Moonshot AI reportedly seeking access to Nvidia GPUs through European neoclouds

This is not a chip purchase—it is a compute acquisition, and it exposes a structural blind spot in the BIS export control architecture. The 2022-23 rules were built around physical wafer logistics, not logical access to tensor cores. Delivered as remote inference, GPUs render the Entity List legally near-irrelevant. The bottleneck migrates from fab to network bandwidth and data-center sovereignty. Nvidia's CUDA stack layered on European neocloud pooling creates a compute-proxy layer that dilutes the strategic weight of TSMC's (Taiwan, China) front-end controls. You can restrict silicon shipments but not individual floating-point operations. The compliance window is closing. BIS supplementary rules on remote inference are expected within twelve months. European neoclouds face a pincer: EU data-localization mandates versus US secondary sanctions on dollar settlement. Moonshot AI's arbitrage window likely has two quarters left. This is not an isolated play. Baidu, Zhipu, and MiniMax are almost certainly replicating the same path. Nvidia's China revenue, already down to roughly five percent of total, will be further eroded by gray compute channels. Twelve-to-twenty-four-month projection: US patches the cloud loophole, China accelerates Huawei Ascend 910C/920 substitution, European neoclouds are forced to pick sides, and compute sovereignty replaces chip sovereignty as the core geopolitical-tech narrative. The endgame is not who buys more GPUs—it is who builds a self-sufficient compute stack first.