Risk-informed multi-objective techno-economic assessment of air-to-liquid retrofits with heat recovery in legacy air-cooling data centers

Driven by the rapid growth of generative AI, many medium-sized legacy air-cooled data centers are facing increasing power density, rising cooling energy demand, and growing cooling-water pressure. However, full replacement is often impractical because of existing infrastructure, electrical-capacity limits, and business-continuity requirements. Under current low-carbon development goals and resource-oriented waste-heat utilization policies, heat recovery has become an important dimension in retrofit assessment. Existing studies mainly rely on deterministic or mean-based comparisons and rarely provide a tail-risk-oriented framework for retrofit decision-making. To address this gap, this study develops a multi-scale techno-economic assessment framework for 1–4 MW legacy air-cooled data centers. The framework couples cold-plate liquid cooling, room-level cooling systems, cooling towers, and heat recovery within a unified model. It further integrates triangular Monte Carlo simulation, VaR/CVaR, random-forest surrogate models, and multi-objective screening. Retrofit options are evaluated under three representative climates, namely Shenzhen, Nanjing, and Urumqi. The results show that the preferred portfolios reduce payback-period tail risk by about 1.69–4.19 years relative to the baseline, whereas the tail risk of PUE changes only marginally. This suggests that retrofit decisions should not focus solely on lowering PUE, but should also consider economic robustness, water-side benefits, and engineering feasibility. Overall, the proposed framework moves beyond static performance comparison and provides quantitative support for staged upgrades and robust retrofit screening under different urban conditions.