Melissa Griffith

Semiconductors have become a central battleground in contemporary geopolitics, with export controls that target strategic adversaries' artificial intelligence efforts (most notably China's) generating enormous public, media, and policy attention. In the last two years, growing concern has centered on the possibility that states and companies have continued to access controlled compute through diversion and export control evasion. Yet, the evidentiary record underpinning these debates remains surprisingly thin.

What types of semiconductor diversion are occurring and at what scale? Setting aside questions over the utility and legitimacy of export controls, where are these chips appearing and when do this transnational chip flows carry the geopolitical weight many in the policy community assume they do. Are restricted chips reaching adversaries in volumes and configurations that meaningfully shift strategic balances? Or are we, in some cases, conflating tactical enforcement problems with strategic threats?  

To date, the dominant framing of evasion and diversion centered heavily on GPUs and the particular realities of AI compute. That concern is real, but that focus risks obscuring cases where the threat picture looks entirely different. For example, the diversion of field-programmable gate arrays (FPGAs) to Iranian drone programs requires far smaller quantities, operates outside the scale and logic of data center buildout, and poses security risks that are tactically concrete rather than strategically diffuse.

This talk asks: what are the national security stakes of semiconductor diversion, and when do transnational chip flows constitute genuine strategic threats and when do they not?