Immersion Cooling vs Air Cooling for Bitcoin Mining: Full Comparison

Why cooling method is a first-order business decision

Every joule your ASIC consumes becomes heat. The only question is how efficiently you remove that heat — and how much extra electricity you burn doing it. Power Usage Effectiveness (PUE) captures this ratio: a PUE of 1.40 means 40 % of your total site power goes to moving air around chips rather than hashing Bitcoin. At Quebec’s commercial hydro rates, that overhead is real money leaving the building every hour.

Three cooling architectures dominate the field today. Each sits at a different point on the CapEx-versus-OpEx curve, and each unlocks (or forecloses) different operational strategies — overclocking headroom, noise floor, waste-heat recovery, rack density, and hardware longevity.

Air cooling vs. single-phase immersion vs. two-phase immersion: full comparison

Criterion	Air cooling	Single-phase immersion	Two-phase immersion
PUE range (real-world)	1.20 – 1.50 (typical hashcenter); well-optimized facilities can approach 1.10 with hot-aisle containment and free-air economization	1.03 – 1.10; Green Revolution Cooling logged 1.03 at a Texas cryptocurrency mine (published 2025 case study)	1.01 – 1.05; LiquidStack quotes 1.02 pPUE for their two-phase platform at warm water temperatures (vendor datasheet, 2025)
Cooling medium	Ambient air, directed by high-speed axial fans built into the ASIC chassis	Single-phase dielectric fluid (mineral oil or engineered fluid such as Engineered Fluids BitCool or Shell Diala S4 ZX); fluid stays liquid at all operating temperatures	Low-boiling-point engineered fluid (e.g., 3M Novec 7100, boiling point ~61 °C / 142 °F); fluid vaporizes at chip surface, condenses overhead, and recirculates passively
Heat transfer mechanism	Convective air — low thermal conductivity (~0.026 W/m·K)	Forced liquid convection — thermal conductivity roughly 10–50× higher than air depending on fluid	Latent heat of vaporization — the phase change absorbs an order of magnitude more energy per unit mass than sensible-heat transfer
Chip junction temperature	85 – 100 °C at stock settings for high-density ASICs; less headroom in summer ambient or at overclocks	Typically 60 – 80 °C under load; substantially more headroom than air	Fluid boils at the surface (≈50–65 °C depending on fluid); chip temperatures are essentially clamped at boiling point regardless of load
Overclocking / underclocking headroom	Limited — thermal ceiling reached quickly on high-density machines (S21 Pro+, S21 XP)	Significant — operators routinely run ASICs 10–30 % above nominal TDP in tested single-phase setups	Maximum — two-phase is purpose-built for extreme density and sustained overclocks; limits become electrical rather than thermal
Noise floor	75 – 85 dB(A) at 1 m for a typical ASIC; unsuitable for residential or office environments	Near-silent; pump noise only (~40 dB(A)); enables home hashcenter and urban facility deployments	Near-silent; sealed tank, vapor-cycle condenser fans are the only moving parts
Hardware longevity	Dust ingestion, thermal cycling, fan bearing wear; typical ASIC fan life 20,000–40,000 hours under load	No dust, dramatically reduced thermal cycling; industry experience suggests 3–5 years of extended useful life vs. air-cooled equivalents (unverified at population scale; operator-reported)	Same advantages as single-phase; sealed environment virtually eliminates atmospheric contamination
Upfront CapEx (equipment only)	Low — no tank or fluid infrastructure; HVAC and air-containment add cost at scale	Moderate — a commercial tank for 20–30 ASICs runs approximately USD 15,000–25,000 before coolant; fluid adds roughly USD 3,000–5,000 per tank (market pricing, 2025; verify before budgeting)	High — proprietary engineered fluids are expensive; sealed-loop condensers add infrastructure cost; economical primarily above ~250 kW per deployment
Facility CapEx impact	Requires HVAC, hot/cold aisle containment, make-up air at scale; HVAC can equal or exceed hardware cost in large builds	Replaces HVAC with dry coolers or cooling towers; CapEx per MW typically lower than equivalent air-cooled build once HVAC is factored in; some operators report 30 % CapEx reduction (cited: Mining Industry Today / EIN Presswire)	Highest tank and condenser CapEx; offsets large HVAC spend; net economics favor two-phase at hyperscale density (>250 kW/rack)
Waste-heat recovery suitability	Low-grade heat (35–50 °C exhaust air); technically capturable but expensive to integrate	Good — fluid exits tank at 45–70 °C; suitable for space heating, greenhouse integration, or district-heat loops (see heat reuse for miners)	Excellent — condenser can reject heat at usable temperature; two-phase systems designed for data center waste-heat programs
Fluid compatibility risk	N/A	ASIC manufacturer warranties are typically voided for immersion-modified machines; verify before deployment; firmware mods may be needed (stock fan controllers expect fan tach signals)	Same warranty caveat; two-phase fluids are compatible with standard PCB materials but require validation for each ASIC model
Best fit	Small home miners; quick-start hashcenters; ops where capital is the constraint	Home miners seeking silence; SMB hashcenters 50 kW–5 MW; waste-heat-recovery projects; sites with noise or HVAC constraints	Industrial-scale deployments (>500 kW); extreme-density builds; operators chasing last-decimal PUE for multi-MW profitability

