Recovering Heat from Bitcoin Miners and Data-Center Chips

Executive summary

The idea is directionally correct: miners and AI/HPC chips turn expensive electricity into heat, and that heat can sometimes be monetized. The mistake is trying to use a normal Wankel engine as the recovery machine.

A Wankel engine is a compact rotary internal-combustion engine. It normally makes power by burning fuel inside rotating combustion chambers, rather than by passively absorbing low-temperature heat from chips. For chip waste heat, the closer concept would be a closed-cycle rotary expander inside an Organic Rankine Cycle, not a gasoline-style Wankel engine. [1]

The hard limit is not mechanical cleverness; it is temperature. Bitcoin miners, GPUs, CPUs, and ASICs usually produce low-grade heat. With a hot loop around 40–80°C and a cold sink around 25°C, even a perfect heat engine has limited headroom. Real systems are much lower than the theoretical ceiling.

2. The physics limit: low-temperature heat cannot make much electricity

Every heat engine needs a hot side and a cold side. The bigger the temperature gap, the more work can be extracted. The idealized upper bound is the Carnot limit:

NASA’s educational material frames the Carnot cycle as a fundamental thermodynamic engine cycle; this report uses the standard Carnot efficiency relation to show why chip heat is difficult to turn back into electricity. [2]

Assuming a 25°C cold side, the theoretical maximum looks like this:

These are theoretical ceilings. Real systems lose energy in heat exchangers, pumps, expanders/turbines, generators, controls, and imperfect heat transfer.

3. What temperatures are realistic?

Bitcoin mining heat is often useful but low-grade. K33 describes common miner heat as roughly 40–50°C, with the possibility of running as hot as 80°C, and notes that immersion cooling makes heat easier to recover than air cooling. [6]

A cryptocurrency-mining liquid/spray-cooling prototype described by IntechOpen heated dielectric coolant through miner hashboards and sent it through a coil in a 190 L water tank. The authors reported maximum coolant temperature reaching 70°C in field testing while staying within safe chip limits. [7]

For AI/HPC data centers, the trend toward liquid cooling matters because it concentrates heat into water loops instead of diluted exhaust air. ASHRAE’s AI data-center framework notes that liquid cooling and higher-temperature GPU cold-plate loops create a path toward warm-water heat recovery or district-scale reuse. [8]

4. Why air cooling is the weaker path

Air carries far less heat per unit volume than liquids, so air-cooled heat is bulky, noisy, and harder to move into a useful secondary system. It can still heat a room, shop, greenhouse, or ducted space, but it is a poor input for a generator.

Liquid cooling changes the economics because it can deliver a more stable heat stream to a heat exchanger. The hotter the return water, the more valuable the heat becomes. This is why direct-to-chip, immersion, spray cooling, and hydro-cooled miners are more interesting than standard fan exhaust.

6. Architecture A: heat-to-electricity recovery

This is the version closest to your original idea. A Wankel-like rotary part could only make sense as the expander, not as a conventional combustion engine.

7. Architecture B: direct heat reuse

This is the version most likely to save real money.

8. Heat-to-electricity savings by scale

The following uses a simple assumption: the compute/mining load runs 24/7, electricity costs $0.13/kWh, and the recovered electricity directly offsets purchased power. The 2% and 4% cases are more realistic for low-grade heat; the 8% case is a stretch case requiring hotter loops and good design.

Interpretation: a single 3 kW miner at 4% recovery only produces about 120 W of electricity and saves about $137/year at $0.13/kWh. That is unlikely to pay for a generator, pumps, heat exchangers, controls, and maintenance. A 1 MW site is different: 4% recovery is 40 kW and about $45,552/year at the same power price.

9. Direct heat value for a 3 kW miner

If a 3 kW miner is used as heat, each hour of operation produces 3 kWh of thermal energy. The table below shows avoided electric-heating value at $0.13/kWh.

COP 3 means a heat pump would have delivered about 3 kWh of heat per 1 kWh of electricity. So miner heat is less financially valuable when it replaces a good heat pump than when it replaces electric resistance heat.

10. What makes the payback work?

• High heat utilization: you need a real heat load during the hours the machines run.
• Liquid capture: direct-to-chip, immersion, hydro, or spray cooling improves heat quality.
• Low parasitic load: pumps, fans, dry coolers, and controls must not eat the savings.
• Minimal downtime risk: a heat-recovery failure should bypass safely without taking miners or servers offline.
• Local heat value: cold climates, pools, domestic hot water, greenhouses, and process loads help most.

12. Why data centers care now

Global data-center electricity consumption was estimated by the IEA at about 415 TWh in 2024, roughly 1.5% of global electricity consumption, with a base-case projection near 945 TWh by 2030. [9]

This growth makes even small percentage improvements valuable. Cooling and environmental control can range from about 7% of total data-center electricity use in efficient hyperscale facilities to over 30% in less-efficient enterprise data centers. [9]

13. Heat networks are often better than generators

Thermal energy networks and district heating systems treat data-center heat as a product instead of a disposal problem. The advantage is that the heat stays heat, avoiding the conversion penalty of turning low-grade heat back into electricity. EESI notes that data-center heat can be transferred through water-pipe networks to buildings and facilities that need heat, and also highlights barriers: connecting producers to heat users, upfront cost, financial risk, and policy support. [10]

For Bitcoin miners, the same logic applies at smaller scale: a miner in a basement, garage, greenhouse, shop, or water-heater loop can be more economically useful than a miner venting hot air outside.

15. Best MVP path

For a practical build or prototype, skip the Wankel engine and build a compute boiler or miner heat module.

• Start with a 1–3 kW miner or compute box in a controlled loop.
• Use immersion, hydro, spray, or direct-to-chip cooling to capture heat in liquid.
• Send the hot loop through a plate or coil heat exchanger into a buffer tank.
• Use the tank for domestic hot-water preheat, garage/shop hydronic heat, greenhouse heat, or pool preheat.
• Add sensors for watts, flow rate, inlet/outlet temperature, tank temperature, ambient temperature, and runtime.
• Publish a dashboard showing compute revenue, heat recovered, avoided heating cost, and net cost per kWh of useful heat.

Appendix A: formulas used

For the tables, electricity price is assumed to be $0.13/kWh and continuous operation is assumed where stated.

Appendix B: decision checklist

• Is the heat captured in liquid instead of loose exhaust air?
• Is the hot-side temperature measured above 60°C?
• Is there a real year-round or seasonal heat customer?
• Can the system bypass safely without downtime?
• Do parasitic pump/fan loads stay small?
• Does recovered heat replace electric resistance, gas, oil, propane, or heat-pump energy?
• Is the site at least hundreds of kW before considering heat-to-electricity machinery?
• Is the maintenance burden acceptable?