With rack densities soaring from 20 kW to over 140 kW, and the 1 MW milestone looming, the industry is transitioning from speculative planning to an urgent, high-stakes deployment phase. Direct liquid cooling (DLC) has become the IT industry’s preferred cooling method for extreme chip power densities.

A DLC system must be specified based on liquid-cooled IT requirements, but also considering your specific data center conditions. The combination of liquid-cooled IT with traditional heat rejection systems does present challenges. The first step, critical for the successful implementation of DLC systems, is specification. In this article, we discuss common challenges operators may face.

Specification challenges of DLC systems

1) Material incompatibility between the CDU and connected components increases the risk of galvanic corrosion, which can create debris that clogs cold plates and damages servers.

When two or more metals with different electrochemical potentials are connected in the presence of a fluid, it sets off a process of galvanic corrosion. These metals become incompatible when the difference becomes large enough to cause issues such as material degradation and redeposition.

Operators must minimize the difference in the anodic index between the materials used for all wetted surfaces. Strictly follow IT manufacturer guidelines for compatible materials and refer to acceptable metals listed by organizations like Open Compute Project (OCP) or American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE).

Keep a comprehensive registry of all materials used, avoid materials like aluminum unless using proper inhibitors, and consider using separate cooling loops if cold plate materials vary across different IT vendors.

2) Opposing liquid and air-cooling requirements force a tradeoff between operating energy savings and the capital costs of chilled water plants.

When you cool your IT equipment with water, you can set your chilled water temperature higher compared to air. For every 0.6°C (1°F) increase in chilled water temperature, the chiller efficiency increases one to two percent. For variable speed chillers, the efficiency increases two to four percent. And this doesn’t include the potential energy savings from economizer hours when the chiller is off or at partial load, which can be significant in cooler climates.

If you want to maximize the efficiency savings from increased chilled water temperatures, you will incur capital costs for a separate chiller plant. This leaves you with two basic choices:

• Use the same “low-temperature” chiller for both loads and forego the energy savings offered by liquid-cooled IT.
• Invest in a separate “high-temperature” chiller plant dedicated to liquid cooled IT loads (higher chilled water temperature) and use the existing chiller for air-cooled IT loads.

The best path forward is to perform a total cost of ownership (TCO) analysis over a 10-year period or more to decide whether to share an existing low-temperature chiller or build a dedicated high-temperature chiller. Because the data center's climate heavily influences this, hotter climates generally justify the investment in a dedicated high-temperature chiller due to increased economizer hours, while colder climates favor a shared low-temperature chiller.

3) Direct coupling between servers and DLC infrastructure complicates specification because the cooling pipes are physically connected to the IT equipment, introducing new variables related to space, leak risks, and specific coolant flow needs.

With air cooling, there might be air ducted to a single rack of servers using rack air containment, but no pipes or ducts direct air to or from an individual server, meaning air-cooled servers and the air conditioning units are only loosely coupled.

In a DLC system, the coolant distribution infrastructure is physically connected to the IT equipment, creating challenges concerning distribution piping, liquid used as the cooling fluid as opposed to air, and the TCS, which must be designed to meet individual server cooling needs.

Distribution piping is necessary for each individual server to connect the CDU to the

server’s cold plates. Collaborate closely with IT and infrastructure vendors to specify compatible quick-connect couplings and dedicate a single cooling loop to a single server model to avoid uneven cooling.

While liquid cooling is more effective, it introduces the risk of devastating leaks. To address those risks, include direct and indirect leak detection systems (such as cable detectors and turbidity sensors) monitored by a centralized control system.

Liquid-cooled server components depend on centralized CDU pumps and controls for cooling, requiring specifically calibrated TCS designs for each server. To manage flow resistance, use constant differential pressure control with energy valves to stabilize pressure across the racks.

4) A lack of CDU system efficiency standards makes it very difficult to compare the capacity and efficiency ratings of different CDUs.

Variables including fluid types, flow rates, fluid temperatures, and ambient temperature influence a CDU’s efficiency or capacity rating, but without a standard, vendors can set their own values for these variables, allowing them to claim the best performance under different operating conditions. As a result, it’s hard to compare two or more CDUs and determine which is best for your design.

The standards exist for air cooling, but not liquid cooling. Until official ASHRAE testing standards are finalized for liquid cooling, request vendors to provide performance ratings based on draft test points. From a design standpoint, limit CDU capacities to support no more than 10-20 percent of your liquid-cooled servers per loop, minimizing the blast radius of a potential disruption should a failure occur.

5) Provisioning IT space for unknown liquid-cooled IT creates the risk of overpaying for oversized piping or stranding cooling capacity, as the system must be prepared for unknown maximum and minimum rack densities.

Data centers may need to provision IT space to support liquid-cooled IT without knowing what specific liquid-cooled IT equipment will show up, a common issue for colocation providers preparing for new tenants. It hasn’t been a problem with air cooling, but DLC directly couples the coolant supply to every single server, which means every CDU, and its TCS loop, needs to be capable of supporting a maximum quantity of racks at the lowest density as well as the minimum number of racks at the highest density. The greater the range, the greater chance of stranding some piping and valves by deploying racks at higher densities.

To mitigate these variations, size the main loop piping diameter for the bulk CDU capacity, ensuring the piping is long enough to accommodate the maximum number of racks (lowest density). However, size the branch piping and manifolds for the highest potential rack density. You can also specify different power densities for distinct areas of the data center or develop reference designs for two or three DLC “pods” at different densities that can be finalized once specific IT equipment is confirmed.

Where the liquid meets the chip

Moving from a conceptual pilot to a robust, production-ready liquid cooling environment begins with managing specification challenges before moving into the installation and operations stages.

With a disciplined approach to specification and a partner like Schneider Electric, you can mitigate the inherent risks of high-density cooling. The transition to liquid is complex, but with the right technical framework, it becomes the foundation for sustainable, high-performance AI growth.

Want to learn more? Read our white paper: Direct Liquid Cooling System Challenges in Data Centers.