As processor power continues to rise and AI clusters push beyond previously accepted density limits, datacentre operators are increasingly evaluating advanced liquid cooling approaches. Among immersion-based solutions, the discussion is now often framed as a choice between single-phase and two-phase cooling. Both approaches move well beyond the limitations of air cooling, yet they are based on fundamentally different physical principles. These differences lead to meaningful divergences in system design, operational complexity, sustainability profiles, and supply chain risk.

Today, operators encounter a growing range of technologies described under this umbrella, ranging from tank-based immersion to evaporative cold plates. Understanding what these terms actually mean—and where each fits—is essential when making architectural decisions intended to last a decade or more.

Single-Phase Immersion

In single-phase immersion cooling, the defining characteristic is thermal stability. Servers are submerged in a dielectric fluid that remains in a liquid state throughout the entire operation. Heat generated by CPUs, GPUs, memory, networking, and power delivery components is transferred directly into the surrounding liquid and carried to a heat exchanger via controlled circulation.

Crucially, the fluid does not boil. There is no evaporation, no condensation, and no vapour handling required. This continuous liquid state means the system operates without the pressure-vessel behaviour found in phase-change systems. This physical simplicity translates into several structural advantages. Because the fluid remains liquid, it provides uniform cooling to all server components rather than just the high-heat processors, and it functions independently of chip package geometry or cold plate designs. This simplicity aligns strongly with operational predictability, making single-phase immersion the most widely adopted approach for scalable, long-term deployments.

Two-Phase Cooling

Two-phase cooling relies on a fluid engineered to intentionally boil at operating temperatures. The heat causes the liquid to transition into vapour, which must then be captured, condensed, and returned to the system as liquid. This cycle—liquid to vapour and back to liquid—creates a highly efficient heat transfer mechanism, but it introduces significant mechanical complexity.

This category includes two distinct approaches often grouped together. Two-phase immersion cooling submerges the entire server in a tank where the fluid boils and condenses on internal coils. While thermodynamically potent, this requires hermetically sealed tanks to prevent the loss of expensive, volatile fluids and careful engineering to manage turbulence and vapour pressure.

The second approach is two-phase on-chip cooling, often using evaporative cold plates. Here, the boiling occurs inside a sealed plate attached directly to the processor. While this solves the heat flux challenge for the CPU or GPU, it remains a form of direct liquid cooling rather than true immersion. Consequently, the memory, power delivery, and storage components often remain reliant on traditional air cooling and server fans, failing to eliminate the systemic complexity of airflow management.

Feature Single-Phase Immersion Two-Phase Immersion
Thermal Mechanism Heat is moved by circulating liquid through convection without any state change occurring. Heat is removed via phase change cycles of evaporation and condensation.
System Pressure The system operates at atmospheric pressure in open or semi-open tanks, eliminating pressure risks. The process requires hermetically sealed tanks to contain vapour and prevent leaks.
Component Cooling It cools the entire board, including CPU, RAM, and VRMs, with equal uniformity. Boiling is generally localised to hot spots and requires careful flow management to ensure coverage.
Maintenance Maintenance is straightforward because components are accessible and the fluid is non-volatile. Breaking seals for maintenance is complex and risks volatile fluid loss or exposure.

Sustainability and Fluid Risk as Constraints

One of the most consequential differences between single-phase and two-phase cooling lies in fluid behaviour and long-term risk. Two-phase fluids often introduce challenges related to PFAS or PFAS-adjacent chemistries, high global warming potential, and unavoidable vapour losses over time. As ESG governance becomes more formalised, these factors are no longer theoretical. Many enterprises are already disqualifying two-phase solutions due to regulatory uncertainty and environmental liability.

Single-phase immersion fluids avoid these issues by design. They are non-volatile, available from multiple global suppliers, and typically reclaimable. This creates a much lower risk profile regarding future environmental regulations and reporting mandates. Furthermore, the supply chain for single-phase fluids offers "optionality"—a broad ecosystem of vendors—whereas two-phase cooling often depends on a narrow chemical supply base, creating vendor lock-in and potential exposure to discontinuation risks.

Risk Factor Single-Phase Immersion Two-Phase Immersion
Fluid Chemistry It uses synthetic or bio-based fluids that are often non-toxic and biodegradable. It often relies on fluorinated fluids containing PFAS or PFAS-adjacent chemistries.
Regulatory Risk Risk is low as the technology aligns well with current environmental directives. Risk is high due to increasing scrutiny regarding "forever chemicals" like PFAS.
Supply Chain The supply chain is resilient with a broad ecosystem of global fluid suppliers. The supply base is constrained with a high risk of vendor lock-in or discontinuation.
Long-Term Viability Viability is high due to standardised infrastructure that adapts to future hardware generations. Long-term viability is uncertain due to regulatory pressure and potential fluid liabilities.

Conclusion

While two-phase cooling technologies demonstrate impressive thermal performance at the extreme end of the spectrum, they are best suited to highly specialised, niche scenarios. For the majority of datacentres—and especially for operators planning for long-term, heterogeneous, and sustainability-constrained environments—single-phase immersion cooling offers a more balanced architecture. It delivers high performance without introducing volatility, regulatory exposure, or supply chain fragility. In the broader conversation about future-proofing infrastructure, single-phase immersion stands out not only for what it enables, but for the complexity and risk it deliberately avoids.