The data center cooling landscape is evolving, driven by the relentless rise in power density and computing demands. Two-phase cooling has emerged as a highly promising solution and increasingly popular buzzword, being used frequently at industry conferences and discussed in countless whitepapers.

But while the physics holds great promise, real-world implementation still presents significant challenges. This is primarily due to the scale of deployment, integrating far more evaporators that what is usually seen with two-phase loops. Matching the modularity of air cooling seen in data centers today is also incredibly difficult.

In this article, we delve into the science behind two-phase cooling, unpack the critical issues facing its large-scale deployment, and explore how Calyos and the broader thermal community are addressing these barriers together.

Understanding the fundamentals: why 2-phase cooling?

At the heart of two-phase cooling lies a simple but powerful concept: the use of latent heat of vaporization to absorb and transport heat more efficiently than either air or single-phase liquid systems.

To understand the advantage, let’s compare the heat absorption capacities with more traditional methods of cooling in data centers:

• Air (sensible heat): ~1.2 [J/L°C] at room temperature and pressure
• Water (sensible heat): ~4,180 [J/L°C] at room temperature and pressure
• Refrigerant (latent heat): ~180,000 [J/L]at 45°C saturation temperature

Obviously for many years air cooling has been more than sufficient for almost all applications, and with the added performance boost from embedded two-phase systems like traditional heat pipes and vapor chambers, this has been the norm.

The obvious next step was to begin looking at single phase cooling and with many engineers existing experience with water cold plates it felt like the natural step. We see this today with lots of water cooling systems available on the market, but the physics of two-phase cooling remain compelling.

This has led many engineers to develop two-phase cooling systems for racks in order to harness the latent heat of vaporization and today we are seeing this arrive on the market.

This article is striving to look at this objectively, particularly given the large number of papers and presentations released recently that discuss the topic often comparing it against single-phase and highlighting many of the problems adopting such systems.

First, let's breakdown a 2-Phase direct to chip (D2C) system

The fluid (typically a refrigerant) evaporates inside an evaporator at the heat source, absorbing a massive amount of energy without a significant temperature increase. This vapor is then transported to a condenser where it condenses back into liquid dissipating the heat into a secondary fluid (typically the facility water system - FWS).

The thermodynamic performance is described by the classic formula:

Q = h · A · ΔT,

where Q is the power (Watts), h is the heat transfer coefficient [W/m²K], A the surface area [m²] and ΔT the temperature difference [K].

In two-phase systems, the h value is often significantly higher where the phase change (vaporisation and condensation) is occurring, but additional ΔT can occur due to pressure drops in the systems.

Here is some common terminology from OCP used throughout this article:

• Technology Cooling System (TCS):
  This is the cooling system from the CDU to the rack, through the manifold and IT equipment, and back to the manifold and to the CDU.
• Facility Water System (FWS):
  The facility or building cooling system.
• Cold Plates (Evaporator)
  Where the primary heat exchange occurs between the component being cooled and the coolant.
• In Chassis Tubing
  The tubing and connections made between different cold plates within the chassis and the QDs
• QDs
  Quick disconnect couplings are used to quickly disconnect the ITE or its components from the manifold for serviceability.
• Manifold
  Distributes coolant inside the rack from the CDU to the ITE and back again as a liquid or vapor.
• CDU
  Exchanges heat between the heated liquid from the TCS and the FWS.

Boiling modes

Two-phase cooling systems are designed differently regarding the boiling mode which operates in the system:

• In pool boiling, there is no imposed flow. This process forms vapor bubbles that grow, detach, and rise in place, often in a pool of liquid. The vapor typically exits naturally .Pool boiling is better suited to passive systems, like the heat pipes we produce,
• In flow boiling, fluid is actively pushed across a surface. This is by external means, typically a pump. As the fluid flows along the surface a higher percentage of the liquid is vaporised. Flow boiling is better suited to pumped systems.

Flow architecture

Fluid can flow though the evaporators regarding two flow types of architecture:

• In series flow, coolant travels through components one after another, similar to a train moving through stations.
• In a parallel flow, coolant splits across multiple channels simultaneously, similar to traffic on a multi-lane highway.

Feedback from the industry

Despite the advantages, many reputable engineers working in the industry have identified challenges for widespread adoption of two-phase rack cooling systems. Here is the list of publicly available information:

• Controlling Flow Instabilities in Direct-to-Chip  Two-Phase Cooling for High Heat Flux Processors , Ali Heydari et al.
• Thermal Management for Data Centers 2025-2035: Technologies, Markets and Opportunities, Yulin Wang
• An Objective look at 2-phase cooling, Strategic Thermal Labs, Austin M. Shelnutt, P.E, Tyler Duncan.
• Trade-offs between single-phase and two-phase liquid cooling, Reza Najjari

Drawing from our experience we compiled this information identifying six key challenges that we dive into below.

