Key Takeaways:
- Both single-phase and two-phase immersion cooling submerge servers in dielectric fluid – but single-phase fluid stays liquid and is pumped, while two-phase fluid boils and condenses, using latent heat to move significantly more energy per unit of fluid.
- High rack density alone should not drive the decision – fluid chemistry, tank design, serviceability, and long-term fluid availability are the factors that actually separate the right fit from the wrong one.
- Some fluorinated fluids used in two-phase systems face active PFAS regulatory scrutiny, with manufacturers like 3M already phasing out production of certain chemistries – fluid supply chain validation belongs in the design phase, not after a purchase order is signed.
- Triton Thermal’s breakdown of single-phase vs. two-phase immersion cooling goes deeper on how fluid behavior shapes every downstream decision in a deployment.
Both Submerge Servers – Then Everything Diverges
Immersion cooling has a straightforward pitch: pull the servers out of the airstream, sink them in dielectric fluid, and let the liquid do what air never could. What sounds like a single technology is actually two architecturally distinct approaches that share a starting point and very little else.
Single-phase and two-phase immersion cooling both submerge hardware in a non-conductive dielectric fluid. That is the common ground. From there, fluid behavior splits completely – and that split drives tank design, fluid cost, maintenance complexity, facility requirements, and ultimately which architecture belongs in a given deployment. Understanding the difference between pumped liquid flow and a boiling-and-condensing phase cycle is the foundation for making a sound infrastructure decision.
Follow the Heat: Two Paths From Chip to Rejection
The clearest way to separate these architectures is to trace heat from the chip outward. Both paths start at the same place and end at facility heat rejection – what happens in between is entirely different.
Single-Phase: Pumped Liquid, External Heat Exchange
In a single-phase system, servers sit submerged in dielectric fluid that never changes state. It stays liquid throughout the entire cooling cycle. Heat transfers from components directly into the surrounding fluid through contact and convection, and a pump then circulates that heated fluid through an external heat exchanger or cooling distribution unit (CDU). The heat moves into a facility cooling loop – typically a dry cooler, chiller, or similar system – and the cooled fluid returns to the tank to repeat the cycle.
The heat path looks like this: chip → dielectric fluid → CDU/heat exchanger → facility loop → heat rejection.
Two-Phase: Boiling Fluid, Vapor, Condenser
Two-phase systems use a dielectric fluid engineered with a deliberately low boiling point – low enough that it boils directly off hot component surfaces while those components are submerged. That boiling is the mechanism, not a side effect. As the fluid vaporizes, it carries heat away through latent heat transfer, which is fundamentally more efficient than moving heat through bulk liquid circulation alone. The vapor rises into the upper section of the enclosure, contacts a condenser, releases its heat, and drops back into the tank as liquid.
The heat path looks like this: chip → boiling dielectric fluid → vapor → condenser → facility loop → heat rejection.
Some two-phase configurations can run with minimal or no active pumping, relying on that natural boil-and-condense convection cycle. That is part of the efficiency appeal, but it requires a fluid formulated specifically for the target boiling point and enclosure geometry.
Why Fluid Behavior Changes Everything
Latent Heat vs. Sensible Heat Transfer
The physics behind these two systems explain why the fluid state matters so much. Single-phase cooling relies on sensible heat transfer – the fluid absorbs heat and its temperature rises, and that temperature differential drives how much heat can be moved. Two-phase cooling taps into latent heat: the energy absorbed during a phase change (liquid to vapor) is enormous compared to simply warming a liquid by a few degrees. This allows two-phase systems to remove very high heat loads at relatively stable fluid temperatures, which is why they suit extreme component heat flux scenarios particularly well.
Tank Design Follows Fluid State
Because single-phase fluid never produces vapor under normal operation, those tanks do not need to contain or manage a vapor phase. Designs can be open or lightly enclosed, and the surrounding infrastructure – pumps, heat exchangers, fluid lines – is conceptually familiar to facilities teams already running liquid-cooled environments.
