidia’s H100 and H200 processors are designed for demanding AI workloads, including the training and deployment of large models. The H800 was developed for the Chinese market after earlier U.S. restrictions, but subsequent changes to export-control thresholds also affected its availability.

Advanced AI accelerators are valuable because they can perform large numbers of calculations in parallel and provide high-bandwidth memory for the data-intensive workloads used in modern machine learning. Large clusters of these processors can support model training, scientific computing and commercial AI services.

For companies building data centres, access to a reliable supply of compatible accelerators can influence project costs, construction schedules and the choice of software. A project designed around one processor architecture may need expensive changes if the planned hardware becomes unavailable or if licensing conditions change.

National Supercomputer Center in Guangzhou, China
National Supercomputer Center in Guangzhou, photographed on April 17, 2015. Photo: O01326 / Wikimedia Commons, CC BY-SA 4.0. This is illustrative Chinese computing infrastructure and is not identified as a site in the Reuters review. Image source · Licence.

How U.S. Export Controls Have Changed

Washington has tightened controls on advanced computing chips and related semiconductor technology over concerns that high-performance systems could strengthen China’s military and strategic capabilities. The rules have evolved as chipmakers have introduced products with different levels of processing power, memory bandwidth and interconnect performance.

In January 2026, the U.S. Department of Commerce’s Bureau of Industry and Security revised its licensing policy for certain advanced computing chips, including Nvidia’s H200 and comparable products. The policy moved specified exports from the United States to China and Macau from a presumption of denial to case-by-case review, subject to conditions designed to protect U.S. supply and security interests.

The conditions include requirements relating to the availability of supply for U.S. customers, recipient screening and compliance procedures, and independent third-party testing in the United States. The rule does not amount to unrestricted permission to sell advanced processors to any Chinese buyer. Exports, re-exports and transfers from abroad can remain subject to more restrictive treatment under the applicable rules.

The distinction is important because the H100, H200 and H800 do not all have identical regulatory status. The January policy specifically addressed the H200 and similar products under defined conditions; it should not be interpreted as a blanket approval for every Nvidia accelerator or every Chinese customer.

Nvidia Says It Has Effectively Lost China’s Data-Centre Market

Nvidia has told investors that U.S. export restrictions and the regulatory environment in China have prevented it from offering a competitive data-centre product that can receive approval from both governments. In its financial disclosures, the company said it had effectively been foreclosed from competing in China’s data-centre computing market.

Reuters cited an IDC estimate that Nvidia accounted for 34% of Chinese AI-accelerator shipments in the first half of 2026. That figure measures shipments during a specific period; it is not the same as the share of installed computing capacity, the number of chips currently operating in China or the value of the projects identified in the Reuters review.

The difference between shipment data and project plans is central to understanding the market. A project can be designed around a particular processor even if its planned supply is delayed, changed or never completed. Conversely, hardware already installed may continue to operate even when new shipments become harder to obtain.

NVIDIA headquarters building in Santa Clara, California
NVIDIA headquarters in Santa Clara, California, photographed on August 4, 2018. Photo: Coolcaesar / Wikimedia Commons, CC BY-SA 4.0. Image source · Licence.

Projects Face Delays And Redesign Costs

For Chinese institutions and companies, uncertainty over access to advanced processors creates practical challenges. Data-centre operators may need to revise equipment plans, seek alternative suppliers, adjust system architecture or delay the launch of services while they assess which processors can legally and reliably be obtained.

Changing the hardware can also require software adjustments. AI models and applications are often optimised for particular chip architectures and software ecosystems, so switching to a different accelerator can involve new engineering work, performance testing and changes to how systems are deployed.

China has invested heavily in developing domestic semiconductor and AI-computing capabilities. Local processors and software platforms are intended to reduce dependence on foreign suppliers, but building a competitive ecosystem requires time, manufacturing capacity, software support and a large base of developers and customers.

U.S. restrictions therefore create both immediate supply challenges and longer-term incentives for Chinese companies to develop alternatives. The eventual effect will depend on how quickly domestic processors improve and whether they can deliver comparable performance, reliability and cost at scale.

Why The Dispute Matters Beyond The Two Countries

The competition over advanced AI chips has implications for global technology companies, data-centre operators, universities and investors. The United States remains a major source of advanced chip designs and semiconductor software, while China is a large market for computing infrastructure and a significant participant in the global technology supply chain.

Export controls can change where companies build facilities, which processors they use and how they structure international partnerships. Suppliers of power systems, cooling equipment, networking hardware and data-centre construction services may also be affected when projects are delayed or redesigned.

The policy debate is complicated by the fact that advanced processors have commercial and strategic uses. The same hardware can support scientific research, business applications and civilian AI services, while governments are concerned about possible military applications. Regulators must therefore balance national-security objectives with the commercial interests of technology firms and the wider effects on innovation.

HeadlinesXtra Analysis

The Reuters findings show the scale of China’s demand for advanced AI-computing infrastructure, but the headline figure needs careful interpretation. More than 23 billion yuan in identified projects does not mean that China purchased $3.2 billion worth of Nvidia chips. The total includes planned and signed developments, installed systems and supporting infrastructure, so it is a measure of project value rather than a direct measure of chip inventory.

The review also highlights a central challenge for export controls: rules can restrict direct shipments, but they do not automatically erase existing installations, contracts, indirect supply chains or projects already under development. Enforcement requires reliable information about end users, ownership structures, resales, remote access and the actual destination of computing capacity.

At the same time, the findings should not be used to imply that every project involving a restricted processor is illegal. The regulatory status of H100, H200 and H800 products has changed over time, and the H200 has been subject to a specific case-by-case licensing policy since January 2026. A project’s value or hardware design alone does not prove that an export-control violation occurred.

For Nvidia, the risk is strategic as well as financial. If restrictions keep the company out of a large market for long enough, Chinese competitors may gain customers, developers and software support that strengthen their position elsewhere. For Chinese operators, however, replacing advanced foreign chips at scale is not a simple process. Performance, energy use, reliability and software compatibility all matter.

The most useful next step is greater transparency. Regulators should make licensing rules clear, companies should disclose material supply risks accurately, and public procurement records should distinguish planned projects from equipment actually delivered and operating. That would allow policymakers and investors to assess the real impact of export controls without confusing project announcements with completed computing capacity.

Sources And Further Reading

Image Credits And Licences