Data Centre Waste Heat for Building Heating: A Comparative Energy Analysis in Italy

Abstract

The decarbonisation of the building sector represents a key challenge for the European energy transition, particularly in the heating segment, which is still largely dependent on fossil fuels. In this context, data centres (DCs) offer a promising opportunity as local sources of recoverable waste heat. This study investigates the use of data centre waste heat for building heating through a comparative annual energy analysis applied to two building typologies in a Mediterranean climate (Italy): a residential building and a school. Three scenarios are considered: non-integrated scenario S0 (data centre with its own cooling system and buildings with gas-fired boilers), non-integrated scenario S1 (data centre with its own cooling system and buildings with air-to-water heat pumps), and integrated scenario S2 (data centre cooling system coupled with the buildings through waste heat recovery and heat pump technology). A theoretical 300 kW data centre was considered as the waste heat source. The integrated scenario significantly improves system performance. In the residential case, the seasonal COP increases from 2.15 to 4.50, reducing electricity consumption from 289.5 MWh to 128.9 MWh. In the school case, the COP increases from 2.51 to 8.00, with electricity consumption decreasing from 161.3 MWh to 49.1 MWh. These improvements lead to reductions in non-renewable primary energy demand of up to 63% and 79% for the residential and school buildings, respectively, compared to the baseline scenario. The results demonstrate that data centres can act as decentralised thermal sources, supporting the transition towards low-carbon and Nearly Zero-Energy Buildings.

1. Introduction

1.1. European Targets of Decarbonisation and the Role of the Building Sector

The European Union is at a critical stage of the energy transition. This process is driven by the objective of achieving climate neutrality by 2050, as defined in the European Green Deal [1]. To make this target effectively operative, the Fit for 55 legislative package established a binding requirement to reduce net greenhouse gas emissions by at least 55% by 2030 compared to 1990 levels.

The building sector is identified as a key pillar for the success of this strategy [1,2]. Buildings currently account for approximately 40% of total energy consumption in the Union and for 36% of energy-related greenhouse gas emissions [3,4,5]. The challenge is further intensified by the fact that around 75% of the existing building stock shows poor energy performance.

Particular attention must be given to heating and cooling demand, which is about half of final energy consumption in Europe. Despite recent progress, more than 70% of this demand is still covered by fossil fuels, mainly natural gas [2,6]. The decarbonisation of this segment requires a combination of technological innovation and behavioural change. Among the most promising solutions, there is the electrification of heat through heat pumps. The development of next-generation district heating and cooling (DHC) networks also plays a central role [7,8,9,10,11].

Technological innovation is enabling the integration of local and sustainable energy sources, such as waste heat from industrial processes and data centres. At present, this resource remains largely unused despite its significant potential [12,13]. In addition, the adoption of zero-carbon-ready building standards and the “energy efficiency first” principle has become a regulatory requirement. These measures ensure that all infrastructure investments are consistent with long-term climate objectives.

In summary, the European building transition is not only an environmental necessity. It also represents an opportunity to improve citizens’ quality of life, reduce energy dependence, and enhance economic competitiveness through the deployment of clean and integrated technologies.

1.2. The Building Heating Sector

As previously stated, the European building sector is one of the main energy consumers. In response to this critical issue, the European Union introduced the concept of Nearly Zero-Energy Buildings (NZEB) through Directive 2010/31/EU (EPBD) [14], requiring all new buildings to achieve very high energy performance, with an almost zero energy demand largely covered by local renewable sources. However, the transition towards the NZEB standard faces its most critical challenges in the heating sector, which alone accounts for about 69% of total household energy consumption [15].

The main challenges related to heating include: the obsolescence of the building stock, the dependence on fossil fuels, economic and technical barriers (highly efficient technologies such as heat pumps involve high initial investment costs and technical complexity when integrated into existing distribution systems, especially in colder climates where the seasonal performance factor can vary significantly) [16], regional disparities (the selection of the optimal heating system is strongly influenced by infrastructure availability, such as district heating networks, and fuel costs). Among the technologies currently considered most promising for heating decarbonisation, heat pumps play a central role due to their high efficiency and compatibility with the progressive electrification of energy systems. In parallel, solar-based heating technologies, including solar thermal collectors and photovoltaic-assisted systems, can contribute to reducing fossil fuel consumption.

Thus, it is vital to analyse solutions that integrate passive strategies with advanced generation systems to ensure thermal comfort in an economically sustainable manner.

1.3. The Data Centre Sector: Energy Challenge and Opportunity

Over the past two decades, the data centre (DC) sector has experienced rapid growth in the number of installations, computing capacity, and rack power density. The increasing demand for digital services, cloud computing, big data processing, and artificial intelligence has driven this expansion. As a consequence, electricity consumption associated with data centres has risen substantially. Current estimates indicate that data centres account for approximately 1–2% of global electricity consumption, with a continuing upward trend [17,18,19,20]. Modern high-density servers have high thermal design power, with rack power densities commonly exceeding 20–30 kW and potentially reaching 50 kW per rack in future scenarios [21].

