At the extreme end of the rack density spectrum — beyond what even direct-to-chip cooling can comfortably manage — immersion cooling has moved from a curiosity confined to cryptocurrency mining and academic high-performance computing into a credible, if still niche, option for the most demanding AI and HPC deployments.
Two Distinct Approaches, Two Distinct Trade-Offs
Single-phase immersion submerges entire servers in a dielectric fluid — typically mineral oil or a synthetic hydrocarbon — that remains liquid throughout the cooling cycle, supporting rack densities in the range of 100–120 kW. Two-phase immersion uses a different class of fluid, engineered to boil at the chip's operating temperature; the resulting phase change dramatically increases heat transfer efficiency, supporting densities of 150 kW and beyond, with vapour condensing on a cooled surface and returning to the liquid phase in a closed loop.
Two-phase immersion is the more thermally capable of the two, with reported heat flux performance many multiples higher than single-phase water cooling, but it comes with materially higher fluid costs and increasing regulatory attention tied to PFAS-related chemistry in some two-phase fluid formulations, with a 2030 phase-out horizon under discussion in European regulatory contexts.
Why Immersion Is Not Simply "Better" Than Direct-to-Chip
It would be a mistake to treat immersion as a strictly superior technology that direct-to-chip cooling will eventually be replaced by. Immersion requires purpose-built tanks, server designs without fans, and dedicated floor space configured around tank footprints rather than conventional rack rows — making it far better suited to greenfield builds than to retrofitting existing facilities. It also requires a different skill set for maintenance: technicians need training specific to handling dielectric fluids safely, and serviceability of submerged hardware differs meaningfully from servicing a conventional rack-mounted server.
- Immersion is currently concentrated in use cases where density matters more than serviceability convenience — cryptocurrency mining, certain HPC clusters, and edge deployments with severe space constraints
- Chip manufacturer certification has historically lagged for immersion compared with direct-to-chip, though this is improving — Intel's certification of immersion fluids for specific Xeon processor generations, with an accompanying warranty rider, was a notable step toward broader vendor confidence
- The technology's growth rate is among the fastest in the liquid cooling market, even though it remains a small share of total deployments today
Immersion cooling is not a more advanced version of direct-to-chip cooling — it is a different engineering trade-off, suited to a different set of problems.
When Immersion Genuinely Makes Sense
The clearest case for immersion arises when a facility needs to support rack densities that exceed what direct-to-chip cooling can reliably manage, and where the operational complexity of a purpose-built tank environment is an acceptable trade-off against the alternative of simply not being able to support the required density at all. This is increasingly relevant for the most extreme frontier AI training deployments, where every available means of increasing compute density within a constrained power and space envelope has commercial value.
For most AI-ready facilities serving a broader mix of workloads, however, direct-to-chip cooling remains the more pragmatic choice — offering substantial density gains over air cooling without the full operational discontinuity that immersion introduces.
The Regulatory Picture Adds a Layer of Complexity
Two-phase immersion fluids in particular face growing regulatory scrutiny tied to per- and polyfluoroalkyl substances, with European regulatory discussions pointing toward a phase-out horizon around 2030 for some fluid chemistries. This creates genuine uncertainty for operators evaluating long-duration investment in two-phase immersion infrastructure today, since a facility designed around a fluid chemistry that later faces regulatory restriction could face costly remediation or fluid replacement well before the end of its intended operating life. Developers considering two-phase immersion should factor this regulatory trajectory explicitly into investment decisions, rather than treating current fluid chemistry as a permanent design assumption.
Single-phase immersion fluids generally face less acute regulatory pressure, which is one of several reasons single-phase approaches remain more widely deployed than two-phase systems despite the latter's superior thermal performance on paper.
Evaluating Immersion as Part of a Broader Cooling Strategy
DATAPERT advises clients on when immersion cooling genuinely adds value to a facility programme, and when a more conventional liquid cooling strategy better serves the project's commercial and operational objectives. Explore our technology integration and data center development capabilities, or start a project to discuss cooling strategy for an extreme-density deployment.
