Immersion Cooling Explained: Single-Phase vs Two-Phase

Immersion Cooling in 2026: Why UK Data Centres Are Looking Again

AI training and inference clusters have pushed typical GPU rack loads well past what perimeter or in-row air cooling was designed for, and that shift is what is dragging immersion cooling out of the lab and into procurement conversations across UK operators. Where a decade of data-centre design assumed a few kilowatts per rack, current AI deployments routinely demand tens of kilowatts, and liquid — either pumped to the chip or as a full immersion bath — has become the only practical way to remove that heat without oversized air-handling plant.

That said, the brief evidence available in mid-2026 does not support a claim that immersion has become the default choice across UK data halls. Most new high-density UK builds are exploring direct-to-chip liquid cooling first, with immersion piloted in limited AI and HPC environments rather than rolled out across broad production estates — particularly where long-term policy and compliance impacts remain uncertain. Understanding why requires separating the two fundamentally different immersion architectures on the market: single-phase and two-phase.

What Is Immersion Cooling? Single-Phase vs Two-Phase Mechanics

At its core, immersion cooling works by fully submerging servers or IT components in a thermally conductive dielectric liquid that absorbs heat directly from the hardware, which is then rejected via a heat exchanger to a secondary loop. There is no air path, no server fans, and — because liquids conduct heat vastly more effectively than air — the chip-level heat transfer coefficients involved are reported to be 1,000 to 3,000 times higher than air-based convection. That physics figure describes heat transfer at the component level, not facility-wide energy use, but it explains why immersion supports such dense hardware packing.

Single-phase immersion uses a dielectric coolant — typically mineral oil or a synthetic hydrocarbon/ester-based fluid — that never changes state. It stays liquid throughout, absorbing heat as it circulates past submerged hardware, then gets pumped to a heat exchanger and returned cooled to the bath. This is mechanically the simpler of the two approaches and is generally compatible with standard server hardware.

Two-phase immersion instead uses a specialised fluorocarbon-based fluid engineered to boil at a low temperature. Heat from the chip causes the liquid to evaporate locally; the vapour rises, contacts a cooled condenser coil above the fluid surface, and condenses back into liquid, falling back into the bath. Because the phase change itself carries the heat, two-phase systems can rely on passive boiling and condensation to minimise the pumping energy that single-phase systems require.

The Hard Numbers: PUE, Density and Cost

The efficiency case for immersion rests on a set of PUE figures that come from different sources and different scopes, and they should not be blended into a single average. Park Place Technologies, citing The Register, reports a typical traditional air-cooled data centre running at around 1.58 PUE in 2022, against single-phase immersion case studies reaching approximately 1.05–1.10. Separately, a 2025 vendor comparison from Green Revolution Cooling (GRC) places single-phase immersion at 1.02–1.03 PUE against two-phase immersion at 1.01–1.02 PUE under similar high-density design assumptions — a marginal efficiency edge for two-phase, but measured in a different study to the 1.05–1.10 figure above.

Rack density tells a similarly two-tier story. Single-phase immersion typically supports 50–100kW per rack, while two-phase immersion is used for ultra-high-density AI cluster configurations exceeding 200kW per rack, leveraging passive boiling and condensation at very high power. If your workload profile sits closer to the top of that range, it is worth running the numbers through a server room cooling calculator before assuming immersion is even necessary — plenty of 60–80kW AI racks are still served adequately by direct-to-chip liquid cooling.

On cost, a 2024 vendor case study from 2CRSi claims single-phase immersion can deliver roughly 30% total cost of ownership savings, a 20% increase in IT hardware lifespan, around 39% lower carbon emissions, and approximately 91% lower water consumption compared with traditional cooling. These are directional, vendor-reported figures from customer case studies rather than independently verified or regulator-guaranteed outcomes, and UK buyers should treat them as comparative illustrations to test against their own site data rather than numbers to plug straight into a business case.

• •Conventional air-cooled PUE (2022 reference): 1.58
• •Single-phase immersion PUE (case studies): 1.05–1.10
• •Single-phase vs two-phase PUE (GRC comparison): 1.02–1.03 vs 1.01–1.02
• •Single-phase rack density: 50–100kW; two-phase AI clusters: over 200kW

Fluids, System Components and Integration with UK Infrastructure

The fluid itself drives much of the cost and risk profile of each approach. Single-phase dielectric fluids are reported to carry a lower price, zero global warming potential, and no ozone depletion, with a stable chemical state that keeps handling and top-up straightforward. Two-phase fluorocarbon fluids, by contrast, typically cost several times more per litre and are subject to tightening regulatory and supply-chain scrutiny because many fall into the PFAS chemical class — a risk factor rather than a confirmed ban, but one that matters for any UK organisation planning multi-decade asset life.

On procurement-level pricing, industry market data puts two-phase fluorocarbon fluid at roughly US$80–200 per litre, against single-phase hydrocarbon or ester-based oils in the region of US$15–25 per litre. Neither figure includes installation, ongoing maintenance, or end-of-life disposal costs, so any UK financial model needs to build those in separately rather than relying on fluid price alone.

System integration differs meaningfully between the two. Single-phase tanks need pumps to circulate heated dielectric liquid to secondary air-to-liquid or liquid-to-liquid heat exchangers before the cooled fluid returns to the tank — broadly comparable in plumbing complexity to a chilled-water loop, which makes it easier to retrofit into an existing UK facility's secondary cooling infrastructure. Two-phase systems can rely on passive boiling and condensation to cut pumping energy, but they require sealed tanks and vapour-containment engineering that most legacy UK data halls were never designed around. For sites already running or planning direct-to-chip liquid cooling, it is worth comparing that architecture directly against full immersion — see our breakdown of liquid vs immersion cooling for the return-on-capital-employed angle.

