As the United States tightens its blockade on advanced AI chips destined for China, Southeast Asia—particularly Malaysia—has become a critical node in the emerging landscape of compute diversion. A late October Reuters report describes how a Republican senator urged the Trump administration to help Malaysia in cracking down on networks smuggling AI chips through the country into China, underscoring Washington’s growing concern about these “backdoor” channels. At the same time, Malaysian authorities have begun stepping up enforcement. Yet enforcement alone cannot close the loopholes. The real question is how Washington can work with Malaysia to shape the rules of AI infrastructure—rather than policing smuggling from the outside.

In a new policy report published by DSET, a government-funded think tank in Taiwan, we suggest that the US export control regime, besides improving its capacity for investigation, should target the PRC-linked cross-border supply chain in Malaysia and build partnership frameworks to protect overseas AI infrastructures. Building on that report, this article argues that Washington should treat Malaysia not just as a transshipment risk, but as a “compute neutral” whose AI infrastructure choices will determine whether US export controls hold or leak.

Throughout this analysis, we will use the term “compute” which, in the artificial intelligence (AI) and export control context serves as shorthand for effective AI computing capacity — encompassing not only physical chips, but also data center infrastructure and cloud-based access to training and inference resources.

The rise of the “black compute network”

For years, Malaysia has been viewed as a critical node for China’s circumvention of US export controls. Since Washington’s 2022 decision to ban the export of AI-training GPUs to China, related trade volumes have surged in Singapore and Malaysia.

In February this year, Singapore police arrested three men for illegally selling servers containing Nvidia chips that were later re-exported to Malaysia. The case shows Chinese companies leveraged third countries to rebuild a “black compute network” outside China’s borders.

By “black compute network,” we refer to overseas data centers and server supply chains that provide Chinese entities with de facto access to restricted US chips—without any of the visibility or safeguards built into the export control system. In practice, a “compute blockade” that focuses only on where chips are shipped, but not where they are used, is bound to leak.

Malaysia: The hub of “compute neutrality”

Malaysia sits precisely within this gray zone of the supply chain. Leveraging its strategic location, established semiconductor backend industry, and long-standing non-aligned foreign policy, it has become a nexus for two conflicting activities: on one hand, it serves as a logistics hub for illicit chip transshipment; on the other, it attracts Chinese firms legally investing in high-performance data centers. As highlighted in our DSET report, Malaysia has emerged as a symbol of “compute neutrality”—neither an outpost of the West nor a proxy for Beijing, but a strategic crossroads courted by both.

The DSET report classifies AI data center deployment in this region into four operational models—Greenfield, Joint Venture, Colocation, and Hybrid Cloud—each reflecting different levels of supply-chain control and risk exposure. In this article, we focus on two of them—joint ventures and hybrid cloud arrangements—because they already offer concrete examples of compute diversion risk in Malaysia.

From risk to market reality

Beyond risk analysis of these four models, there is ample empirical evidence of chip diversions and computing power leakages.

First, in a joint venture case, Malaysia’s YTL and China’s GDS co-developed a large-scale data center deploying Nvidia’s Blackwell-based servers, serving clients such as Singapore-based Sea Ltd., a company with partial Chinese ownership. While the partnership is framed as contributing to Malaysia’s digital economy, it simultaneously enables Chinese-linked actors to access advanced compute resources outside China’s borders. The contrast is notable: numerous data centers in China remain underutilized due to their inability to obtain restricted GPUs.

Second, in the hybrid cloud case, The Wall Street Journal reported in March 2025 that several Chinese engineers carried roughly 4.8 petabytes of training data—stored in high-capacity SSDs—into Malaysia, where they rented Nvidia-powered servers for model training before returning to China. This episode demonstrates how hybrid cloud arrangements allow data and compute to flow across borders even when hardware exports are blocked—turning physically located export controls into a largely virtual problem.

Kuala Lumpur’s strategic paradox

Meanwhile, Malaysia’s shifting policy stance shows growing awareness of these risks. In April, China and Malaysia signed several MOUs to deepen bilateral cooperation amid trade tensions. Yet only a month later, following updated guidance from the US Bureau of Industry and Security (BIS), the Malaysian government distanced itself from the plan to use Huawei’s Ascend chips in its national “sovereign AI” initiative. By July, Malaysia’s Ministry of International Trade and Industry (MITI) announced cooperation with US authorities in investigating illegal GPU smuggling and introduced a licensing system for US-origin advanced chips.

Taken together, these developments reveal Malaysia’s strategic paradox: it seeks to capture the economic benefits of the AI boom and work with economic and financial interests from both Western and Chinese sides while avoiding entanglement in great power competition. For Washington, Malaysia is not merely a frontline against chip smuggling—it has become a testing ground for AI infrastructure diplomacy. The country’s long-standing non-alignment stance now sits precariously between compute mobility and geopolitical gravity.

The traditional US export controls directly on Chinese firms—while temporarily slowing China’s technological ascent—have spawned alternative paths for Chinese firms to leverage gray-market supply chains and third-country compute markets that dilute policy effectiveness. We suggest that verification measures for the supply chain and investigation of indirect transfers should be the foundation for the goal of building partnerships on AI infrastructures with these countries, avoiding pushing them deeper into China’s economic orbit.

Strengthening US-Malaysia cooperation and supply chain transparency

To sustain export controls without alienating regional partners, Washington must leverage American technological leadership with clearer regulatory requirements in partnership agreements, setting up a healthy environment for exporting the American AI technology stack. For example, following the United States–Malaysia Agreement on Reciprocal Trade (ART), Washington is set to expand trade and investment with Malaysia, valued at roughly $150 billion. The US has also included clauses to ensure Malaysia does not jeopardize essential US interests. While it is a sweeping alignment test for Malaysia, we encourage US policymakers to recognize the necessary rules and regulations to safeguard American AI and a transparent AI supply chain.

To address this, the US and Malaysia should jointly establish an AI Infrastructure Partnership Framework, protecting overseas data centers with US-origin technologies and services by disclosing their supply chains and end-client structures. Besides leveraging the American technological leadership, policymakers can also embed due diligence requirements on key American suppliers to allow Washington to oversee outbound investment, ensuring that American technology does not indirectly empower Chinese-linked entities. At the same time, the partnership framework would encourage Malaysia to align with transparent, rules-based investment practices.

Other than the partnership framework, we also encourage Washington to launch a Trusted AI Infrastructure Whitelist Program, combining with a Usage Declaration Cross-Verification System, to promote transparency, detect diversion, and protect privacy. Together, these measures would strengthen US oversight of outbound investment, close regulatory blind spots, and foster trusted AI infrastructure in Southeast Asia.

Ultimately, maintaining technological security while pursuing targeted cooperation will be essential. Building secure and transparent AI infrastructure abroad will strengthen supply-chain resilience and reaffirm American leadership in defining the standards of the next AI era.

Jin Chian Seer ([email protected]) is a policy analyst at the Research Institute for Democracy, Society, and Emerging Technology (DSET), focusing on economic security and emerging technologies. He is a co-author of the DSET report “A Shared Future?,” referenced in this article.

Chris Chih-Hua Tseng ([email protected]) is a non-resident fellow at DSET and a sociology Ph.D. Candidate at the University of California, Irvine. He is a co-author of the DSET report “A Shared Future?,” referenced in this article.

Main image: Nikkei

The Pilot commentaries and responses represent the views of the respective authors. Alternative viewpoints are always welcomed and encouraged. Please write to [email protected] for more information on how to contribute.