Commentary: AI server ODMs face margin squeeze as memory costs soar

The margin squeeze on AI server ODMs reflects a structural shift: value is migrating upstream to component monopolists. Soaring HBM and GPU costs not only erode assembly profits but also force ODMs to invest heavily in advanced thermal, power, and PCB designs—raising system integration barriers and sidelining smaller players. Geopolitical risks amplify supply chain fragility, with U.S. export controls and Korea’s HBM concentration limiting cost hedging. Facing pricing dominance by NVIDIA and AMD, top ODMs like Quanta and Wistron will likely pursue vertical integration—into CoWoS packaging or proprietary liquid cooling—to reclaim bargaining power. Over the next 12–24 months, expect strategic pruning of low-margin orders and a pivot from 'manufacturing for hire' to 'engineering for hire.' Only those mastering heterogeneous integration and thermal management will survive.