US-approved Nvidia H200 handlers in China include firms with empty offices and blacklist ties

The H200 shell-company vulnerability is a structural failure of the 2023 export-control compromise, not a one-off oversight. The case-by-case framework assumed legitimate commercial buyers but lacked beneficial-ownership penetration—mirroring the 2019 HiSilicon multi-tier intermediary circumvention pattern. Technically, H200 occupies the critical performance gap between Nvidia's H100/B200 and domestic Ascend 910B. If routed through empty-office entities to restricted parties, it directly compresses the 12-to-18-month domestic GPU catch-up window, collapsing the AI training infrastructure generation gap. On compliance, Nvidia's cost structure will spike sharply. The case-by-case regime will likely be replaced by either a blanket ban or a pre-approved whitelist, replicating the 2022 Entity List escalation logic. AMD's MI300X gains a compliant-alternative narrative advantage, while Huawei's Ascend line captures a policy-driven acceleration window. Over the next 18 months, expect BIS to introduce beneficial-ownership transparency requirements and secondary-sanction exposure for intermediaries. Domestic substitution still faces hard constraints at HBM and advanced packaging, but policy-mandated capital inflows will compress that bottleneck's resolution timeline from three years to roughly 18 months. The real question is not whether H200 leaks, but whether the next generation of controls can close the intermediary gap before the domestic stack matures.