Nvidia’s dominance in the high-performance semiconductor market has positioned U.S. export controls as a critical lever in Washington’s strategy to maintain its technological edge over China in the burgeoning artificial intelligence race. However, reports indicate that despite these stringent U.S. restrictions on the export of advanced Nvidia chips, including the highly sought-after GB300 series, several Chinese firms have managed to secure access to their substantial computing power through data centers located in Southeast Asia.
This circumvention strategy has drawn significant attention. Less than a week after Moonshot AI unveiled a new AI model in July, a White House official publicly accused the company of leveraging Nvidia’s GB300 chips via a facility situated in Thailand. Moonshot’s Kimi K3 represents one of a new generation of Chinese AI models that have demonstrated remarkable performance gains in recent months, underscoring the escalating AI supremacy contest between the United States and Beijing. Companies like DeepSeek and Alibaba have also recently launched sophisticated AI systems that have achieved impressive scores on industry performance benchmarks.
Industry analysts suggest that access to advanced computing resources, often facilitated by overseas cloud service providers, is a pivotal factor enabling Chinese AI models to rapidly enhance their capabilities. While legislative discussions are underway in the U.S. to address this perceived loophole, significant hurdles remain before any new regulations can effectively curb this practice.
**Navigating the Global Compute Landscape: Chinese Firms and Remote Nvidia Access**
The U.S. export control framework currently targets the physical shipment of Nvidia’s most advanced AI chips to China. While less powerful semiconductor offerings are permitted, the cutting-edge processors that are crucial for training state-of-the-art AI models remain off-limits.
Cassia King, a senior researcher focusing on compute policy at the Institute for AI Policy and Strategy, explained to CNBC that Moonshot’s reported access to computing resources via a Thai facility would likely be considered legal as long as Moonshot itself is not directly purchasing or owning the physical hardware. She elaborated that the U.S. export control regime’s scope is confined to physical AI chips, and it does not extend to the remote utilization of these chips.
When queried about Chinese entities accessing Nvidia’s computing power through international channels for AI model training, a White House representative reiterated the administration’s commitment to maintaining robust export controls, stating, “The Trump administration has implemented the most rigorous export control regime in modern history, and remains committed to safeguarding America’s national and economic security.” The U.S. Department of Commerce and the Bureau of Industry and Security (BIS) have not yet responded to requests for comment on this matter.
Reports suggest that major Chinese technology companies, including ByteDance, Alibaba, and Tencent, have been utilizing remote access to Nvidia chip compute power through cloud providers in other Asian nations such as Thailand, Malaysia, and Japan. ByteDance and Tencent have not commented on these reports, while Alibaba declined to provide a statement.
According to sources familiar with the situation who spoke on condition of anonymity due to the private nature of the information, ByteDance has reportedly collaborated with Aolani, a Singapore-headquartered cloud provider that utilizes Nvidia chips, to access computing resources in Malaysia. This arrangement was first reported by The Wall Street Journal in March. Aolani, in a statement to CNBC, indicated that it serves “a global and diversified customer base spanning customers from North America and Asia.” The spokesperson emphasized that the companies they service “do not have ownership, potential future claim or physical access to the chips that power our solutions,” and that “Any permitted access to our services, infrastructure or technology is fully compliant with all applicable regulations.”
The burgeoning demand for advanced computing power is fueling a significant expansion of AI infrastructure across Southeast Asia. Real estate advisory firm JLL projects that global data center capacity could nearly double to 200 gigawatts by 2030. Data compiled by DC Byte reveals a substantial pipeline of new data center developments, with 31 planned facilities exceeding 100 megawatts in Malaysia, Indonesia, and Thailand, a stark contrast to the mere two such centers currently in operation.
**The Proposed Remote Access Security Act: Addressing the Loopholes**
Michelle Nie, a visiting fellow specializing in technology and national security at the Center for a New American Security, believes this loophole poses a “threat to U.S. national security.” She articulated, “The point of chip export controls is to deny China the ability to train frontier AI using advanced U.S. chips.”
A legislative proposal, the Remote Access Security Act (RASA), aims to broaden the scope of U.S. export controls to encompass remote, cloud-based access to critical hardware and software. This bill successfully passed the House of Representatives in January but is still awaiting consideration in the Senate. Nie anticipates potential resistance from the industry, noting that “Cloud providers would bear the compliance burden of any KYC and customer verification requirements mandated by the bill.”
Furthermore, Nie cautioned that the passage of RASA alone would not fully resolve the issue. While it would grant the U.S. government “the authority to regulate remote access,” the administration would still need to develop specific regulations for the export control of remote access to advanced chips. King suggested that the Bureau of Industry and Security (BIS) could potentially expedite such a rule, possibly “in a matter of days” with backing from the White House. “The challenge,” King added, “will be in making a rule that’s effective and enforceable. Policymakers will need to decide what compute is covered, who should be prohibited from remotely accessing the compute, and how to implement a robust know-your-customer scheme.”
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/25005.html