Real-World Challenges for UK Operators

Even where the efficiency numbers are compelling, UK operators in regulated sectors cite the same practical friction points repeatedly. Retrofit complexity is real: an immersion tank changes rack form factor, floor loading, and service access in ways that a like-for-like server swap does not. OEM warranty support for submerged hardware is not universal, and operational teams need new skills for fluid handling, sampling, and disposal that most UK facilities teams have not built up. Integrating immersion into established resilience and compliance frameworks — change control, incident response, business continuity testing — takes real project time, not just a hardware order.

Analysis from Asperitas frames this well: two-phase immersion demonstrates genuinely impressive thermal performance for extreme workloads, but it is best suited to highly specialised, niche scenarios. Single-phase immersion, by contrast, offers a more balanced architecture for long-term, heterogeneous, sustainability-constrained environments — precisely the profile of most UK enterprise and regulated-sector estates, which run mixed hardware generations for many years rather than a single homogeneous AI cluster. That is the practical answer to when single-phase beats two-phase and vice versa: two-phase wins on raw density and marginal PUE for a concentrated, short-cycle AI island; single-phase wins on fluid cost, regulatory exposure, and compatibility across a mixed, long-lived fleet.

As of mid-2026, this is exactly the pattern playing out across UK regulated sectors: immersion piloted for targeted AI or HPC islands, while mainstream racks default to direct-to-chip liquid cooling, with operators keeping a watching brief on PFAS regulation, vendor ecosystem maturity, and forthcoming guidance from sector regulators before committing at scale.

Sustainability and ESG: Meeting UK Net Zero Targets

The sustainability case for immersion is where UK ESG reporting pressure and technical efficiency genuinely align. UK data centres are increasingly measured against PUE and water-use metrics as part of ESG reporting, and the documented PUE improvements — from around 1.58 in legacy air-cooled sites down to 1.05–1.10 or better in optimised immersion deployments — feed directly into SECR and TCFD-style disclosure expectations for organisations tracking energy intensity and net-zero progress.

Single-phase immersion cabinets can also directly cut water consumption by enabling hybrid cooling towers or adiabatic dry coolers with very low water use, alongside indirectly reducing water use at power plants through lower electricity demand — reinforcing the vendor-claimed 91% water consumption reduction and 39% carbon emissions reduction cited above, albeit as directional case-study figures rather than sector-wide averages. Single-phase fluids' stable, non-ozone-depleting, zero-GWP profile also sits more comfortably with UK sustainability narratives and heat-reuse or district-heating ambitions than two-phase fluorocarbon fluids, which carry PFAS-related regulatory uncertainty that regulated UK sectors are actively watching rather than resolving.

Choosing Your Solution: Fluids, Vendors and Selection Criteria

The vendor landscape documented in current technical and market literature spans both architectures: LiquidStack and Araner detail the mechanics of both single- and two-phase systems; Submer and GRC are prominent single-phase and comparative-testing voices respectively; Asperitas and 2CRSi publish suitability and TCO analysis; and independent market research from ResearchIntelo tracks fluid cost and density trends across the sub-market. Note that vendor material inevitably carries some bias toward the architecture that vendor sells, so cross-check any single source's efficiency or cost claims against independent data before committing capital.

A practical UK selection checklist should cover: fluid properties (GWP, PFAS classification, stability, and cost per litre); system design (pumped single-phase vs passive two-phase, and the resulting facility loop and CDU requirements); rack density needs against the 50–100kW single-phase and 200kW-plus two-phase benchmarks; hardware compatibility and OEM warranty terms for submerged operation; end-of-life fluid disposal and regulatory compliance planning; and — critically for regulated UK organisations — how the chosen architecture fits existing resilience, change-control, and audit frameworks rather than sitting outside them. For teams still scoping whether immersion or direct-to-chip is the right starting point for a specific workload, our air vs liquid cooling comparison is a useful next step.

Conclusion: Is Immersion Right for Your UK Data Centre in 2026?

The technical case for immersion cooling is no longer in question: the PUE, density, and vendor-reported TCO figures are real and well documented. What remains unresolved for UK regulated sectors is not whether immersion works, but whether the fluid-cost, disposal, warranty, and regulatory exposure — particularly around PFAS-class two-phase fluids — are acceptable trade-offs against a multi-decade asset horizon. For most UK organisations running mixed, long-lived hardware estates, single-phase immersion is the more defensible starting point; two-phase remains a specialist tool for the narrow set of ultra-high-density AI clusters where its marginal PUE and density advantage justify the fluid cost and regulatory watch-list status. Expect that split — pilot islands rather than fleet-wide adoption — to persist through 2026 and into 2027 until PFAS policy and vendor ecosystem maturity settle further.

Sources

Every figure in this article traces to the sources below.

• •LiquidStack — mechanics of immersion cooling and heat exchange
• •LiquidStack — single-phase fluid types and system design
• •Submer — single-phase immersion mechanics
• •Araner — two-phase immersion mechanics
• •ResearchIntelo — PUE, rack density and fluid cost market data
• •Park Place Technologies — conventional vs single-phase PUE (citing The Register)
• •Science and Technology for Energy Transition (STET) — heat transfer, fluid GWP, water savings
• •Asperitas — single-phase vs two-phase suitability
• •2CRSi — single-phase TCO, water, carbon and lifespan claims
• •GRC (Green Revolution Cooling) — two-phase vs single-phase PUE comparison