Cooling: Air vs Hydro vs Immersion

ASIC cooling is how a miner removes the heat its chips produce — and the method you choose sets your noise, energy efficiency, hardware lifespan, and cost. Three approaches dominate in 2026: air cooling (cheapest upfront, fine for home and small setups), hydro cooling (the new professional standard), and immersion cooling (the most efficient and quietest, but the most expensive). None is universally right. The correct choice depends on your scale, climate, capital, and how long you plan to run.

Key takeaways

• Cooling is a hashrate enabler, not just heat removal. Bitmain’s S23 air model does 318 TH/s; the hydro version of the same chip family does 580 TH/s at a record-low 9.5 J/TH — the first Bitcoin miner below 10 W/TH.
• Lifespan roughly doubles per 10°C cooler. Liquid-cooled chips run 45~60°C versus 75~85°C on air, which is why hydro and immersion fleets last years longer.
• Air is cheapest to deploy (~$50~100/kW) but loudest (~75~80 dB) and least dense; immersion is quietest (~30~40 dB) and most efficient (PUE ~1.02~1.08) but costliest.
• Liquid cooling unlocks heat-recovery revenue. 50~65°C fluid is genuinely useful — one Finnish project heats homes for 11,000+ residents from mining heat.
• Two-phase immersion has a supply problem. 3M discontinued the Novec/Fluorinert fluids that made it work; the last order date was March 2025.

Mining is the world’s strangest space-heater business: you buy a device whose entire job is to turn electricity into hashes — and, as a byproduct, into heat. Nearly 100% of the power an ASIC draws ends up as heat. A single rack of modern hydro units puts out the thermal load of a small office building. The number you don’t see in mining marketing is the cost of moving that heat away from the chips fast enough to keep them under throttle thresholds. Pick wrong and either your power bill explodes, your noise becomes a lawsuit, or your hardware dies in two years instead of seven.

Why does cooling matter more than miners realize?

Cooling is decisive for three reasons that hit revenue directly:

1. Throttling. Modern ASICs throttle hashrate when chip temperatures climb past ~85°C. Throttling means lost hashrate, which means lost revenue. A miner held at 60°C delivers full advertised hashrate; the same unit at 90°C may deliver a fraction of it.
2. Longevity. Semiconductor wear follows an Arrhenius-style relationship — as a rule of thumb, every 10°C reduction in operating temperature roughly doubles expected lifespan. A chip at 50°C can outlast the same chip at 75°C by years. Cooling investment is hardware-life investment.
3. Operational stability. Hot chips fail unpredictably. Failed hashboards mean RMA cycles, downtime, and replacement costs. A well-cooled fleet simply has fewer surprises.

Rule of thumb: keep chips cool and you’ll see both higher sustained hashrate and substantially longer hardware life than the same compute run hot. Past a certain scale, cooling pays for itself.

Air cooling — the legacy default

Air cooling has been the default since the first GPU rig: heatsinks bonded to chips, fans pushing air through them, ductwork carrying hot air outside. Simple, universal, and increasingly strained by modern hardware density.

How it works

Each Antminer or Whatsminer ships with internal fans (typically 2~4 per machine, 6,000+ RPM) pulling cool intake air across heatsinks on the ASIC chips; hot air exhausts from the rear. In a data center, rows of machines create hot/cold-aisle separation: cold air enters one side, hot air exits the other, where it’s either ducted outside (cool climates) or processed by HVAC (hot ones).

The good: lowest upfront cost (just fans and ductwork), fastest deployment (containers in days), universal ASIC compatibility, easy hands-on maintenance, and a natural fit for cold climates (Iceland, Scandinavia, Canada) that get free cooling from ambient air.

The bad: loud (75~80 dB is industrial-only — residential deployments draw noise complaints fast), heat-density limited (~5~10 kW/m² versus 50~100+ for liquid), climate-sensitive (a Texas or Dubai summer needs expensive HVAC), higher PUE (cooling adds 10~40% to energy use), and dust accumulation that demands regular fan and heatsink cleaning.

