An investigation by The Wall Street Journal has identified a supply chain allegedly used by Shanghai-based AI firm INF Tech to access NVIDIA A100 graphics processing units (GPUs) via an Indonesian data centre. This route is reported to bypass US export controls aimed at limiting China’s access to advanced AI hardware.
INF Tech focuses on developing AI models. The A100 GPUs are subject to US restrictions that prohibit their sale to Chinese entities that could advance military capabilities.
Bypassing US export controls
The data centre in question is operated by PT Riung Mitra Lestari (RML) in Indonesia. RML provides cloud services and data centre infrastructure. The investigation suggests third-party facilities like RML’s are used by companies to obtain restricted hardware and circumvent US controls. This alleged circumvention underscores the complex geopolitical landscape surrounding advanced semiconductor technology and its dual-use potential. The A100, a powerful chip designed for high-performance computing and AI training, is a critical component for developing sophisticated artificial intelligence systems, including those with potential military applications. The US government’s export controls are a direct response to concerns that such technology could bolster China’s military modernization efforts, creating a significant strategic advantage.
NVIDIA’s A100 GPUs are used for training large-scale AI models. US export controls on these chips to China began in late 2022, with further restrictions in October 2023 impacting more advanced chips, including NVIDIA’s H100. These controls are designed to prevent the transfer of technology that could be used to develop advanced weaponry, surveillance systems, or other capabilities that could threaten US national security interests or those of its allies. The continuous evolution of these restrictions reflects the dynamic nature of technological advancement and the ongoing efforts by governments to manage its proliferation.
Indonesian data centre loophole
The reported pathway involves INF Tech contracting RML for computing resources. RML allegedly procured NVIDIA chips, potentially through intermediaries or by sourcing them in markets with less stringent export controls. RML’s Indonesian operations are cited as a location outside direct US export enforcement reach, enabling INF Tech’s use of the hardware. This strategy leverages the global nature of cloud infrastructure and data centre services. By operating in a jurisdiction not directly subject to the same level of US oversight, companies like RML can potentially facilitate access to restricted technologies for entities that would otherwise be unable to acquire them. The specifics of how RML allegedly obtained the GPUs remain a key point of the investigation, with possibilities including diversion through third-party distributors or acquisition in regions with less restrictive trade policies.
The technical specifications of the A100 GPU make it highly desirable for AI workloads. Its architecture is optimized for parallel processing, crucial for the computationally intensive tasks involved in training deep learning models. This includes tasks like image recognition, natural language processing, and complex simulations. The ability to access such hardware is a significant enabler for AI research and development, allowing companies to accelerate their innovation cycles. For INF Tech, gaining access to these chips would be vital for their stated goals in autonomous driving and advanced NLP, areas where cutting-edge AI is paramount.
Global supply chain challenges
This situation highlights enforcement challenges for governments regarding advanced technology export controls, particularly with global data centre operations. Companies can use international infrastructure to acquire hardware otherwise inaccessible. The interconnectedness of the global technology supply chain presents a formidable challenge for regulators. Identifying and intercepting the flow of restricted components requires sophisticated monitoring capabilities and international cooperation. The alleged use of an Indonesian data centre by a Chinese AI firm illustrates how geographical dispersion and the outsourcing of computing power can create blind spots for enforcement agencies. This necessitates a more adaptive and collaborative approach to export control, potentially involving greater scrutiny of data centre operators and their procurement practices.
INF Tech aims to develop AI models for autonomous driving and natural language processing. High-performance GPUs are essential for training these AI systems. Circumventing US export controls would allow INF Tech to continue R&D without direct sanctions. The enterprise implications are significant; if INF Tech can continue to develop advanced AI models using restricted hardware, it could gain a competitive advantage in the global AI market, potentially impacting the market share of companies operating within the bounds of export controls. The timeline for such developments is difficult to ascertain without direct access to the hardware, but the ability to train larger and more complex models more rapidly is a clear benefit.
PT Riung Mitra Lestari stated to The Wall Street Journal that its operations comply with all applicable laws and regulations. NVIDIA confirmed adherence to export control regulations and is investigating the matter. The company’s response indicates a commitment to compliance, but the investigation will likely delve into the specifics of their supply chain and customer vetting processes. The US Department of Commerce oversees export controls and is increasing monitoring and enforcement of AI hardware restrictions. The alleged use of third-country data centres represents a potential loophole for regulators to address. How regulators will adapt their strategies to counter such circumvention tactics remains a key question, potentially involving enhanced due diligence requirements for companies involved in the international trade of sensitive technologies and increased collaboration with foreign governments to ensure compliance.
