Explainer-Tougher curbs on AI chip sales to China likely after Nvidia CEO resigns - The Mighty 790 KFGO

Nvidia's CEO transition isn't a personnel shuffle—it collides with the precise moment export controls are escalating from chip-level to architecture-level restrictions. On the technical stack, H200's 4nm process is locked to TSMC (Taiwan, China) CoWoS packaging, while HBM3E supply remains constrained by entity-list rules. If regulators extend scrutiny to system-level interconnect bandwidth, NVLink itself becomes a compliance target, forcing a protocol redesign and a minimum two-quarter product slip. Compliance costs are compounding. The A100→A800→H20 downgrade-for-market playbook has hit its ceiling. A leadership vacuum stretches the legal-and-government-affairs decision chain; a single misstep now carries penalty exposure in the 5-10% of revenue range, not the single-digit millions of 2022. Competitively, AMD's MI300X is exploiting the governance gap to lock in secondary procurement from Meta and Microsoft. Huawei's Ascend 910B has already crossed 30% penetration in domestic training clusters. Intel's Gaudi 3, despite a clear generational gap, retains appeal in Middle East and Southeast Asian markets through its non-US supply chain narrative. Within 18 months, the AI chip market will bifurcate from one superpower to two parallel tracks. If Nvidia doesn't complete its leadership handoff and secure a compliant path for the Rubin architecture by Q3, its share in the Global South will irreversibly slide below 20%.