Megaspeed and the battle to get high-end Nvidia chips into China - Deutsche Welle

"Megaspeed" is the generational upgrade of the grey-market transshipment model that flourished during the 2018 Huawei sanctions era, now targeting spec-level downgrading rather than geographic routing. The technical implication is stark: if intermediaries can re-spec H-series silicon below the 4,800-TOPS threshold, the US architecture of single-parameter export controls is structurally porous. Nvidia faces a paradox — China once represented roughly a quarter of its data-center GPU revenue; direct sales are now illegal, yet grey channels keep that demand cycle alive. Huawei's Ascend 910B and Cambricon's Siyuan 590 are squeezed from both sides: CUDA ecosystem lock-in persists, and grey-market flow continues to extend the lifecycle of Nvidia hardware in Chinese data centers. Over the next 12–24 months, expect BIS to push tracking granularity down to the chiplet-packaging level, while Chinese hyperscalers accelerate software-compatibility layers built on domestic silicon. The Megaspeed phenomenon is not a loophole to be patched; it is a stress test proving that when a value chain spans 15+ jurisdictions, component-level export controls cannot achieve engineering closure. The real question is no longer "can chips be smuggled?" but "can the control framework survive its own enforcement costs?"