A recent report by the United States Congressional Executive Commission on China (USCC) for the year 2026 has raised alarms over the rapid integration of artificial intelligence (AI) into weapons and operational systems by the Chinese Communist Party (CCP), a move that is pushing the concept of “intelligent warfare” forward. The publication highlights not only the growing demand from the Chinese military for high-end NVIDIA chips but also the new challenges faced by U.S. export controls. Even if these chips are not directly exported to China, the CCP could still access such technological resources through third-party companies, overseas data centers, and remote computing capabilities.
Experts have analyzed that the presence of overseas computing power presents numerous risks, further complicating the situation for the Chinese government as it attempts to navigate its AI strategy.
The report (Chapter 8, Section 3) explicitly states that the Chinese military is seeking advanced NVIDIA chips, which it intends to integrate into various military applications, including weapons platforms, command and control systems, kill chains, and logistics support.
According to an assessment conducted by Georgetown University’s Center for Security and Emerging Technologies in April 2026, documents pertaining to public procurement by the Chinese military explicitly name the acquisition of high-end NVIDIA chips for satellite reconnaissance image analysis, drone swarm coordination, and missile target identification. Furthermore, Chinese companies are utilizing data centers in third-party countries like Malaysia to remotely access the computing resources of NVIDIA chips.
In recent years, the Chinese regime has shifted its strategy towards “renting overseas computing power.” This means that when high-end chips are unavailable, they establish or rent data centers in locations such as Singapore, Malaysia, and Indonesia, allowing domestic enterprises and institutions to remotely connect via the internet to utilize the computing power of these chips.
This approach implies that as long as one can rent a server equipped with NVIDIA chips overseas, they can access the necessary computational resources. Various methods, including smuggling, third-party transshipment, and shell companies, are all interconnected. However, renting computing power is the hardest to trace, as procurement, deployment, and use happen across different countries.
Dr. Lin Song from the University of New South Wales pointed out that using overseas rented computing power involves significant risks concerning data breaches and confidentiality. For the Chinese government, maintaining a stable and long-term ability to process sensitive information within a controlled environment is far more appealing than relying on external resources.
Concrete examples further illustrate this dynamic. A report from Reuters on November 11 indicated that Northwestern Polytechnical University procured two NVIDIA H200 chips through a Chinese supplier on October 21, totaling approximately 874,700 yuan. This university is one of the “Defence Seven Sons,” which has been on the U.S. Entity List since 2001. This year alone, it rented servers equipped with H200, A100, and A800 chips through five contracts.
Dr. Lin notes that procurement announcements do not equal delivery, and renting does not mean ownership. The difference between the two encapsulates the predicament faced by the CCP.
Cases such as Megaspeed, registered in Singapore, exemplify these practices. Professors like Xie Tian from the University of South Carolina reported that Australian media discovered Megaspeed Australia is a shell company with no employees. The Wall Street Journal reported on November 14 that this company arranged for nearly $12 billion worth of NVIDIA chips to be directed towards Chinese enterprises between March 2024 and June 2025.
Since February 2025, Singapore has been investigating this shell company but found insufficient evidence to charge its former head, Huang Le, although the investigation is ongoing. Megaspeed has denied any wrongdoing amid speculation about Huang’s ties to intelligence activities. Huang is reportedly a recent mistress of Jensen Huang, the founder and CEO of NVIDIA, who has since resigned.
The scale of the CCP’s “rental computing power” strategy is larger and more challenging to trace than smuggling activities. Data from Wirescreen revealed that the company examined around 3,800 procurement records from the Chinese military, uncovering over 500 instances between 2019 and 2025 where military units in China sought NVIDIA chips, including models like A100, A800, H100, and H800. These records reflect procurement requests rather than confirmed deliveries.
Researcher Zeng Zhi-Hua from Taiwan’s Center for Technology, Democracy, and Society emphasized that the distribution of procurement, deployment, and usage across different countries complicates verification of end-users.
Dr. Lin believes that the U.S. has increasingly identified how the CCP circumvents sanctions. Initially, it was obvious that third-party companies were acquiring these technologies. However, when that route was closed off, they discovered fraudulent companies in Singapore and Malaysia, ultimately leading to the rental of computing capabilities through overseas data centers.
He mentioned that the more routes that the U.S. blocks, the more alternative paths the CCP will take. Yet, with each alternative comes increased costs and risks, leading to more unstable computing power. Military training necessitates large-scale, long-duration, low-latency computing clusters, while rented computing capabilities are often scattered and struggle to support a sophisticated intelligent warfare framework.
