Immersion cooling is moving from a niche HPC and crypto-mining play to a mainstream data-center thermal strategy, with the global market tracked at USD 0.57 B in 2025 and projected to reach USD 2.61 B by 2032, a 24.2% CAGR over 2025-2032 [S2]. North America is currently the largest regional consumer due to dense hyperscale and AI build-outs [S2].
The technology submerges IT hardware in electrically insulating (dielectric) fluids that absorb heat directly from components, replacing or supplementing air-side heat exchange. Two commercial formats dominate: single-phase systems, where the fluid stays liquid and circulates through a heat-exchanger loop, and two-phase systems, where the fluid boils at the chip surface and condenses on a tank lid for passive return. Both formats enable higher component density, lower PUE contributions from fans, and longer component service life versus forced-air racks [S2].
Single-Phase vs Two-Phase: Where Each Format Fits
Single-phase immersion uses high-boiling dielectric fluids (commonly synthetic hydrocarbons or fluorocarbons) that remain liquid across the operating range, with heat rejected via a fluid-to-water heat exchanger outside the tank [S2]. Procurement advantage: simpler tank mechanics, no condensation management, and easier field service, which is why GRC (Green Revolution Cooling) targeted single-phase deployments in its October 2023 Middle East & Africa rollout with Dell Technologies and DCV Industries [S2].
Two-phase immersion relies on fluids engineered to boil at chip-junction temperatures, delivering higher heat-transfer coefficients per unit area and allowing denser ASIC or GPU packing, at the cost of more complex tank sealing, condensate return paths, and fluid management [S2]. A useful engineering rule of thumb — not a vendor claim — is that two-phase systems become economically attractive once per-rack heat flux exceeds the practical limits of single-phase fluid loops, a threshold that AI training accelerators routinely cross. For facilities planning mixed HPC and AI clusters, the dominant pattern is single-phase for general compute density and two-phase reserved for accelerator-dense AI pods.
Dielectric Fluid Selection: The New Procurement Battleground
Cooling-fluid chemistry is now a first-tier spec line, not an accessory. MarketsandMarkets segments the fluid market into synthetic fluids, mineral oil, fluorocarbon-based fluids, and others, with synthetic fluids forecast to lead the market through 2032 on efficiency, low maintenance, and hardware reliability [S2]. Mineral oil remains the lowest-cost option but carries viscosity and longevity trade-offs at elevated temperatures.
Fluid-manufacturer partnerships are the most visible 2026 signal: Shell and Asperitas announced a bespoke immersion cooling fluid for data centers in the lead-up to the 2026 Lube Focus Week (June 9-11), positioning lubricant majors as co-engineers of tank fluids rather than commodity suppliers [S1]. For buyers, the procurement checklist now reads: dielectric strength (kV per mm), kinematic viscosity at operating temperature, boiling point (single vs two-phase compatibility), material compatibility with PCB coatings and connector polymers, flash point, and end-of-life fluid-recycling pathway. Skipping any one of these is a known path to unplanned downtime when rubber swell or vapor-lock surfaces 12-18 months into operation.
Hyperscale, HPC, and Crypto: Who Should Specify Immersion — and Who Should Not
Immersion cooling is a strong fit for high-performance computing clusters, AI training halls, edge-computing pods in space-constrained sites, and ASIC/GPU cryptocurrency mining farms where maximizing hash rate per square foot is the dominant economic lever [S2]. The financial case is strongest where power-density per rack exceeds 50 kW, a threshold that conventional air or even rear-door heat exchangers struggle to handle economically.
It is a poor fit for small enterprise server rooms with mixed-vintage hardware, sites lacking the structural floor loading for filled tanks (a fully populated single-phase tank can exceed 1,500 kg per rack), and any deployment where the operations team lacks fluid-handling training. For general office and light commercial loads, conventional precision air conditioning remains the lower-TCO path. Operators weighing a move should also factor in the secondary heat-reuse opportunity: heated fluid exiting the rack can feed district heating or absorption chillers, a pattern visible in European pilot deployments tied to waste-heat recovery mandates.
Standards, Safety, and the Open Specification Gaps
No single global standard yet covers immersion cooling end-to-end the way ASHRAE TC 9.9 covers air-side data-center thermal envelopes, which is why most operators currently combine vendor test data with general electrical-safety frameworks (such as IEC 60079-series for hazardous-area classification where fluids are classified, and IEC/UL 60950-1 or IEC 62368-1 for IT-equipment safety) and facility-level fire-suppression codes [S2]. The fluid itself becomes the dominant fire-safety variable: a high-flash-point synthetic dielectric behaves very differently from a low-flash-point mineral cut in a leak scenario, and that single parameter is driving the synthetic-fluid segment's projected lead [S2].
For sensor and instrumentation integration, the same industrial controls that supervise conventional chilled-water plants — pressure transmitters, flow meters, and industrial valves on the secondary loop — are repurposed to monitor tank inlet/outlet temperatures, fluid level, and coolant flow, with loop pressure and dielectric-fluid conditioning reported to the PLC that orchestrates the heat-rejection plant. Operators planning retrofits should plan a clean instrumentation survey of the secondary loop before tank commissioning, since most first-year reliability issues trace to sensors that were never re-calibrated for dielectric-fluid service.
2026 Trackable Signals for Procurement
Two indicators will tell you whether immersion is moving from pilot to standard build: (1) hyperscale operators publicly disclosing PUE figures attributable to immersion tanks rather than to whole-facility averages, and (2) lubricant majors — Shell, Castrol, ExxonMobil — releasing tank-fluid product lines with published dielectric-strength and material-compatibility datasheets rather than bespoke OEM co-developments [S1]. Both are already partially in motion as of mid-2026, but neither has reached the level of a published, multi-vendor comparable spec sheet. Until that lands, treat fluid selection as a single-source engineering decision and lock in long-term supply and recycling commitments alongside the tank order.
Background reading: Laser Marker Price 2026: Cost Drivers, Spec Tiers, and TCO Map.