PUE vendor sources: LiquidStack two-phase platform datasheet (2025); Green Revolution Cooling case studies; Delta Power Solutions HPC PUE white paper. Air cooling ranges from Terahash.space PUE optimization guide and CBECI methodology documentation. All PUE figures are ranges — actual site PUE depends on climate, facility design, and load factor. Verify vendor datasheets before infrastructure decisions.

What PUE actually means for your electricity bill

PUE is not an abstract benchmark. Every tenth of a point is real kilowatt-hours purchased — or saved.

Consider a 100 kW ASIC load (roughly 35–40 modern miners running flat-out):

• Air cooling, PUE 1.40: total site draw = 140 kW; 40 kW of that goes purely to cooling overhead
• Single-phase immersion, PUE 1.07: total site draw = 107 kW; only 7 kW for cooling overhead
• Two-phase immersion, PUE 1.02: total site draw = 102 kW; 2 kW for cooling overhead

The gap between air cooling (PUE 1.40) and single-phase (PUE 1.07) is 33 kW per 100 kW of ASIC load. Over a year that is 33 kW × 8,760 hours = 289,080 kWh of pure cooling overhead.

Quebec hydro math: why these numbers hit different in Quebec

Quebec’s hydroelectric grid is one of the most competitive power markets in North America for energy-intensive operations. Hydro-Québec’s commercial Rate M (medium-power business tariff) prices the first block of consumption at approximately 5.03 ¢/kWh and additional consumption at approximately 3.73 ¢/kWh (rates effective 2025–2026 per published Hydro-Québec tariff schedules; confirm current rates at hydroquebec.com before project modelling).

Important rate-policy note: Hydro-Québec has proposed a phased rate increase specifically for large blockchain and cryptocurrency operations, targeting approximately 19.5 ¢/kWh over three years. Regulatory approval and implementation timeline were not confirmed as of June 2026 — verify with Hydro-Québec or legal counsel before committing infrastructure. Existing contracts may be grandfathered under different conditions.

Using the conservative Rate M base of ~5 ¢/kWh CAD as a modelling input:

Cooling method	PUE (midpoint)	Total site draw per 100 kW ASIC	Annual cooling overhead kWh	Annual cooling overhead cost (@ 5 ¢/kWh CAD)
Air cooling	1.35	135 kW	306,600	~CAD 15,330
Single-phase immersion	1.07	107 kW	61,320	~CAD 3,066
Two-phase immersion	1.02	102 kW	17,520	~CAD 876

Per 100 kW of ASIC load, single-phase immersion saves roughly CAD 12,264 per year in cooling overhead compared to air cooling at the Rate M base rate — before accounting for overclocking gains, longer hardware life, or waste-heat revenue offsets. Scale this linearly to your facility size. At 1 MW ASIC load, the annual overhead saving reaches roughly CAD 122,000. These are modelling figures, not guarantees; your actual PUE and tariff will differ.

