Data centers (DC) are synonymous with high energy consumption and heat dissipation. These infrastructures are practically heat factories, leaving a high carbon footprint. As the popular parlance goes, “with great power comes great responsibility.”

Quite literally, the responsibility, herein, is cooling and energy efficiency owing to the recent advances in technology, including the high demand for AI-powered systems. Consequently, these advancements have fuelled the demand for a more efficient power and cooling approach, alongside reducing the Power Usage Effectiveness (PUE) within DCs.

This industry piece was written by Solomon Obadimu & Jones Udo-akang

Interestingly, while more efficient cooling strategies have long been the prevalent area of discussion amongst data center professionals, that mindset is gradually shifting towards combining cooling with recycled waste heat from these heat factories - data centers.

This “rejected stone,” waste heat - an untapped resource - could become the “cornerstone” of energy efficiency in this ever-growing industry. Technologies such as absorption and adsorption chillers, desiccant systems for dehumidification, and the Organic Rankine Cycle (ORC) for power generation have been proposed as potential solutions to the aforementioned heat problem, reducing operational cost and lowering PUE.

How significant is the impact of this shift from theoretical triage to actual implementation? Can we rely on these proposed technologies to lower PUE in this ever-growing industry? Will the increasing complexity of DC infrastructure accommodate these technologies? Can DC waste heat be potentially used to power its own systems? This article addresses these questions.

It also aims to briefly introduce the technologies mentioned above by exploring and drawing from recent key research-based studies within the last five years. To further answer these questions, this article focuses exclusively on experimental and simulation-based research studies that investigate the potential reuse of waste heat generated by DCs internally, rather than externally.

Data Center Internal Waste Heat Reuse Pathways

1.1 Cooling Pathways

Absorption/Adsorption Chillers for DC Cooling

Gupta and Puri (2021) conducted a simulation-focused feasibility study on internally reusing waste heat from water-cooled racks within data centers (DC) by integrating a silica gel-water adsorption chiller. The proposed system utilises hot water discharged from liquid-cooled racks to drive the adsorption chiller.

This chiller then produces chilled water for in-row cooling units serving nearby air-cooled racks. By reducing reliance on conventional vapour-compression chillers, this approach offers energy, environmental, and economic benefits.

Overall, following sensitivity analysis, inlet water temperatures, airflow, load sharing factor, and chilled water setpoints were identified as the primary parameters that significantly affect the performance of the proposed adsorption chiller system. By integrating both air- and water-cooled systems, the study demonstrates significant energy savings efficiency (ESE) improvements of up to 22.5 percent and annual CO₂ emissions reductions of 104 tons.

However, a key operational trade-off was identified: operators must choose between maximising energy savings or maximising computational power. Running the system with lower-grade waste heat (e.g., 40°C inlet water) optimizes chiller efficiency and ESE, but forces servers to run hotter.

Conversely, using higher-grade heat (e.g., 65°C) improves chiller drive temperature but can degrade server computing performance (GFLOPs/W) by up to six percent. The authors report that potential investors can expect to see a return on investment in a little over a year (379 days) for a 30kW adsorption chiller and under a year (285 days) for a 90 kW chiller.

In a similar vein, Amiri et al. (2021) built upon the work of Gupta and Puri (2021). The authors simulated the potential applications of absorption chillers in data centers compared to the adsorption chiller modelled by Gupta and Puri. The absorption chiller requires a liquid salt solution - lithium bromide (LiBr) as its absorbent, while the adsorption chiller employed by Gupta and Puri (2021) uses a solid sorbent such as silica gel.

While both the aforementioned chillers use water as their refrigerant, the key difference lies in the heat grade required. The absorption chiller requires a relatively high-grade waste heat, while the adsorption chiller can operate with a lower-temperature waste heat.

When modelled for a data center, the study by Amiri et al. (2021) revealed that a 4.5MW and 13.5MW capacity data center could achieve energy savings of 4.3 to 13.0 GWh annually. This reduction in energy consumption was accompanied by a significant decrease in CO₂ emissions, with an annual reduction of 3,068 to 9,208 tons. The study also found that the use of absorption chillers reduced the workload of conventional compression chillers, contributing to overall energy efficiency. Economically, the system could potentially achieve payback periods of 2.56 to 2.76 years, with performance improving with the scale of the DC.

