The U.S.-banned NVIDIA top-tier B300 chips, which China is prohibited from importing, were reportedly leased by Moonshot AI as early as August this year. According to the latest report by Bloomberg, documents submitted to Beijing regulators reveal that Moonshot AI, the company behind Kimi, signed a contract in August to lease 32 Supermicro servers equipped with NVIDIA B300 GPUs. The B300 belongs to NVIDIA’s most advanced Blackwell series and is subject to U.S. export controls on China. It is important to note that the 32 refers to the number of servers, not individual GPUs; the disclosed contract details are insufficient to confirm whether the equipment has been physically delivered or deployed. Interestingly, this is not the first time Moonshot AI has been exposed for seeking access to restricted NVIDIA computing power. In July, U.S. White House officials accused Moonshot AI of using Blackwell chips through a data center in Thailand. Just last week, CNBC reported that Moonshot AI had discussed leasing a cluster of at least 8,000 Blackwell GPUs with European cloud providers, primarily for model inference but also involving training capacity needs. The European arrangement remains under negotiation and cannot yet be considered finalized. When viewed together, these reports reveal an increasingly clear reality: the U.S. is striving to restrict the flow of advanced chips, while Chinese AI companies are actively seeking access to high-performance computing resources. These two objectives are not entirely equivalent. While chips can be banned from export, computing power can be reorganized through server leasing, overseas cloud services, and third-party infrastructure into a commercial offering. Of course, this does not imply all such transactions are legal, nor that regulators cannot further close these loopholes. Meanwhile, Beijing is pushing for domestic alternatives in computing power, yet leading AI model companies continue to seek NVIDIA’s most advanced GPUs globally. This underscores that, at least in the race for frontier models, replacement of foreign computing power remains far from complete. The U.S. aims to control the spread of AI capabilities by restricting chip supply—but what truly needs to be controlled may no longer be chips alone, but the entire global computing services market. As hardware increasingly becomes a service, export controls are no longer faced with just a single customs checkpoint.