For context on Quebec energy economics and the full sovereignty angle — including renewable energy, energy independence, and compute density — see our energy for compute deep-dive.

Single-phase immersion: how it works

The ASIC is fully submerged in a non-conductive (dielectric) fluid — typically a mineral-oil derivative or a purpose-engineered fluid such as Engineered Fluids BitCool or Shell Diala S4 ZX. These fluids do not boil at normal operating temperatures; they simply circulate through the tank, absorbing heat from chip surfaces, and exit through a heat exchanger (typically a dry cooler or cooling tower) where the heat is rejected before the fluid recirculates.

Key operating points:

• Fluid operating temperature typically 35–60 °C; chip junction temperatures well below air-cooled equivalents
• Fan controllers on stock ASICs expect fan tach signals — most single-phase deployments require a fan-emulator board or custom firmware to satisfy the controller
• Fluid is reusable for years; top-up is needed as trace amounts evaporate or are carried out on serviced boards
• Warranty: ASIC OEM warranties are generally voided by immersion modifications — a standard operating assumption in the industry

For a full setup walkthrough, see our immersion cooling guide.

Two-phase immersion: where physics does the work

Two-phase immersion replaces the sensible-heat transfer of single-phase with a phase-change process. The dielectric fluid — most commonly a fluorocarbon engineered fluid such as 3M Novec 7100 (boiling point approximately 61 °C / 142 °F) — boils directly at the chip surface. The latent heat of vaporization absorbs a far greater quantity of energy per unit of fluid than a temperature rise alone would.

The vapor rises to a condenser coil at the top of the sealed tank, condenses back to liquid, and falls back onto the hardware passively — no pump required for the primary cooling loop in most architectures. This is the thermodynamic elegance of two-phase: the cooling loop is self-driving.

Trade-offs:

• 3M Novec fluids carry GWP (global warming potential) values; 3M announced a phase-out of its PFAS-based Novec line; alternative two-phase fluids from vendors such as LiquidStack and others are in active development — verify current product availability before specifying
• Tank and infrastructure cost is substantially higher than single-phase
• Fluid loss through vapor leakage is a maintenance consideration in imperfectly sealed systems
• Dense packing (250 kW per enclosure demonstrated by BitFury using 3M technology) makes it viable at hyperscale; the economics are challenging below ~250 kW

Waste heat: the hidden revenue variable

All three methods generate the same amount of heat per joule consumed — the physics is identical. What differs is how usable that heat is.

• Air cooling rejects heat at 35–55 °C as exhaust air. Capturing this for space heating is possible but involves ductwork and heat recovery ventilators; efficiency is moderate.
• Single-phase immersion rejects heat at 45–70 °C via a fluid loop that can feed a heat exchanger directly integrated into building heating, greenhouse systems, or process-heat applications. This is the most commercially mature waste-heat-recovery configuration for mining.
• Two-phase immersion can reject heat at similarly useful temperatures via the condenser loop; system design determines temperature lift.

Waste-heat revenue — even modest space-heating offset — can meaningfully shift the economics of immersion’s higher CapEx. For Quebec operations, where winter heating loads are substantial, this is an underappreciated lever. See our waste heat reuse guide for a full treatment.

Decision framework: which method fits your situation

Choose air cooling if: you are starting out with 1–5 machines, capital is the primary constraint, you have adequate ventilation and noise tolerance, and you need the ability to modify or resell hardware without warranty complications.

Choose single-phase immersion if: you are operating 10+ machines in a noise-sensitive or thermally constrained space; you want to overclock sustainably; you have a waste-heat application; or you are building a 50 kW–5 MW hashcenter where HVAC would otherwise be a major CapEx line item.

Choose two-phase immersion if: you are deploying at multi-hundred-kW or MW scale; you are targeting maximum efficiency for a long-duration operational commitment; you have access to qualified facilities engineering; and the economics of proprietary fluid and sealed-tank infrastructure pencil out at your scale and power tariff.

Frequently asked questions

What is PUE and why does it matter for Bitcoin mining?