Very recently, Cui et al. (2025) proposed a waste heat-driven cooling system for data centers. This system uses a triple-stage LiBr-H₂O absorption chiller designed to operate with ultra-low-grade waste heat at 50°C. The system captures heat from the liquid-cooling loop and uses it to cool air-cooled components, reducing the dependence on conventional chillers. A novel stage-switching feature allows the chiller to maximize its cooling capacity from a limited waste heat source by continuously adjusting its operational modes as cooling water temperature decreases.

To demonstrate the system’s effectiveness, Cui et al. (2025) used the case study of a DC with 200 kW liquid-cooling and 100 kW air-cooling capacity. The proposed chiller achieved an 80.5% reduction in mechanical cooling load and an annual electricity savings of 78 MWh. Overall, the authors noted that the solution would be particularly effective in hot, humid climates and is compatible with existing free cooling strategies.

1.2 Humidity Control Pathways

Desiccant-Based Dehumidification for DCs

In an interesting experimental study conducted by Huang et al. (2025), desiccant-coated heat exchangers (DCHEs) were used to dehumidify a liquid-cooled DC. According to the authors, the system achieves continuous dehumidification by alternating between two sets of DCHEs, driven by 50°C waste heat water recovered from the liquid cooling loop and 15°C ambient cool air. Following a series of experiments, the authors reported approximately 42% heat recovery efficiency at 4500 m³/h airflow and an average coefficient of performance (COPd) of up to 8.85 at 2,700 m³/h airflow.

Moreover, they observed an increase in dehumidification COP and a decrease in waste heat recovery efficiency with a reduction in airflow and vice versa. Overall, they found the proposed DCHE to be 1.36 times more energy efficient than a conventional dehumidification system, reporting a moisture removal rate of 7.07 kg/kWh of electricity consumed.

Threading the same path of DC dehumidification using waste heat, Okposio (2020) established the possibility of using desiccant-based direct evaporative cooling systems within data centers following experimental analysis. This system absorbs moisture from the DC-cooled air and then regenerates the desiccant using warm return air (waste heat) from the DC. The authors proposed and tested a condensate

recovery system to capture water desorbed from the desiccant material, demonstrating its potential reuse in the evaporative cooling process to reduce overall water consumption in DCs. Experimental results showed a correlation between pad depth (thickness) and evaporative cooling, with the latter improving up to 81 percent as pad thickness increased. The proposed condensate recovery method was found to recover an average of 50 percent of lost moisture. Overall, the author reported a significant improvement in PUE owing to the system’s operational efficiency.

1.3 Power Recovery Pathways

Organic Rankine Cycle (ORC)

Data centers are heat factories. However, the main focus has been on solving their cooling challenge, rather than viewing the heat itself as a resource. The Organic Rankine Cycle (ORC) has been identified as a potential solution to harness this low-grade server heat and put it back to work as power for the data center’s own use.

Ancona et al. (2022) conducted an investigation to assess the feasibility of potentially integrating micro ORCs into DC cooling systems. Employing a 3 kW-size ORC test rig, the authors varied and analysed: thermal input power, expander output power, net power production, and the overall efficiency. Experimental analysis revealed a correlation between heat source temperature and the thermal input power of the ORC system. Numerical analysis revealed that as the heat source temperature increased from 40°C to 55°C, thermal power rose from 10 kW to over 20 kW, consequently increasing the ORC mass flow rate— a trend consistent with experimental results.

The flow rate of the heat source, however, was found to have a negligible effect on input power. Regarding expander output power, the authors found that both the heat source temperature and ORC mass flow rate had a positive impact on the expander power, with power increasing from 200W to 550W. Furthermore, net power production was found to increase as heat source temperature (DC waste heat) increased, while a negative correlation was observed for cold sink temperature (representing ambient air temperature). Overall, using numerical modelling, the ORC system was found to be the most efficient in converting DC waste heat to electricity when R1234ze(E) refrigerant was employed, indicating that the ORC system’s performance is fluid-dependent.

Marshall and Duquette (2022) built upon the work of Ancona et al. (2022) by conducting a system-level modelling for the potential integration of ORC systems into DC operations. Their proposed model, a refrigerant-based heat pump-assisted Organic Rankine Cycle (HPA-ORC), advances the basic ORC concept proposed by Ancona et al.

