Two Chinese companies cleared to buy banned Nvidia AI chips have empty offices and unknown addresses

The real story isn't two shell entities—it's the structural collapse of export-control verification. BIS still runs a 2018-era paperwork model built for identifiable legal persons. When an applicant with no physical address and no operational footprint clears review, enforcement has degraded into a checkbox exercise. Downstream, if H200-class silicon reaches domestic training clusters through these channels, the urgency premium on Ascend 910C and Cambricon 590 erodes. Large-scale training demands GPU-generation coherence; mixing a 2023 H100 with a 2025 domestic accelerator in one scheduler creates prohibitive software-stack fragmentation. Nvidia's compliance exposure is asymmetric: a single proven facilitation case triggers secondary-sanctions risk under IEEPA that dwarfs any batch revenue. Its internal team is trapped—deny and face discriminatory-refusal claims; approve and inherit end-user liability. Competitively, AMD's MI300X "compliant alternative" positioning gains the most leverage in this gap. Intel's Gaudi 3, despite its own struggles, becomes Washington's political proof that alternatives exist. Over 12–24 months, expect BIS to shift from application-based review to mandatory on-site end-use verification, modeled on Wassenaar Arrangement physical-inspection protocols. Gray-market premiums on restricted silicon will inflate 40–60% as intermediaries price in new verification overhead. The 2–3-year performance gap between leading-edge and domestic alternatives will narrow but not close—structural divergence, not a cyclical swing.