PUE (Power Usage Effectiveness) is the ratio of total site electrical consumption to the power consumed by IT equipment alone. A PUE of 1.0 would mean 100 % of electricity goes to the ASIC miners — impossible in practice, but the target. For mining specifically, every point above 1.0 represents electricity you pay for but that does not contribute to hashrate. In a margin-compressed mining environment, PUE is a direct cost-per-petahash variable.

Can I run my home ASIC miners in a single-phase immersion tank?

Yes — single-phase DIY immersion setups for 1–4 miners are well-documented and achievable for a motivated builder. The main considerations are: selecting the right dielectric fluid, installing a fan-emulator or using firmware that suppresses fan-fault errors, and building or purchasing a heat exchanger adequate for your thermal load. Budget for the tank, fluid, pump, and heat exchanger; expect a higher upfront cost than air cooling but substantially lower noise and better thermal performance.

Does immersion cooling void my ASIC warranty?

In the overwhelming majority of cases, yes — ASIC manufacturers (Bitmain, MicroBT, and others) do not support immersion-modified hardware under their standard warranties. This is an accepted trade-off in professional hashcenter operations. Factor refurbishment and repair costs accordingly. D-Central’s repair services cover immersion-modified hardware on a case-by-case basis; see ASIC repair services.

What PUE should I assume for budgeting a new hashcenter in Quebec?

For budget modelling: use 1.35–1.45 for a well-designed air-cooled facility with hot-aisle containment; 1.05–1.10 for single-phase immersion; and 1.01–1.03 for two-phase. Use the midpoints as your base case and model sensitivity. Actual PUE depends heavily on ambient climate, facility design, and load factor. Quebec’s cold winters provide a natural advantage for air economization and dry-cooler efficiency in immersion loops.

Is 3M Novec fluid still available for two-phase immersion?

3M announced in 2022 a commitment to phase out PFAS-based fluorochemical products, including the Novec line used in two-phase immersion. As of mid-2026, alternative two-phase fluids from vendors such as LiquidStack and others are commercially available or in qualification. If you are planning a two-phase deployment, verify current fluid availability and supply-chain risk with your system vendor before committing. This is an evolving area of the market.

Can waste heat from immersion cooling pay for itself in Quebec?

It depends on your specific application and whether you have a use for process heat. Space heating in Quebec’s winter is one of the most straightforward applications — the overlap between peak mining profitability seasons and peak heating demand is real. Greenhouse heating, aquaculture, and district-heat integration are more complex but have been demonstrated at commercial scale. The economics depend on your alternative heating cost (electric baseboard vs. natural gas vs. heat pump) and the thermal integration cost. See the heat reuse guide for worked examples.

What is the difference between single-phase and two-phase immersion cooling?

In single-phase immersion, the cooling fluid remains liquid throughout the entire process. Heat is absorbed by the fluid, raising its temperature, and then rejected via an external heat exchanger. In two-phase immersion, the fluid is selected to boil at chip-surface temperatures. The phase change (liquid to vapor) absorbs heat via latent heat of vaporization — a far more energy-dense mechanism. The vapor rises, condenses on a cool surface inside the sealed tank, and falls back as liquid. Two-phase achieves lower PUE and more stable chip temperatures, at higher fluid and infrastructure cost.

How does cooling method affect Bitcoin mining profitability?

Three channels: (1) PUE / electricity cost — lower PUE means more of your electricity bill goes to hashrate rather than cooling overhead; at Quebec rates even modest PUE improvement yields measurable annual savings at scale. (2) Overclocking — immersion unlocks thermal headroom to run ASICs above stock TDP, increasing hashrate per machine at the cost of higher power draw; the net effect on efficiency (J/TH) depends on tuning. (3) Hardware life — reduced thermal stress in immersion potentially extends ASIC service life, lowering capital depreciation per BTC mined. None of these effects is guaranteed; model conservatively.

Further reading

• Complete ASIC immersion cooling guide — setup, fluid selection, tank builds, common failure modes
• Waste heat reuse for miners — space heating, greenhouses, and district-heat integration
• Energy for compute — Quebec hydro, renewable power, and the economics of energy sovereignty
• Energy independence — solar, off-grid, and hybrid power strategies for miners
• ASIC repair services — D-Central’s repair services for air-cooled and immersion-modified hardware