Qisda's Aewin targets 200kW liquid-cooled AI servers

200kW per rack is a structural break in data center physics, not a spec bump. Once thermal density exceeds air-cooling's ceiling, two-phase DLC becomes the only viable path. Aewin's pairing of Arivor's phase-change refrigerant with AMD EPYC 9006 (Zen 5, 192 cores) is a deliberate CPU-centric AI inference stack, countering NVIDIA's GPU-cluster monopoly. Ripple effects extend far beyond thermals. At 200kW, PDUs, busbars, fiber routing, and rack structures demand ground-up redesign—ODMs shift from assemblers to system architects. Arivor as a single-source coolant provider creates lock-in reminiscent of Intel's 2019 Foveros internalization squeezing OSATs. Supply-chain risk is acute: a three-node chain (Taiwan, China: Qisda; US: AMD; Arivor's coolant patents) means any export-control escalation halts production. Tighter entity-list rules on EPYC 9006 zero out deployment in affected markets overnight. Supermicro and Dell have shipped 100kW liquid-cooled racks, but 200kW remains hyperscaler-only (Microsoft, Meta, AWS). Aewin's OCP positioning is less a revenue play than a standards-capture maneuver to set the architectural ceiling before enterprise demand matures. 12–24-month outlook: two-phase DLC defaults for racks above 80kW; ODMs capture 30%+ of AI server BOM value (from ~15%); AMD's data-center CPU share crosses 25%; rack-level procurement replaces per-server purchasing as the dominant commercial model.