Foxconn unveils Vera Rubin AI racks, liquid-cooling systems at OCP Summit

Foxconn's OCP showcase is not another server launch — it is the completion of a structural shift from component OEM to AI infrastructure system integrator. Vera Rubin NVL72's thermal density has broken the air-cooling ceiling; liquid cooling is now a physical prerequisite, not a differentiator. FIT's simultaneous positioning across interconnect, power delivery, and thermal management signals that margin is migrating from the chip layer to the system-architecture layer — the same dynamic that shifted NVMe controller value to OEMs in 2020. Risk calculus: staging the demo in San Jose is an explicit US-market signal, yet the manufacturing base remains anchored in Taiwan, China and mainland China. If high-speed connectors or optical modules get added to export-control lists, FIT's supply chain becomes a single point of failure. The 2022 Entity List episode proved component-level restrictions can collapse rack-level delivery within weeks. Competitive landscape: Supermicro's CoolIT alliance and Quanta's Dell channel are direct threats, but the real adversary is Nvidia's own MGX reference design, which aims to commoditize the rack. Foxconn's moat lies in system-level IP — thermal modeling, power sequencing, interconnect topology — far harder to replicate than any single component. 12–24 month call: CPO will displace discrete transceivers in NVL72-class systems; liquid cooling will converge on DTC; the rack, not the GPU, becomes the unit of commerce. Whoever owns system-level integration owns AI infrastructure pricing power.