Compared to Ancona et al.’s configuration, the HPA-ORC system collects waste heat from DC servers, and the integrated heat pump then upgrades this low-grade heat, thereby making it useful. At this stage, the captured heat is delivered to the ORC system—a small power plant - where a refrigerant fluid boils at a relatively low temperature, creating a high-pressure vapour to spin and expand through a turbine to produce electricity. The vapour is then cooled back to repeat the same cycle.

The authors modelled a 1,000-server data center, tested various refrigerants, and compared the proposed HPA-ORC system with an air-source heat pump (ASHP) chiller system to evaluate the potential economic benefits of ORC systems. Modelling results confirm a correlation between refrigerant fluid and HPA-ORC performance, with R161 and pentane reported as the top performers.

As expected, the HPA ORC system was found to outperform its ASHP counterpart. Additionally, when operated under the same conditions, the HPA-ORC system was found to be more economically viable, with profitability rising with an increase in server utilization. Despite the higher initial capital and maintenance costs, the study concludes that large-scale implementation promises considerable economic benefits.

Comparison and Insights

From the reviewed literature, it is apparent that the potential utilisation of data center waste heat has shifted from theoretical triage of heat conversion from server racks to the implementation of practical systems engineering—shifting from theory to action. Much more attention is being paid to reusing waste heat within DCs, rather than externally.

The body of work reviewed spans different technological approaches, ranging from absorption/adsorption chillers and desiccant-based dehumidification systems to power generation through ORC systems. Collectively, these studies show that server-generated waste heat is not just a byproduct; it is recoverable and can be put to good use for cooling and power generation, promoting energy efficiency within DCs.

Regarding heat reuse for cooling within DCs, the multi-stage absorption chiller proposed by Cui et al. (2025) appears to be quite sophisticated and impressive owing to its ultra-low grade heat requirement (as low as 50°C) and its multi-stage process. Although its operating principle is similar to that of the adsorption chiller proposed by Gupta and Puri (2021), which utilises heat from the liquid cooling loop, the model could potentially be deployed in high-temperature and high-humidity regions. Overall, these studies have shown the feasibility of thermal reuse within DCs and potentially lowering cooling load as well as improving PUE via the proposed methods.

Beyond cooling, the management of relative humidity has been identified as a key area for efficiency gains. The use of desiccant-coated heat exchangers (DCHEs) has been validated at a practical level to be a viable solution to dehumidification issues within DCs.

The system proposed by Okposio can simultaneously remove adsorption heat and dehumidify the air, placing less demand on the chillers and saving energy, whilst improving DC PUE. The authors noted that striking the right balance between airflow rate and fan speed will further optimise the system. Collectively, these studies show that reusing data center waste heat for cooling and humidity control is no longer a theoretical concept—it is becoming a real-world solution with strong economic and environmental benefits.

In contrast to the direct thermal reuse discussed above, the Organic Rankine Cycle (ORC) represents a different approach. In this case, the same low-grade heat employed for cooling and dehumidification could also be directly converted to electricity. In simple terms: server heat boils a refrigerant → expanding vapour spins a turbine → electricity is produced. More importantly, the electricity generated is not theoretical; it can be used within DCs to power some systems, including control valves, fans, building management system (BMS) sensors, and ancillary loads. However, when scaled across multiple racks, the collective energy recovery could potentially reduce a data center's PUE by partially self-powering its own thermal management subsystems, although efficient energy conversion from heat to power remains a key challenge.

In summary, three key insights have been identified from the body of work reviewed:

• Temperature range dictates technological viability: Every technology proposed has a minimum temperature requirement to operate efficiently. Absorption and adsorption chillers can be operated at moderate temperatures (50–70°C). Desiccant systems, on the other hand, thrive at lower temperatures (35–55°C), while ORCs can operate across a broad low-to-medium temperature range.
• Viable integration opportunities within data centers: A hybrid of the aforementioned systems within a data center will offset electrical demand in all aspects, yielding compounding efficiency gains over time.
• Economic and environmental potential: The internal reuse of waste heat for cooling and dehumidification will reduce over-reliance on primary chillers and decrease water consumption, as demonstrated by the proposed condensate recovery system (explained above). This will improve overall energy circulation, where recovered heat offsets both electrical and cooling loads.

Conclusion

Once seen as a burden, the rejected stone—waste heat—is now an operational asset. The reviewed studies provide a blueprint for using it internally to directly lower PUE and build a more circular, sustainable data center energy model.

As data centers continue to evolve into heat factories, those that successfully transform this 'rejected stone' into an 'operational asset' will lead the industry towards both economic and environmental sustainability.