EU needs only ‘very trusted partners’ on AI chips, tech chief says amid Nvidia crisis - POLITICO.eu

The EU's "very trusted partners" framing is not diplomatic hedging—it is a de facto market-access gate for AI silicon. On the technical stack, Nvidia's CUDA moat is being eroded by compliance overhead: H200's CoWoS packaging depends on Taiwan, China capacity, and if "trust" is defined by geopolitical alignment rather than performance, TSMC's leading-edge nodes become a risk variable in themselves. Upstream, ASML export controls have already tightened; downstream, European cloud providers must redraw their AI infrastructure roadmaps. Compliance is essentially GDPR projected onto hardware: chips are no longer commodities but data-sovereignty carriers. Nvidia will need a standalone EU compliance entity, adding 15-25% per-unit cost. Strategically, AMD's MI300X will exploit the de-Nvidia narrative for government and enterprise contracts; Intel's Gaudi 3 gains implicit points in Brussels for domestic manufacturing. The deeper signal: the EU is converting its €43B Chips Act into a geopolitical hedge, not merely a capacity play. 12-24 month outlook: expect a Trusted AI Chip certification regime analogous to automotive WVTA but targeting compute. If Nvidia has not restructured its EU compliance architecture by 2026, its European data-center share could slide from 70%+ to below 50%. This is not market selection—it is institutional displacement.