Chinese AI Firms Scramble for Compute Abroad as US Export Controls Tighten

Chinese artificial intelligence companies are pursuing a new strategy to circumvent U.S. export controls: renting computing power in third countries rather than purchasing hardware directly. According to sources familiar with the matter, firms are targeting data centers in Southeast Asia and the Middle East to gain access to Nvidia's upcoming Rubin architecture GPUs, which remain restricted for direct sale to China.

The move represents a significant shift in how Chinese AI firms are navigating the increasingly complex landscape of U.S. semiconductor restrictions. Rather than attempting to acquire restricted hardware through gray-market channels—a practice that has drawn increased scrutiny from U.S. authorities—companies are instead seeking to rent compute time from international data centers that can legally purchase Nvidia's latest chips.

This approach allows Chinese AI firms to access the computational power needed for training large language models and other advanced AI systems without violating export regulations. The strategy is particularly crucial for companies like Zhipu AI and Alibaba's cloud division, which have publicly warned that the technological gap between Chinese and U.S. AI capabilities is widening due to restricted access to cutting-edge hardware.

The urgency stems from the impending release of Nvidia's Rubin architecture, which represents the next major leap in GPU performance for AI workloads. While the H100 and H200 chips have been the workhorses of AI development in recent years, Rubin promises significant improvements in efficiency and processing power. Chinese firms fear that without access to this technology, they will fall further behind U.S. competitors who are already securing their allocations.

U.S. companies like OpenAI, Google, and Meta have been first in line for Nvidia's latest hardware, with the chipmaker prioritizing orders from its largest and most strategic customers. This has created a two-tier system where American AI firms can access the most advanced chips while Chinese companies face increasingly stringent restrictions.

The rental strategy introduces new complexities. Data centers in Southeast Asia and the Middle East must carefully navigate their own regulatory environments while ensuring they don't violate U.S. secondary sanctions. Some countries in these regions have already faced pressure from the U.S. to restrict Chinese access to advanced computing resources.

For Chinese AI companies, the stakes are particularly high. Zhipu AI, one of China's leading foundation model developers, has been vocal about the challenges of competing without access to the same hardware as U.S. rivals. The company has developed impressive models despite these constraints, but executives have warned that the gap could become insurmountable without access to next-generation chips.

Alibaba's cloud division faces similar challenges. As China's largest cloud provider, it needs to offer competitive AI services to its enterprise customers, but the hardware restrictions limit what it can deliver. The company has been investing heavily in developing its own AI chips, but these efforts are still years behind Nvidia's technology.

The rental market itself is emerging as a new business opportunity. Data centers in regions with more favorable regulatory environments are positioning themselves as compute hubs for international clients. Singapore, Malaysia, and the United Arab Emirates are among the locations being considered, though each presents its own regulatory challenges.

This development also highlights the evolving nature of technology export controls. Traditional restrictions focused on physical hardware sales, but the shift toward service-based access (renting compute time) creates new enforcement challenges. U.S. authorities may need to develop new frameworks to address this loophole.

The implications extend beyond immediate business concerns. If Chinese AI firms can successfully maintain access to cutting-edge compute through international data centers, it could undermine the effectiveness of U.S. export controls. Conversely, if these efforts are blocked, it could accelerate China's push toward semiconductor self-sufficiency and alternative AI hardware architectures.

The situation remains fluid as both companies and regulators adapt to the changing landscape. Chinese AI firms are likely to continue exploring creative solutions to access advanced computing resources, while U.S. authorities will be watching closely to ensure export controls remain effective.