Javad VASHEGHANI FARAHANI
Modul University Vienna, Austria
E-mail: javad.vasheghani@modul.ac.at
Received: 21 Feb. 2026 /Revised: 12 Mar. 2026 /Accepted: 16 Mar. 2026
/Published: 23 Mar. 2026
______________________________________________________________________________
Abstract: In terms of thermodynamics, proof-of-work blockchain mining is comparable to an electric resistive load whose electrical input is nearly completely dissipated as low-grade heat. This characteristic makes it possible to create mining–heating setups that serve two purposes: space heating and cryptographic computing. Nevertheless, rather than relying just on gross power use, the circumstances of the marginal grid, market exposure, and enforceable operational limits determine the environment and economic performance of such systems. In this study, mining-heating systems with AI are evaluated under certain legislative and economic constraints. A Bitmain Antminer S21 Pro (3.51 kW, 234 TH/s) is simulated using physics spanning realistic room volumes (60–340 m³) in a European comfort zone of 20–23 °C. The same thermal and economic assumptions are used to benchmark four control strategies: traditional electric resistance heating, hybrid modulation, reinforcement learning (Q-learning), and bang-bang. Price-aligned operation, marginal emissions conditions, and other governance-relevant situations are evaluated in addition to operational profit and comfort criteria. Bang-bang control yielded the largest comfort-valid profit under baseline deterministic settings (€108.41 at 160 m³), reinforcement learning (€103.73), hybrid modulation (€100.24), and electric resistance heating (−€116.77 at 60 m³). In its best comfort-valid scenario, reinforcement learning maintained 99.94 % time-in-band comfort and reduced duty to 95.69 % while achieving almost maximum profit. While bang-bang control generated the greatest best-case comfort-valid profit, hybrid control obtained the maximum comfort-feasible share over the entire simulated scenario set. Building-physics characteristics had lower secondary impacts, while network hashrate (−3.124 % profit per +1 % input), power price (−2.271 %), Bitcoin price (+2.115 %), and block reward (+2.115 %) dominated economic viability, according to the local elasticity analysis. Emissions results are conditional: mining-heated only lowers net system emissions when it replaces fossil fuel-based heating and is time-aligned with low-carbon or excess renewable power. The findings demonstrate that AI-controlled mining-heating systems may operate as programmable electric loads with potential flexibility value, but their feasibility is dependent on market alignment, marginal system conditions, and enforceable governance constraints. Reinforcement learning improved operational smoothness in some scenarios under deterministic training and assessment, but cross-seed robustness remained subpar, with 91.70 % ± 11.38 % time-in-band at 160 m³ and 21 °C.
Keywords: Bitcoin mining, Dual-purpose heating, Demand-side flexibility, Marginal emissions, Energy market integration, Climate governance.
______________________________________________________________________________
