This webinar examines how rising rack power densities, driven primarily by AI and accelerated computing workloads, are fundamentally changing data center cooling strategies. The discussion contrasts traditional air-cooled architectures with emerging liquid cooling approaches and outlined how Munters-supported systems are applied across these evolving thermal regimes.

The webinar began by establishing the baseline “as-is” cooling configuration still prevalent in many data centers today. In this model:

  • Servers rely on internal fan-driven airflow
  • Cool air is supplied through cold aisles and drawn through server racks
  • Heated exhaust air is contained in hot aisles
  • Air returns to CRAH/CRAC fan arrays
  • Heat is transferred to a facility chilled water loop
  • Chillers reject heat to ambient air, typically via air-cooled equipment

This approach has historically supported rack densities in the 10–20 kW per rack range, with incremental improvements enabling higher densities through increased airflow, higher ΔT, and larger fans.

The webinar highlighted the physical and operational constraints that limit air cooling scalability:

  • Above roughly 40 kW per rack, airflow requirements increase rapidly
  • Aisle air velocities above ~600 feet per minute can cause airflow instability and rack “starvation”
  • Increasing corridor width, fan power, and temperature differentials yields diminishing returns
  • Higher fan energy significantly degrades overall efficiency metrics
  • Reliability risks increase as airflow margins shrink

While air cooling can be extended to approximately 60–80 kW, and in some cases near 100 kW per rack, these designs are generally inefficient and difficult to operate consistently. Compounding the issue, chip manufacturers (like NVIDIA) are mandating liquid cooling for next-generation GPUs and accelerators. This effectively removes air cooling as an option for many future high-density deployments.

The presenter next turned to the most widely adopted high-density cooling approach: single-phase direct-to-chip liquid cooling. Key characteristics include:

  • Cold plates mounted directly on CPUs and GPUs
  • Liquid coolant absorbs heat via conduction
  • Typical temperature rise across cold plates of 18–20°F (with observed ranges from ~15°F to 27°F depending on flow rates)
  • Heat rejection via Coolant Distribution Units (CDUs)

Approximately 80% of rack heat load is removed via liquid cooling, with the remaining ~20% handled by air cooling for components such as memory, power electronics, and network hardware. This hybrid approach allows existing air-cooled infrastructure (CRAHs/CRACs) to remain relevant, even as rack densities increase substantially.

A critical distinction emphasized in the webinar is the separation between facility water loops and technology cooling system (TCS) loops:

  • TCS loops are closed, localized systems serving racks and CDUs
  • Typical coolant is ~25% propylene glycol / water
  • Filtration levels commonly range from 25–50 microns, with 25 microns becoming standard
  • The loop is designed for precise control of:
    • Pressure
    • Flow rate
    • Supply temperature

Using a TCS loop rather than facility chilled water directly reduces leak risk, limits the “blast radius” of potential failures, and allows faster, more precise thermal control. Large facility water loops have higher thermal inertia and slower response times, making them poorly suited for direct rack-level cooling.

The presenter then introduced a Munters-specific alternative to traditional chilled water systems: refrigerant-based cooling architectures integrated with liquid-cooled racks.

In this design:

  • The TCS loop still delivers coolant to cold plates
  • Instead of rejecting heat to a chilled water loop, CDUs interface with refrigerant-based evaporators
  • Heat is rejected via one-to-one refrigerant piping to rooftop condensers
  • Refrigerant lines are low-pressure and gravity-fed, reducing installation complexity

Several technical benefits of refrigerant-based systems were highlighted:

  1. Tighter approach temperatures
    Refrigerant systems enable closer approach to ambient conditions, improving free cooling performance.
  2. Expanded free cooling viability
    Free cooling becomes feasible in climates previously considered marginal (e.g., Dallas, Atlanta, Kansas City), not just cold-weather regions.
  3. Reduced infrastructure expansion
    For retrofits or mid-build pivots, refrigerant systems avoid the need to:

    • Upsize existing chilled water plants
    • Add a second facility water loop

This makes refrigerant-based systems particularly attractive for incremental capacity expansions, such as adding an additional 10 MW of high-density load to an existing data center.

As an interim or constrained option, the webinar discussed rear-door heat exchangers:

  • Liquid-cooled servers reject heat to an air-to-liquid heat exchanger mounted on the rack
  • No full TCS loop is required
  • Air temperature rise across the rack increases substantially

This approach reduces infrastructure requirements but is limited in achievable density. It is not expected to support 250–300 kW per rack and is generally positioned as a transitional solution rather than a long-term strategy.

Two-Phase Direct-to-Chip Cooling

The session briefly addressed two-phase direct-to-chip cooling, an emerging technology using dielectric refrigerants:

  • Refrigerant evaporates at the cold plate, absorbing latent heat
  • Vapor is condensed remotely and returned to the rack
  • No water is used anywhere in the system

Key technical advantages include:

  • Dielectric fluid eliminates catastrophic damage risk from leaks
  • Superior thermal performance due to phase change
  • Completely waterless cooling architecture

However, market adoption remains limited due to system complexity and industry conservatism, similar to early resistance seen with single-phase liquid cooling.

Munters’ scope focuses on heat rejection and distribution infrastructure, not server internals. This includes:

  • Water-to-water and water-to-refrigerant CDUs
  • CRAHs and fan arrays
  • Air-cooled and water-cooled chillers
  • Dry coolers
  • Rooftop condensers
  • Refrigerant-based economization systems

Systems are broadly grouped into:

  1. Chilled water solutions – optimized for high ambient conditions and peak power reduction
  2. Refrigerant-based systems – optimized for free cooling and reduced infrastructure
  3. Evaporative systems – historically significant but declining due to water consumption concerns

Evaporative cooling remains highly energy efficient but is seeing reduced adoption due to sustainability and water availability constraints.