The US let Nvidia sell chips to an empty office in China - The Verge

Core judgment: An "empty office" entity clearing US export approval isn't an enforcement glitch—it's the structural ceiling of a name-based licensing regime. Technical ripple: If H100/A100 units flow through shell entities, China's actual AI compute penetration is systematically undercounted. For TSMC in Taiwan, China, the signal is paradoxically stabilizing—so long as licenses persist, fab utilization holds. But the deeper consequence: the urgency narrative around Huawei Ascend and Cambricon weakens, potentially deferring capex on advanced nodes. Compliance exposure: Entity lists are a naming game; shell companies are its natural predator. Nvidia's per-transaction due diligence cost escalates exponentially when registrants sit in BVI or Cayman. Precedent from ZTE and SMIC cases shows "should-have-caught" penalties can hit 15% of revenue. Market dynamics: AMD will weaponize "supply-chain transparency" as a pricing premium for MI300. Domestic Chinese chipmakers face an awkward window—Nvidia products still arriving undermines the substitution thesis in the eyes of enterprise buyers. 12–24 month outlook: BIS will pivot from entity lists to transaction-behavior audits, demanding end-user telemetry logs. Compliance authority migrates from legal to CTO offices. Nvidia's counter isn't retreat—it's a compliance-audit SaaS layer, socializing regulatory cost onto downstream customers.