AMD's Lisa Su says AI demand is 'absolutely real' as CPUs return to focus

Lisa Su’s insistence on 'absolutely real' AI demand signals a strategic pivot: CPUs are reclaiming centrality in AI infrastructure beyond GPU hype. Technically, this accelerates adoption of chiplet designs and CXL interconnects, forcing co-design across memory, power delivery, and thermal subsystems. Geopolitically, while U.S. and EU subsidies ease capex, tightening export controls compel AMD to rebalance its reliance on TSMC’s advanced nodes with onshore packaging capabilities. Countering NVIDIA’s inference dominance, AMD is betting on MI300X + EPYC synergy to capture enterprise training and edge AI workloads. Over the next 18 months, as models shift from massive pretraining to efficient fine-tuning, CPU roles in heterogeneous compute will gain weight—reshaping server BOM economics and incentivizing Tier-2 cloud providers to actively diversify away from NV-centric stacks.