A coolant distribution unit, or CDU, is the equipment that stands between a building's water and the clean loop reaching the servers. In a few years these have grown from a few hundred kilowatts to multi-megawatt modular plants. Two kinds of company sell them: liquid-cooling specialists and broad infrastructure vendors.
Specialist suppliers
CoolIT, LiquidStack, Motivair, Accelsius, Chilldyne, and ZutaCore bring architecture-specific products. Their differentiation often sits in cold-plate integration, controls, pressure strategy, or phase-change design.
Infrastructure suppliers
Vertiv, Schneider Electric, nVent, STULZ, Rittal, Modine, and Delta can connect CDUs to broader rack, power, plant, and service portfolios.
Normalize the bids
Require net capacity at common supply and return temperatures, one-failure redundancy, total pump power, approach temperature, water pressure, fluid quality, controls points, certification, and footprint.
Method and inclusion criteria
This page includes companies with a publicly identifiable coolant distribution unit for data center liquid cooling or an integrated product line that performs that function. General pump, heat-exchanger, or building-control suppliers are not included unless they package and support the unit as a system. A reference design without a commercial product is labeled separately in the wider database. Inclusion requires a company and product source, not a minimum advertised megawatt figure.
- Current rack, row, or facility-scale distribution product
- Public specifications or product evidence
- A defined interface between facility and technology cooling loops
- Commercial sales or project engagement route
This is a landscape, not a capacity ranking
Advertised capacity is not comparable until temperatures, flow, fluid, pump state, and redundancy match. A smaller unit at a warm duty point may deliver more useful capacity than a larger nameplate tested with colder water. The page groups suppliers by scope and buyer fit. It does not award a numerical rank or infer quality from product size.
Choose rack, row, or plant scale
Rack units limit a fault to one cabinet and can enter halls without broad pipework, but they multiply assets and maintenance. Row units serve a pod and match phased deployment. Multi-megawatt platforms reduce unit count and can improve factory standardization, while increasing the shared failure domain. Select the format from deployment phases, plant routes, and acceptable blast radius before comparing vendors.
Fix the duty point
Issue primary supply and return temperatures, secondary supply and return, design flow, fluid, pressure, and worst-day drift. Require net capacity and pumping power at that point, both normally and with the agreed component unavailable. Ask for curves across the operating range rather than one number. The heat-exchanger approach connects every rating to facility performance and should be stated.
Review hydraulics and materials
The unit must hold flow and pressure across manifolds, hoses, quick disconnects, and narrow cold-plate channels without exceeding server limits. Request the pump curve, control range, expansion strategy, filters, particle limits, chemistry, wetted materials, and maximum pressure events. Match these with the server vendor's requirements in one signed compatibility schedule.
Test controls and failure behavior
Specify protocols, points, alarms, local overrides, setpoint ownership, and behavior on lost communication. Factory and site tests should cover pump, controller, power, sensor, valve, filter, and facility-water failures. Verify leak isolation and the signal that tells computing to throttle or shut down. Redundant labels are less useful than a drawing and witnessed sequence.
Assess manufacturing and service
Ask where the quoted model is assembled, what is actually in production, current lead time, test capacity, and which long-lead components are shared across units. Confirm local commissioning staff, response time, remote support, pump and seal spares, software support, and replacement compatibility over the server life. Factory area and aggregate orders are demand signals, not proof that a unit can ship on the project date.
Evidence and award
Prefer named sites operating at comparable scale and temperatures. Use pilots and anonymous orders for the narrower facts they support. Score compliance, evidence quality, delivery, service, lifecycle cost, and commercial ownership after mandatory thermal and warranty gates. Tie award payments to factory testing, site capacity, fluid cleanliness, controls integration, and failure-state acceptance.