China is finding ways around U.S. restrictions on advanced artificial intelligence chips by using Southeast Asian data centers as intermediaries. Rather than importing restricted Nvidia chips directly, Chinese companies are renting computing power or installing server cabinets in Singapore, Malaysia, and other countries to train their AI models overseas, then importing the finished models back into China. This approach sidesteps export controls that focus on hardware shipments but don't restrict the movement of AI data or trained models. Why Are U.S. Chip Export Controls Failing to Stop China? The fundamental problem with current U.S. export controls is their narrow scope. Gary Wojtaszek, a director at GDS Holdings, a major data center operator in China, explained the loophole clearly: "The US chip rules are pretty nebulous. In China, you can't import chips, but you can export your data to train your model. Then you import it back." He added that "water always finds the level playing field. People may try to control or slow a trend, but it will eventually find a natural balance". The U.S. government has implemented multiple layers of restrictions over the past few years. In October 2022, the Biden administration banned exports of Nvidia's H100 and A100 chips to China. This was expanded in October 2023 to include the H800 and A800 chips. In early 2025, the Biden administration introduced the AI Diffusion Rule, which created a three-tier framework categorizing countries by their access to U.S. AI technologies. However, the Trump administration withdrew the AI Diffusion Rule after concluding it was too broad and difficult to enforce. Now Washington allows exports of Nvidia's H200 chips to China, but not the most advanced Blackwell graphics processing units (GPUs). Despite these restrictions, Chinese firms have found a practical workaround that doesn't technically violate any law. How Are Chinese Companies Accessing Restricted AI Chips? - Southeast Asian Data Centers: Major Chinese companies including Alibaba and ByteDance have been training AI models in Southeast Asian data centers to gain access to Nvidia's high-end chips, according to reporting from the Financial Times. - Renting Computing Power: Chinese firms can rent computing capacity from data center operators in Singapore, Malaysia, and other Tier 2 countries that face tighter controls but still have access to advanced chips. - Installing Server Cabinets: The Wall Street Journal reported that four Chinese engineers flew from Beijing to Kuala Lumpur carrying hard drives loaded with AI training data and rented about 300 Nvidia AI servers at a Malaysian data center to process the data. - Exporting Data, Not Hardware: Chinese companies export their raw data to overseas locations for training, then import the finished AI models back into China, avoiding restrictions that target chip shipments. This strategy has made Singapore particularly attractive as an AI hub for Chinese firms. Nvidia's revenue surged 65 percent year-on-year to $215.9 billion for the financial year ended January 25, 2026, with about one-fifth of the company's revenue coming from Singapore, making it Nvidia's second-largest market after the United States. What Does This Mean for the U.S.-China AI Race? The implications are significant for American national security and geopolitical dominance. Wojtaszek noted that the global AI competition has been dominated by the United States and China, with the U.S. holding advantages in cutting-edge graphics processing units (GPUs) and large language model (LLM) algorithms. However, China is catching up quickly and holds advantages in energy production needed to power AI infrastructure and far larger data pools for AI training. China's DeepSeek AI model exemplifies this progress. Wojtaszek observed that "DeepSeek is ingenious. They did something interesting at significantly lower cost and with lower power," adding that China will continue to reinvent other people's inventions and make improvements to them. The stakes are particularly high given recent developments in military AI applications. The Iran conflict has demonstrated AI's transformative power in warfare. According to U.S. Central Command, America's military is using AI to support the initial screening of incoming intelligence data, with Anthropic's Claude identifying and prioritizing key bombing targets in Iran. AI-powered bombing is now considered quicker than the "speed of thought" by some experts, shortening the "kill chain" from target identification to legal approval to strike launch. This military advantage could be undermined if China gains access to the same advanced semiconductor technology. Security experts warn that the federal government must deny adversaries, particularly the Chinese Communist Party, the capacity to turn advanced semiconductor chips into military weaponry. Anthropic recently accused China's DeepSeek, Moonshot AI, and MiniMax of using roughly 24,000 fraudulent accounts to siphon American technology through a coordinated "distillation" campaign intended to extract high-value model outputs for Chinese use, military or otherwise. What New Threats Are Emerging? A new layer of complexity has emerged with proposed legislation. On January 14, the U.S. House of Representatives passed the Remote Access Security Act, a bipartisan bill that modernizes the Export Control Reform Act by expanding federal authority to restrict foreign adversaries' ability to access sensitive technologies remotely through cloud computing services. If this act becomes law, using U.S. AI chips or computing services via overseas data centers could fall under the same export-control rules that govern hardware shipments, potentially closing the loophole that Chinese firms currently exploit. However, enforcement remains challenging. Wojtaszek emphasized that enforcing global chip controls is inherently difficult because hardware can easily change hands across borders. He predicted that China would ultimately overcome such restrictions either through domestic technological development or by sourcing computing resources elsewhere. The bottom line is that current U.S. export controls, while well-intentioned, have significant gaps that allow determined actors to work around them. As the AI race intensifies and military applications become more critical, policymakers face mounting pressure to close these loopholes before they fundamentally shift the balance of technological power between the United States and China.