Challenges and potential resolutions

01 – Semi-Vaporization Limits Uniform Cooling

In series configurations the coolant partially vaporizes in each evaporator in the chain, as there must be more fluid available to vaporize in the following evaporator. This means each evaporator operates with a different vapor quality, and therefore a different thermal performance. In turn this leads to different case temperatures.

The most common resolution to this is to move to parallel configurations.

02 – Instabilities in parallel layouts risk dryout

In parallel flow setups, high-power components vaporize more fluid, increasing pressure drop at their respective evaporators. This causes the liquid to divert toward lower-power evaporators, leading to imbalances in the fluid flow and potential dry-out as less fluid is fed to the evaporators that need it most.

To solve this some companies are adding restrictive or regulatory valves to stabilize the fluid flow. While this works, it introduces cost and complexity and often requires a complex control system to manage each device. Alternatively moving from flow to pool boiling can avoids parallel flow instabilities altogether, particularly if you use capillary structures (like we do at Calyos 😉).

03 – High vapor-phase pressure drop

It is commonly known that vapor has significantly higher pressure drop than liquid when flowing down a tube. This increased pressure drop during the transportation of the vapor can cause unfavourable temperature delta's in sections of the system, reducing overall system performance. The vapor quality will directly affect the performance of this.

Therefore most engineers design systems to have the shortest vapor flow routing in their systems. Another alternative is to use use direct-contact condensation (a technique borrowed from nuclear power plants) where the vapor begins to condensate directly inside a liquid. This can be extremely effective, but requires strict design of the manifold.

04 – CDU and QD limitations

Many existing CDUs and Quick Disconnects (QDs) have been designed for single-phase systems and they do not necessarily work optimally in two-phase systems.

• Typically smaller pumps are required as the flow rate is less in 2-phase.
• Accumulators are required to compensate for volumetric changes to the fluid inside the system. This can result in a larger CDU.
• Greases/oils used for o-rings inside QDs are broken down by refrigerants used inside 2-phase systems resulting in failiures.

Companies are beginning to design their own CDUs or with partners who have experience producing HVAC systems using the same or similar fluids. Valeo, Accelsius, and Staubli already making headway on these topics and this is all likely to be resolved soon.

05 – Sustainability and Safety of Fluids

Concerns include high fluid costs, for example ~€30/L for R1233zd(e) vs ~€5/L for PG25. Also some of the fluids have high GWP, ODP, TFA, atmospheric lifetimes. The fluids can also be toxic, and may be subject to ever evolving government regulations.

Modern refrigerants like R1233zd(E) and R515B offer lower GWP, and bring the benefit of being dielectric (protecting expensive hardware in leaks), and have growing support from manufacturers under HVAC mandates. While it is easy to put together a list of many dangerous, unsustainable fluids there are options that are orders of magnitude lower in sustainability scores.

At Calyos we expect these fluids to stick around particularly for HVAC applications (nobody want's to give up Air Conditioning) and therefore there will be fluids available for these two-phase cooling systems. It is our job as suppliers to choose the most environmentally friendly ones.

06 – Simulation Complexity

We all know how difficult two-phase cooling is to simulate, both vaporization (bubble nucleation) and condensation are chaotic and highly complex. This makes it challenging for engineers to objectively compare systems without purchasing one to test.

At Calyos we are creating in-house physics-based models that combine with CFD and FEA/FEM analysis. They are validated against real test data in order to maximise accuracy of the results. Maybe AI can begin to help us solve our problems while we solve its.

Apples and oranges and the problem with comparisons

Many papers attempt direct comparisons between single-phase and two-phase systems. In principle this is a good idea, but it often results in misleading or inaccurate comparisons. The two types of systems are fundamentally very different, while in single-phase systems the fluid temperature increases with fluid flow, whereas in two-phase the fluid maintains a constant temperature and simply changes phase.

Key comparison pitfalls:

• Case-to-fluid resistance always favors two-phase due to higher HTC (vaporisation).
• Fluid-to-FWS inlet resistance often favors single-phase due to lower pressure drop in the liquid lines vs vapor.

Direct one-to-one benchmarks can lead to poor conclusions. Comparing a single-phase CDU to a two-phase one is less valuable for end users. We need to focus solely on case-FWS inlet temperature as a metric for comparison as this factors in the full system.

It would be great to see comparisons of cooling systems on identical ITE but it is almost impossible to run them with the same loads, same ambient conditions, same inlet temperatures and therefore comparing real cooling systems on real hardware is going to continue to be a challenge. Maybe this is something data center operators can help us with.

Looking forward: collaboration is key

The only path to successful adoption of all liquid cooling is through close collaboration between different actors in the ecosystem.

At Calyos, we’re contributing through innovative system designs and patented technologies that address problems like instabilities, sustainability, and vapor pressure drop.

As AI workloads soar and thermal constraints tighten, two-phase cooling presents one of the most technically promising solution. But its success depends not only on physics—but on partnerships, smart system integration, and ongoing innovation.

We’re excited to be part of this journey—and we hope you are too.

‍