Best fit: home miners with 1~5 rigs in a garage or dedicated room; cold-climate operations with naturally low ambient air; modular containerized farms expecting a 3~5 year refresh cycle; and small devices like Bitaxe / NerdQAxe / NerdOCTAxe that are too small to benefit from liquid infrastructure.

Hydro cooling — the new professional standard

Hydro cooling moves heat from chips into a circulating liquid (usually water with corrosion inhibitors and antifreeze). Cold plates bond directly to the chips; coolant flows through channels in the plates, absorbs the heat, and carries it to an external heat exchanger or cooling tower.

How it works

A hydro miner uses internal cold plates instead of (or alongside) air heatsinks. A pump circulates roughly 8~10 L/min of coolant through them. Hot coolant exits at 50~60°C, dumps its heat at an external exchanger (to ambient air, a water source, or a downstream heat-recovery system), and returns cooler to the miner.

The good: quiet (a roughly 25 dB drop versus air is a 5~6× reduction in perceived loudness, since every 10 dB is about a doubling); high density (50+ kW/m²); lower PUE; better chip cooling (direct liquid contact holds junction temperatures 15~25°C below air, translating to multi-year lifespan gains); heat-recovery friendly (50~60°C exit coolant is genuinely useful); and a real efficiency edge — the Antminer S23 Hyd runs at 9.5 J/TH, well below the air-cooled S23’s 11 J/TH, because liquid cooling lets the chips hold higher clocks without throttling.

The bad: higher upfront cost ($200~400/kW versus $50~100 for air — for a 1 MW facility, several hundred thousand dollars more); specialized hardware (only certain models support hydro: S19 XP Hyd, S21 XP Hyd, S21 Pro Hyd, S23 series, M63, M66S Hyd); maintenance complexity (pumps, plumbing, cold plates, leak detection, coolant chemistry); three-phase power requirements; non-zero leak risk near electronics; and slower deployment (weeks to months versus days).

Best fit: industrial operations of 500 kW+ where the cost amortizes across thousands of machines; hot climates where air needs expensive HVAC anyway; heat-recovery setups; noise-sensitive locations; and long-term holdings where 5~7 year hardware life justifies the premium.

Immersion cooling — the apex approach

Immersion cooling submerges entire mining hardware in a non-conductive dielectric fluid that absorbs heat directly from every chip, component, and PCB surface at once. Hot fluid rises, gets pumped through a heat exchanger, and returns cool. There are two variants:

Single-phase immersion

The fluid stays liquid; pumps circulate it through external heat exchangers. Options range from mineral oil (cheap, low-tech) to engineered synthetic dielectrics (pricier, better thermal performance). This is the common, practical approach in 2026.

Two-phase immersion

The fluid boils at 56~61°C; the phase change absorbs enormous heat via latent heat of vaporization. Vapor rises, condenses on cooled lids, and drips back — no circulation pumps needed. It offers the highest thermal performance but the highest cost, and it now faces a serious supply problem (see below).

The good: best thermal performance (chip temps 30~40°C below air, maximum overclock headroom); near-silent (30~40 dB); lowest PUE (1.02~1.08, so almost all energy goes to mining — recovery-focused immersion containers have reported PUE near 1.02); maximum density; dust-proof sealed hardware; longest hardware life (immersion-cooled miners commonly last 2~3× longer than air-cooled units); and climate independence.

The bad: highest upfront cost ($300~600/kW plus the fluid itself); frequent hardware modification (fans removed, sometimes warranty-voiding retrofits — though models like the S23 Immersion are purpose-built); fluid degradation and periodic quality testing; messy maintenance (pulling an oil-coated miner for repair); spill/environmental risk with hundreds of liters of fluid; two-phase fluid scarcity; and slow deployment (months).

