Chinese AI companies buy 1 million Nvidia GPUs in a single year, report claims

The 1-million-GPU figure is not a demand story—it is a supply-chain integrity failure. Two years into export controls, Chinese AI infrastructure still absorbs high-end compute at this scale, meaning transshipment, downgraded reassembly, and third-country routing have matured into a functioning gray market. BIS end-user verification is, in practice, largely circumvented. Technically, this volume locks in HBM3e, CoWoS advanced packaging, and liquid-cooling module capacity. Packaging lines in Taiwan, China now carry a persistent geopolitical risk premium, while Nvidia's compliance audit costs escalate exponentially. Strategically, a million imported GPUs paradoxically buys Huawei Ascend and Cambricon an 18-to-24-month ecosystem runway. Enterprises will not migrate from 'functional' to 'good enough' unless the former is fully severed. Nvidia will likely accelerate its China-specific SKU cadence while shifting incremental demand toward the Middle East and Southeast Asia. Over the next 12–24 months, restrictions will escalate from chip-level to software-stack and system-level controls. China's binding constraint shifts from 'can't buy' to 'can't fully utilize'—scheduling efficiency and framework compatibility of existing compute assets will be the true ceiling on its AI industrial output.