The report also notes that Chinese AI chip manufacturers like Huawei’s Ascend and Cambricon’s Tao cannot compete with the likes of NVIDIA due to bottlenecks in advanced manufacturing processes, high-bandwidth memory, semiconductor equipment, and NVIDIA’s CUDA software ecosystem.
Professor Shen Ming-Shih, affiliated with Taiwan’s National Defense Security Research Institute and Tamkang University’s Institute of International Affairs and Strategic Studies, asserted that a key issue lies in China’s inability to produce high-end chips domestically. NVIDIA’s H200, H800, and Blackwell primarily utilize TSMC’s 2-nanometer and 3-nanometer manufacturing processes, which, under restrictions, leave the CCP with few options: smuggling, illicit procurement, or utilizing third-party transshipment. Military educational institutions may seek to acquire chips under the guise of academic cooperation; however, major universities closely affiliated with the military, like the “Defence Seven Sons,” ultimately procure chips for defense or military purposes.
Shen pointed out a crucial detail: the lithography machines produced domestically by the CCP are cobbled together, functional but inferior to the original products. However, through repetition and optimization, their performance can gradually improve. This concern is precisely what the U.S. fears — that lagging behind is not a static condition and that stolen technologies could be absorbed, imitated, and reproduced. Conversely, the act of stealing itself is evidence of underdevelopment: a nation capable of manufacturing high-end chips independently wouldn’t need to risk sanctions through shell companies and third-party rentals. The greater the urgency from the CCP to acquire such technology, the more it underscores its limited access.
Xie Tian contended that the CCP still has significant capability gaps in the high-tech field, with the acquisition of U.S. technology representing a critical avenue for its catch-up efforts.
Dr. Lin’s assessment is more direct: the CCP’s chips lag significantly behind comparable NVIDIA chips, with NVIDIA’s most advanced offerings yet to be released. He believes there is little hope for the CCP to catch up but anticipates that they will continue to aggressively pursue advancements, even if the gap between them and leading players may only widen.
The U.S. response is also catching up to these evolving challenges. The eighth key recommendation outlined in the report suggests that Congress revise the 2018 Export Control Reform Act so that exporters must prove that exports of AI chips to China do not risk relinquishing essential capabilities for the U.S. market, with false declarations bearing criminal penalties. The third recommendation targets the access of Chinese companies to weight-restricted AI models, urging the Department of Commerce to identify sensitive uses within 180 days, ultimately prohibiting new deployments in critical infrastructure.
On November 16, the Department of Commerce’s Bureau of Industry and Security released temporary final rules requiring data centers that provide advanced chips capable of more than 12 million total processing capabilities to restricted entities like China must first obtain a permit. This threshold is approximately equivalent to 750 H100 or H200 chips. This marks the first time the U.S. has expanded its regulatory focus from simply “selling chips” to includes “renting computing power.”
On November 17, NVIDIA’s acting CEO, Collette Kress, confirmed that the company has ceased shipments of data center chips, including the H200 to China. The concurrently released financial report indicated that NVIDIA generated revenues of $115.8 billion, reflecting a 103% year-on-year increase. Even without the Chinese data center market, NVIDIA continues to experience rapid growth, diminishing the claim that “regulations will cripple American companies,” thus reducing opposition to tightening controls in Congress.
Two crucial developments will be closely monitored: first, whether the gap in remote computing power truly narrows following the 90-day approval window stipulated by the new Department of Commerce regulations; and second, on December 1, whether Jensen Huang’s testimony before Congress will help solidify temporary regulations into a long-term structure.
Dr. Lin expressed skepticism regarding the extent of the loopholes and the CCP’s benefits, indicating that the truth might remain elusive. However, one fact stands clear: while the CCP can rent access to computing power, it cannot reacquire the chips, and the struggle between control measures and countermeasures has now shifted from the chips themselves to determining who can effectively harness the computing capabilities.
In summary, the analyses of the three experts ultimately converge on one core issue: the CCP’s drive toward military modernization necessitates the acquisition of advanced chips and computing power from external sources. The success or failure of U.S. export controls will hinge on the ability to trace chip transactions back to cross-border computing capabilities, and subsequently verify the end users and their intended applications.
Looking ahead, it will be crucial to observe not just whether the Chinese domestic chip manufacturing capabilities can narrow the technology gap but also whether the U.S. can effectively close off channels for third-party transshipments and remote computing power access.