Immersion cooling sounds simple: put the servers in dielectric fluid and remove the heat. The real decision starts with what that fluid does next.
Single-phase immersion cooling and two-phase immersion cooling both submerge your hardware directly in a dielectric fluid instead of cooling it with air—but that’s roughly where the similarity ends. The fluid behaves completely differently in each system, and that difference drives almost every other decision: what tank you need, how much fluid you’re buying, what maintenance looks like, and how much density you can actually support per rack.
If you’re evaluating immersion cooling for the first time, or you’ve heard both terms used interchangeably and want to know what actually separates them, this is the distinction that decides which one fits your deployment.
Follow the Heat
The clearest way to see what separates these two architectures is to trace the heat from the chip to the point of rejection:
Single-phase: chip/server → dielectric fluid → tank loop → heat exchanger/CDU → facility loop → heat rejection
Two-phase: chip/server → boiling dielectric fluid → vapor → condenser → facility loop → heat rejection
Both start at the chip and end at facility heat rejection — the difference is entirely in what happens to the fluid in between.
What Single-Phase Immersion Cooling Actually Does
In a single-phase system, servers sit submerged in a dielectric fluid that stays liquid throughout the entire cooling cycle—it never boils, never changes state. The dielectric fluid absorbs heat directly from submerged components, with natural and forced convection moving that heat through the fluid before it is transferred out of the tank. The fluid then circulates through a heat exchanger or CDU, where heat is transferred to the facility cooling loop before ultimately being rejected through equipment such as a dry cooler, chiller, or other heat-rejection system.
Because the fluid is not intended to boil during normal operation, single-phase systems generally do not require the same vapor-containment and condensation architecture used in two-phase immersion. That makes the mechanical design simpler, the fluid itself more forgiving to work with, and the maintenance routine closer to what a facilities team already understands from other liquid-cooling systems—pumps, heat exchangers, fluid quality checks.
What Two-Phase Immersion Cooling Actually Does
Two-phase immersion uses a dielectric fluid engineered with a low boiling point—low enough that it boils directly off the surface of hot components while they’re submerged. That boiling is the whole point: as the fluid vaporizes, it carries heat away far more efficiently than simple liquid contact can. The phase change takes advantage of latent heat transfer, allowing significant heat to be removed at relatively stable fluid temperatures and supporting very high local heat fluxes.
The vapor rises into the upper portion of the immersion enclosure, where a condenser removes the heat and condenses the vapor back into liquid. Two-phase systems therefore require engineered vapor containment, condensation, and fluid-management provisions. Some two-phase designs can run largely on that natural convection cycle with minimal or no pumping, which is part of the efficiency appeal — but it requires a fluid formulated specifically for that boiling point and thermal behavior.
The Core Difference
| Factor | Single-Phase | Two-Phase |
| Fluid behavior | Stays liquid throughout the cycle | Boils and condenses (phase change) |
| Heat transfer mechanism | Direct liquid contact + pumped circulation | Latent heat via boiling/condensing |
| Tank design | Open or lightly enclosed | Sealed — must contain vapor |
| Heat flux capability | High — capable of supporting high-density IT depending on fluid, flow, tank and heat-exchanger design | High — phase-change heat transfer can support very high component heat flux depending on fluid chemistry and system design |
| Fluid cost | Varies by fluid chemistry; multiple dielectric-fluid options are available | Varies significantly by fluid chemistry, boiling point, availability and regulatory profile |
| Maintenance profile | Closer to conventional liquid-cooling upkeep | More specialized — sealed system, engineered fluid handling |
| Typical fit | Most enterprise/hyperscale racks at today’s densities | Extreme-density racks pushing the outer edge of limits |
Neither one is a strictly “better” technology—they’re built for different points on the density curve, and the fluid technology itself is the reason.
Why This Matters for Your Density Target
Both single-phase and two-phase immersion can support very high-density AI and HPC deployments. The choice is not simply determined by which architecture supports the highest rack density. Fluid chemistry, server compatibility, heat flux, facility-water temperatures, heat-rejection design, serviceability, operational model, cost, and long-term fluid availability all influence which approach is the better fit.
Fluid chemistry deserves particular attention in any immersion design. Certain fluorinated dielectric fluids historically used in two-phase applications have faced increased environmental, PFAS-related, and supply-chain scrutiny. Fluid chemistry, regulatory status, availability, and long-term vendor support should therefore be validated as part of the design process.
How to Decide
A few practical questions we walk through with clients evaluating immersion cooling:
- What’s your actual per-rack density target, today and in 2–3 years? Both architectures can support high-density deployments, so density alone shouldn’t be the deciding factor.
- What’s your team’s comfort level with a sealed, vapor-based system? Two-phase maintenance is a different skill set than most facilities teams already have from air- or single-phase-liquid-cooled environments.
- Does your budget support the fluid cost and chemistry your deployment requires? Fluid pricing varies meaningfully by chemistry for both single-phase and two-phase—that has to be weighed against the specific fluid, tank, and heat-exchanger design you’re evaluating.
- Is fluid chemistry availability and long-term supply a factor for your timeline? Given the regulatory attention on some fluorinated fluid chemistries, it’s worth confirming current fluid availability and any transition plans with your vendor before locking in a design.
Facility and Environmental Considerations
Beyond the fluid itself, the two architectures place different demands on the facility around them. Single-phase’s pumped-loop design integrates fairly directly with cooling infrastructure most facilities already have some version of—a CDU or dry cooler on the other end of the loop, similar in concept to how a chilled-water system moves heat off-site. Two-phase’s sealed-tank, natural-convection design can reduce or eliminate active pumping in some configurations, which lowers electrical load for that part of the system, but it shifts the engineering burden onto the tank seal integrity and condenser sizing instead.
Floor loading is worth checking either way — immersion tanks, especially at scale, carry substantial fluid weight that most raised-floor data centers weren’t originally designed around. That’s a structural conversation with a facilities engineer before tank placement, not an afterthought once equipment shows up on site.
Serviceability is the other practical difference that doesn’t always come up early enough in the planning process. Pulling a server for maintenance in a single-phase tank means lifting it out of liquid — straightforward, if a little messier than air-cooled service, and it’s a routine your team can learn quickly. In a two-phase tank, opening a sealed vessel to service hardware means managing vapor containment during the process, which is a different maintenance discipline and typically requires either specialized training or vendor-supported service contracts, at least early in a deployment.
Common Deployment Patterns
Single-phase immersion is widely used for high-density AI, HPC, enterprise, and edge applications because of its relatively straightforward fluid-management and heat-exchange architecture. Two-phase immersion is another high-density option that uses boiling and condensation rather than pumped bulk-fluid circulation. Which architecture fits best depends on the workload, server design, component heat flux, fluid chemistry, service model, facility infrastructure, and project economics — not rack density alone.
Mixed deployments — single-phase across most of the floor, two-phase reserved for a specific zone — aren’t unusual either, especially in facilities that serve multiple workload types under one roof. The architecture doesn’t have to be a single, facility-wide decision; it can be matched zone by zone to what’s actually running there and what that workload requires.
Immersion isn’t the only path to higher rack density, either. Many of the same facilities running immersion in one zone are running rear door heat exchangers in another, where a full tank isn’t practical. Which rack-level architecture fits depends on the same variables — fluid or water infrastructure, serviceability, and how the workload is actually distributed across the room.