Chinese companies hope to deploy about 115,000 Nvidia H100 and H200 GPUs—chips which have been restricted from sale to China by the U.S. government since 2022 over security concerns
In sum – what to know:
Unlicensed AI chip plans – Chinese firms aim to build data centers in Xinjiang using 115,000 Nvidia chips, despite strict U.S. export restrictions on such hardware.
Energy-rich desert strategy – Xinjiang’s access to solar, wind and coal power makes it a strategic location for energy-hungry AI compute, with most capacity focused in a state-backed compound.
Feasibility in question – U.S. officials are skeptical about China’s ability to secure so many banned chips, with Nvidia warning that smuggled products don’t support scalable or efficient data center deployment.
Chinese firms are reportedly planning to build dozens of data centers in Xinjiang, potentially housing nearly 115,000 restricted Nvidia AI chips, according to a Bloomberg investigation.
The report is based on a review of investment approvals, tenders and corporate filings. The only company named in the report is Nyocor, an energy firm based in Tianjin.
In its investigation, Bloomberg states that in the fourth quarter of 2024 alone, local governments in Xinjiang and nearby Qinghai greenlit 39 data center projects. As of June 2025, at least seven of those projects were either under construction or had secured tenders for AI compute services. Around 70% of the computing power linked to these data center developments is reportedly centered in a single compound backed by the Xinjiang regional government.
While the compound wasn’t specifically named in Bloomberg’s reporting, the Xinjiang government had previously announced a plan in April 2024 to build a large-scale “converged computing center” in Urumqi. The facility would offer supercomputing, AI computing and data storage and was described as having national-level computing capacity.
Based on Bloomberg’s review of investment plans, the involved Chinese companies hope to deploy about 115,000 Nvidia H100 and H200 GPUs—chips which have been restricted from sale to China by the U.S. government since 2022 over security concerns. In April 2025, the U.S. expanded restrictions to include the H20 chip, a China-specific version designed to comply with earlier export rules. “As a result of these new requirements, Nvidia incurred a $4.5 billion charge in the first quarter of fiscal 2026 associated with H20 excess inventory and purchase obligations as the demand for H20 diminished,” Nvidia previously stated.
Despite the ban, some Chinese firms are reportedly still attempting to procure the restricted hardware. Nvidia began developing a new line of chips for China in May 2025 that would align with current U.S. export controls.
Bloomberg consulted individuals familiar with ongoing U.S. investigations into chip exports, as well as sources knowledgeable about China’s black-market hardware channels. Those sources said they were unaware of the reported scale of the Chinese data center plans and doubted the existence of a network capable of smuggling over 100,000 high-end processors. U.S. officials estimate that only around 25,000 restricted Nvidia chips have made it to China.
Nvidia responded to Bloomberg saying that a company merely posting online inquiries for restricted GPUs does not equate to building a functioning data center. The company also stressed that constructing a modern data center requires extensive support and engineering, and added that the firm does not offer services or repairs for restricted chipss. It added that relying on outdated or smuggled chips to power data centers is neither efficient nor cost-effective, especially given the local availability of Huawei chips.
The investigation also noted that the Xinjiang government and China’s Ministry of Industry and Information Technology did not respond to requests for clarification or comment about these projects.