Two-phase tanks are a different story. Vapor containment is non-negotiable. The enclosure must be engineered to keep vapor inside the system until it condenses, which means sealed designs, specific condenser sizing, and more demanding requirements for seal integrity over the system’s life. The fluid cannot be a standard dielectric option – it has to be formulated for the right boiling point, which narrows the field considerably.
Density Alone Won’t Make the Decision
A common misconception is that rack density is the primary decision variable – that once a deployment crosses some threshold, two-phase is the only answer. That is not how the decision actually plays out. Both architectures can support very high-density AI and HPC deployments. Single-phase immersion cooling has been documented supporting rack densities of 100kW and above, a meaningful data point that single-phase is not a compromise for lower densities.
The Real Decision Factors
The factors that actually differentiate a good fit from a poor one include:
- Per-rack heat flux targets – both today and 2-3 years out, since architecture decisions are hard to reverse at scale
- Facility infrastructure – single-phase integrates more directly with cooling infrastructure most facilities already have; two-phase shifts engineering burden to tank seal integrity and condenser design
- Serviceability model – pulling a server from a single-phase tank is straightforward; opening a sealed two-phase enclosure requires managing vapor containment and typically specialized training or vendor support contracts
- Budget – fluid pricing varies meaningfully by chemistry for both architectures, but two-phase fluids engineered for specific boiling points often carry a cost premium that compounds at scale
- Floor loading – immersion tanks carry substantial fluid weight that many raised-floor facilities were not originally designed for; this is a structural conversation that belongs early in planning, not on installation day
PFAS Regulations and Fluid Availability
Fluid chemistry deserves particular scrutiny for two-phase deployments. Many fluorinated dielectric fluids historically used in two-phase systems contain PFAS (per- and polyfluoroalkyl substances), which face increasing regulatory pressure from the EPA and state-level agencies. 3M has already announced the phase-out of certain PFAS-containing chemistries – including Novec-branded products widely used in two-phase applications. If the fluid underpinning a system design is being discontinued or restricted, long-term supply chain viability becomes a real planning risk. Confirming current fluid availability and any vendor transition plans before locking in a design is a straightforward step that is easy to skip and costly to regret.
Maintenance and Serviceability Realities
Single-phase maintenance is broadly familiar to facilities teams with any liquid-cooling experience: pumps, heat exchangers, and periodic fluid quality checks. Fluid does not evaporate under normal operation, so top-offs are minimal, and the open or lightly enclosed tank design means hardware pulls are relatively routine – lift the server out, service it, resubmerge it. The process is messier than air-cooled service, but learnable quickly.
Two-phase maintenance operates differently. Opening a sealed vapor-containment vessel to access hardware means actively managing the vapor environment, which is a distinct skill set. Most organizations deploying two-phase for the first time rely on vendor-supported service agreements, at least during early operation. That is a real operational cost and dependency that should be priced into the total ownership model from the beginning – not discovered after the first hardware failure.
Choose the Architecture That Fits Your Deployment
Neither single-phase nor two-phase immersion is universally superior. They are optimized for different positions on the heat flux curve, and the fluid technology itself is the reason those positions exist. Single-phase works exceptionally well across a wide range of high-density AI and HPC environments, offering simpler fluid management, broader chemistry options, and a maintenance model that does not require specialized vapor-handling skills. Two-phase makes sense at the outer edge of component heat flux, where latent heat transfer provides capabilities that pumped liquid flow physically cannot match – assuming the fluid chemistry, supply chain, and operational model all align.
Mixed deployments are not unusual either. Facilities serving multiple workload types often run single-phase across most of the floor and reserve two-phase for a specific high-flux zone. The architecture does not have to be a single facility-wide choice; it can be matched zone by zone to what is actually running there.
The starting point is always the same: define what is in the rack, what the heat flux target is, what the facility can support, and what the team can realistically operate. Once those variables are on the table, the right architecture tends to become clear.
Triton Thermal
3350 Yale St.
Houston
Texas
77018
United States
