Chinese technology giants like Alibaba and ByteDance have adopted a new strategy to circumvent US export restrictions on advanced Nvidia GPUs critical for AI training. Due to US controls banning the sale of Nvidia’s most powerful chips to China, these firms are increasingly leasing data center capacity in Southeast Asia, such as in Singapore and Malaysia, to remotely train their large AI models using Nvidia hardware. This approach enables them to access cutting-edge Nvidia H100 GPUs, which remain unavailable domestically because of export curbs and China’s own bans on foreign AI chips in state-funded centers.
While China has made progress in developing its own AI chips, these domestic alternatives mainly handle inference tasks and lag behind Nvidia’s GPUs in training performance, which requires extensive computational power. By training AI models offshore, Chinese companies maintain competitiveness in developing large language models, AI services, and cloud computing offerings—and continue growing their global market share.
This move illustrates the complex interplay between US export policies, China’s domestic chip ecosystem ambitions, and multinational cloud infrastructure dynamics. The ability of Chinese firms to lease Nvidia-powered servers abroad effectively sidesteps US restrictions without violating them directly. It highlights the challenges of regulating critical AI hardware in a globally interconnected tech environment and the lengths to which companies will go to secure access to leading-edge technology crucial for AI innovation.
