Who makes ems, system integration & rack-level test for AI data centers

EMS providers sit downstream of the board and tray builders (4.03, 4.04): they take built boards, populate chassis, route cabling, install the liquid-cooling loop hardware from layer 5, and run rack-level test and burn-in before a system leaves the factory. Several of the same companies that design and build the boards also run their own integration lines, while pure-play EMS firms take on integration and test work for OEMs that don't want to run factories themselves. AI racks make this step harder and more valuable: a liquid-cooled, multi-tray cabinet with a large number of internal cables and connectors has far more failure points to catch before shipment than a conventional 1U server, and a leak or a bad connection discovered in the field is expensive to fix. That raises the bar for test equipment and floor space, and it means integration capacity — not just chip supply — can gate how fast a rack generation ships. For investors, the signal to watch is which integrators win capacity commitments tied to a new rack generation, since qualification and yield on rack-level test tend to lag the underlying silicon ramp.