The fluid supply problem (important for 2026). Two-phase immersion depended heavily on 3M’s Novec and Fluorinert fluids. In December 2022, 3M announced it would exit all PFAS manufacturing, and per DataCenterDynamics the phase-out was tied to mounting PFAS “forever chemical” regulation and a multibillion-dollar water-contamination settlement. The last order date for Novec was March 31, 2025, with production ceasing by year-end. Replacements (from Solvay, Chemours and others) exist but aren’t yet at volume, and many are themselves PFAS. Practical fluid pricing in 2026: engineered single-phase fluids run roughly $150~400 per gallon, while mineral oil is far cheaper at ~$20~40 per gallon. If you’re planning two-phase immersion today, fluid sourcing is the first thing to solve, not the last.

Best fit: maximum-density operations where space is constrained and capital isn’t; hot, dusty, or harsh environments; long-horizon holdings (5~10 years); aggressive overclocking (immersion can sustain ~1.3× nameplate); and noise-critical or stealth sites.

Same chip, different cooling: the S23 comparison

Bitmain’s S23 line is a clean controlled experiment — the same chip generation across air, immersion, and hydro variants, so the only thing that changes is heat removal:

Same silicon, hashrate ranging from 318 to 1,160 TH/s. The S23 Hyd 3U delivers 3.6× the air model’s hashrate in a comparable footprint — entirely because cooling lets the design pack more chips and clock them higher without throttling. The lesson: at the modern hardware tier, cooling is no longer just a heat problem. It’s a hashrate enabler — and the S23 Hyd’s 9.5 J/TH is the first sub-10 W/TH efficiency Bitmain has shipped, a level air cooling cannot reach.

1 MW comparison: hashrate and cost by cooling

Compare what one megawatt of ASIC power buys, by cooling method. Hashrate per megawatt is governed by efficiency (J/TH) — and the best efficiencies only exist in liquid-cooled models:

The headline: the same megawatt produces ~57 PH/s on air-cooled S21 hardware versus ~105 PH/s on the most efficient hydro hardware — roughly 85% more hashrate from the same power draw, because liquid cooling is what lets the chips run at 9.5 J/TH without throttling. Add lower PUE (less power wasted on cooling) and longer hardware life, and the higher upfront cost of liquid cooling is recovered over a multi-year horizon. For a 1 MW build, the economics now favor hydro or immersion; air remains competitive mainly at smaller scale (under ~100 kW) or in cold climates with cheap power. (Exact dollar outcomes depend on hardware prices, power cost, and Bitcoin price — model your own case in the SoloFury profitability calculator.)

Hardware support matrix (2026)

Not every ASIC supports every method:

Some manufacturers ship immersion-only models — no fans, sealed packaging optimized for a fluid bath. These won’t run on a desk; they require a tank.

Can you turn waste heat into income?

Yes — and it’s one of the main reasons operations move to liquid cooling. Because nearly 100% of an ASIC’s power becomes heat, hydro and immersion produce hot fluid at 50~65°C that is genuinely useful, not just thermal pollution. Real deployments in 2026:

• District heating. A Finnish project channels mining heat into a municipal network warming homes for 11,000+ residents, displacing fossil-fuel heating, as documented in industry coverage of heat-reuse projects.
• Greenhouses. A 3 MW pilot in Manitoba, Canada captures around 90% of server electricity as heat (75°C+) to preheat water for a tomato greenhouse, cutting its heating costs by up to 55%.
• Pools and spas. Vancouver’s MintGreen heats a public swimming pool and commercial buildings with recovered mining heat.
• Industrial drying and aquaculture. Wood, grain, and fish drying, plus temperature-controlled fish farms, all use low-grade mining heat that’s effectively “free” once you’ve paid for the electricity.
• Home heating. If you’d be heating the space anyway, a home ASIC’s heat offsets your heating bill — the electricity does double duty.

Heat recovery only works well with hydro or immersion: air-cooled exhaust at 30~40°C is too low-grade for most uses. The shift to liquid cooling is partly driven by this second revenue stream.

How does climate change the answer?

Cold (Iceland, Norway, Canada, Russia): air cooling shines — free natural cooling drives HVAC needs near zero. Liquid still helps with noise and density, but the air advantage is strong, so most cold-climate operations stay air-cooled.

