Who makes rack manifolds for AI data centers

The manifold sits between the CDU's supply and return lines and the individual trays: it splits one large coolant flow into the branch lines that feed each cold plate's quick disconnects, then recombines the return flow for the trip back to the CDU. Because it touches every tray in the rack, the manifold's port count, spacing, and pressure rating are set by the rack's specific tray layout and GPU count, which ties the part to a reference design in the same way cold plates are. As racks pack more accelerators per U and run hotter, manifolds need more ports, higher flow rates, and tighter tolerances on the quick disconnect interfaces at each branch — a manifold designed for a lower-density generation does not simply carry over. Investors should watch how much of this part rack integrators build in-house for their own designs versus source from independent thermal vendors, and whether manifold makers also supply the CDU and cold plate to capture more of the loop.