Home Blog GreenTech and Sustainable IT Immersion cooling for high-density compute: ROI analysi...
🌱 GreenTech and Sustainable IT March 27, 2026 12 min read

Immersion cooling for high-density compute: ROI analysis

GreenTech and Sustainable IT Enterprise Guide 2026 SCALE D2C GreenTech and Sustainable IT Enterprise Guide 2026

Immersion cooling for high-density compute β€” submerging servers in dielectric fluid tanks β€” has moved from niche curiosity to serious ROI analysis for enterprises building or expanding AI GPU infrastructure in 2026. With NVIDIA H100/H200 rack densities exceeding 100kW and air cooling hitting its physical limits, immersion cooling delivers PUE as low as 1.03, 30–50% energy reduction vs air, and up to 25% higher sustained GPU performance through better thermal management. This guide covers the ROI calculation, deployment architecture, and the enterprise decision framework for immersion vs DLC.

Single-Phase vs Two-Phase Immersion

Immersion Cooling Types
Single-phase immersion: servers submerged in a dielectric mineral oil or synthetic hydrocarbon fluid. The fluid remains liquid throughout the heat transfer process β€” it circulates out of the tank to a heat exchanger (CDU) that removes heat to a facility water loop, then returns cooled. Simpler to implement, lower fluid cost. Two-phase immersion: servers submerged in a low-boiling-point dielectric fluid (3M Novec or similar). The fluid boils at the chip surface, vapour rises to a condenser lid that liquefies it and returns it to the bath. More efficient but Novec fluid has sustainability concerns and is being phased out β€” two-phase immersion is shifting to alternative fluids.

ROI Analysis: Immersion vs DLC vs Air

DimensionAir CoolingDirect Liquid CoolingSingle-Phase Immersion
PUE achievable1.4–1.81.05–1.151.03–1.05
Max rack density15–20kW80–100kW250kW+
CapEx per rackBaseline+$15,000–25,000+$40,000–80,000
GPU performance upliftBaseline+10–15%+15–25%
Server maintenanceStandardRequires draining before serviceMust remove from tank β€” fluid recovery needed
Payback periodN/A18–36 months vs air24–48 months vs air; 12–24 months vs DLC for very dense workloads
1.03
PUE achievable with single-phase immersion cooling β€” the lowest practically achievable PUE, approaching the theoretical limit of 1.0 (zero cooling overhead)
25%
GPU performance uplift from sustained lower junction temperatures with immersion cooling β€” GPUs boost clock frequencies higher and longer when thermal headroom permits, recovering performance lost to thermal throttling in air-cooled environments
250kW
Achievable rack density with single-phase immersion β€” enabling 20+ H100 systems per tank footprint, dramatically increasing compute density per square metre of data centre floor space
01
Decision Framework
When to Choose Immersion over DLC

Choose immersion over DLC when: (1) rack density exceeds 100kW (beyond DLC's practical maximum for air-residual components); (2) data centre floor space is constrained and immersion's 250kW+ density per rack justifies the higher CapEx; (3) your workload runs 24/7 at full utilisation β€” the energy savings compound fastest at continuous high utilisation; (4) you have a team experienced with fluid management (or engage a managed service). For most enterprises: start with DLC for H100 deployments, evaluate immersion when you need density above 100kW per rack or are building a new greenfield AI data centre.

100kW+ density trigger24/7 utilisationNew greenfield builds
02
Vendors
Leading Immersion Cooling Vendors

Single-phase: GRC (Green Revolution Cooling) β€” ICEtank series, most enterprise deployments; Submer β€” European market leader, good sustainability story. Two-phase (alternative fluid): LiquidStack β€” moving away from Novec to HFE alternatives. Managed service: Nautilus Data Technologies and Aligned Data Centers offer immersion as a colocation service β€” avoiding the operational complexity of self-managed immersion. Evaluate: fluid chemistry and sustainability, server compatibility (not all server form factors work), warranty implications, and service provider track record.

GRC ICEtankSubmer SmartPodManaged service option
AI Data Centre Cooling Strategy

Our DevOps and digital transformation teams advise on AI GPU infrastructure including immersion cooling ROI analysis and deployment planning. Book a free advisory session.

Frequently Asked Questions

End-to-end GreenTech and Sustainable IT strategy, implementation, and optimisation. Contact us for a free consultation.

Strategy: 4–8 weeks. Full implementation: 3–12 months.

Yes β€” D2C brands to enterprise. View our pricing.

GREENTECH AN

Ready to Implement GreenTech and Sustainable IT?

Our specialist team delivers measurable ROI for enterprise and D2C brands.

Free Audit