Temperate (Europe, US Northeast, Northern China): mixed. Air is viable in cooler months but expensive in summer; hydro is becoming the default for new installs; immersion appears at commercial scale.

Hot (Texas, UAE, North Africa, SE Asia): air alone is impractical without heavy HVAC. Free natural cooling is impossible above ~30°C ambient, so hydro is the practical floor and immersion is increasingly standard for new builds.

Indoor / urban / regulated: noise rules effectively mandate liquid. Many municipalities cap operations at >55 dB at the property line, which rules out air-cooled mining in residential or mixed-use areas.

What about home and solo miners?

For 1~5 ASICs at home or in a small office, air cooling almost always wins: the cost of hydro infrastructure (often $2,000+ for a small setup) doesn’t pay back at that scale, noise can be managed with a basement/garage/shed, heat output is modest, and maintenance stays simple (no fluid handling). Hydro or immersion is only worth it at home if you face strict indoor noise limits, a very hot ambient, plans to scale to 10+ rigs, or a concrete heat-recovery use (pool, greenhouse, workshop).

Small devices like Bitaxe, NerdQAxe and NerdOCTAxe were designed for desktop air cooling and run cool by mining standards — no special cooling needed. Some hobbyists submerge a Bitaxe in a mineral-oil tank for the look, but the practical benefit is minimal.

For SoloFury miners specifically, cooling rarely makes or breaks the operation. A single S21+ at home mines BCH equally well in air or hydro — solo outcomes are dominated by network probability, not chip temperature (the math behind that is in our guide to mining variance and Poisson math). Cooling matters most when you run fleets, when margins are thin, or when hardware needs to last seven years instead of three. Whatever cooling you choose, the network treats every hash equally.

Operational factors beyond raw cooling

Maintenance. Air: fan replacement, dust cleaning, ductwork checks — any ASIC tech can service it. Hydro: pump inspection, cold-plate cleaning, coolant-chemistry monitoring, leak detection. Immersion: fluid quality testing every 6~12 months, sealed-component procedures, and fluid-handling compliance.

Insurance. Air uses standard data-center coverage; hydro may need water-damage riders; immersion can carry environmental-liability exposure for fluid spills and often requires specialized coverage.

Resale value. Air-cooled ASICs have a universal resale market. Hydro variants sell to a smaller industrial pool (often a 10~20% discount). Immersion-modified units have the narrowest market (20~30% discount), and some immersion-native models have essentially no secondary market.

The decision framework

For a new operation, answer in order:

1. How big? Under 100 kW: air almost always. 100 kW~1 MW: hydro becomes interesting. 1 MW+: hydro or immersion for competitive economics.
2. What’s the climate? Cold (avg <15°C): air viable, hydro for density. Temperate (15~25°C): hydro recommended. Hot (>25°C avg): hydro mandatory, immersion preferred.
3. How long will you run? Under 3 years: air. 3~7 years: hydro. 7+ years: immersion.
4. Can you sell the heat? No: air or hydro. Yes (district heating, greenhouse): hydro or immersion.
5. How much upfront capital? Constrained: air. Moderate: hydro. Abundant: immersion.

Don’t over-engineer and don’t under-engineer. The economics work out if you size cooling to your scale, climate, capital, and time horizon — then point the hardware at the chains you want to mine with the setup wizard.

The bottom line

Mining is a heat business as much as a hash business: the SHA-256 computation is exactly what turns electrical energy into thermal energy, so cooling is the discipline of making sure that heat doesn’t bottleneck your hashrate. Air is the past — still right for small operators and cold climates, but strained by modern density. Hydro is the present — fast becoming the default for serious operations, with the best shipping efficiency (9.5 J/TH). Immersion is the future — most expensive, most efficient, quietest, and best for heat recovery, though its two-phase variant must navigate the Novec supply gap.

For solo miners, cooling rarely decides the outcome at home scale; for fleets and thin margins, it pays for itself many times over. Match the cooling to your situation, size it correctly, and the rest of the math takes care of itself.