The combination of increasing computational loads, continuous 24/7 operation, and high equipment concentration results in a substantial rise in cooling demand. This is primarily due to the intrinsic characteristic of data centres: almost all electrical input is instantaneously converted into heat, with a 1:1 ratio. Consequently, cooling systems play a central role in overall energy performance. Several studies report that cooling can account for up to 40–50% of total data centre electricity consumption when thermal management is inefficient [20,22,23,24].

Beyond increasing electrical demand, the growing cooling requirement leads to a larger availability of low-temperature waste heat. Heat rejected by data centre cooling systems represents a thermal resource that is currently largely underutilised, yet it holds significant potential for recovery and external use. Increasing attention to energy efficiency, decarbonisation, and circular economy principles has promoted a shift in perspective: data centres are no longer viewed solely as energy-intensive loads or contributors to the so-called “urban heat island effect”, but also as potential thermal energy sources.

Depending on the cooling technology adopted, the temperature of recoverable waste heat typically ranges from 25 to 35 °C (or higher) in air-based systems and can reach 50–60 °C in liquid cooling configurations [25,26].

Recent contributions have explored the role of data centres within integrated energy systems, analysing their interaction with building energy demand and highlighting their potential as active components in smart energy networks [27,28].

1.4. Waste Heat Recovery from Data Centres for Building Heating

Recent studies highlight the significant potential of data centre waste heat recovery for building heating applications, especially in the context of low-temperature heating systems. This has enabled a paradigm shift in which data centres can be re-conceptualised as distributed heat sources embedded within or near buildings, supplying recovered heat directly to local heating networks. Several system configurations for integrating data centre waste heat into buildings have been proposed and analysed.

Lu et al. [29] proposed an active heat-exchanger waste heat recovery concept that reallocates heating demand across networks using a small heat pump. Over a 5-month winter period, the system recovered 98.4% of available DC waste heat (~90 MWh), while the heat pump consumed 7.5 MWh with an average COP of 4.75. Lu et al. [30] compared two HX integration schemes for recovering waste heat from a 25 kW liquid-cooled rack to supply a building space-heating network designed at 45 °C supply/30 °C return. For residential buildings, hybridisation with heat pumps and storage can address the mismatch between low-grade heat and higher heating-temperature requirements. Bai et al. [31] noted that DC waste heat is often in the range 35–45 °C, which may be insufficient for direct hydronic heating in many residential contexts; heat pumps are therefore commonly used to upgrade temperature, typically achieving COP > 3.0. In their case study, they proposed a cascade system combining water-source heat pumps, heat exchangers, and thermal storage to provide space cooling/heating and domestic hot water by reusing IT waste heat, and report that the proposed configuration could meet the energy demands of a residential area of 86,000 m². Complementary evidence comes from district- and neighbourhood-scale modelling of DC waste heat coupled to building heating systems with seasonal storage. Sun et al. [32] simulated a coupled system integrating DC waste heat + solar thermal + seasonal soil heat storage + ground-source heat pumps for 20 ten-storey residential buildings (total heating area 200,000 m²) in Beijing. Assuming that 97% of server electricity becomes heat and that 60% is recoverable, the available recoverable DC heat was estimated at 727.5 kW. Over one year, DC waste heat contributed 2.10 × 10⁶ kWh (about 14.7% of the 1.43 × 10⁷ kWh annual heating load), reducing reliance on the ground-source heat pump. At the system level, the overall decarbonisation potential depends not only on technology (HX/HP/storage) but also on matching load profiles and minimising periods of heat rejection. Woodruff et al. [33] show—through an organisational modelling perspective—that distributing computing loads to match building heating demand can yield time-averaged energy savings of roughly ~10% under typical assumptions, with better performance in apartment buildings due to their more constant domestic hot water needs.

1.5. Novelty and Contributions of the Present Study

Despite the growing body of literature on data centre waste heat recovery, some relevant gaps still remain. Most existing studies focus on district-scale applications, specific technological subsystems, or cold-climate conditions, while comparatively fewer contributions investigate building-scale integration strategies in Mediterranean climates. In particular, the influence of different building typologies, heating temperature levels, and load profiles on the effectiveness of waste heat recovery systems remains insufficiently explored. Furthermore, annual analyses comparing integrated and non-integrated configurations under consistent boundary conditions are still limited in the current literature.

In this context, the present study investigates the building-scale integration of data centre waste heat in a Mediterranean climate (Florence, Italy) through a comparative annual energy analysis applied to two markedly different building typologies: a residential building and a school. The analysis is based on a representative data centre configuration and explores its potential role as a local thermal energy source within an urban context. The study adopts an annual energy balance perspective, enabling the identification of periods of surplus and deficit and providing a more comprehensive assessment of system effectiveness than analyses limited to peak or winter conditions. From a methodological standpoint, the paper combines building energy modelling with a scenario-based assessment of waste heat integration, allowing the quantification of primary energy savings, reduction in fossil fuel consumption, and potential contribution towards NZEB targets.

Unlike most previous studies, this work provides a building-scale comparative perspective under consistent boundary conditions, highlighting how temperature levels and building use strongly influence system performance. Overall, the study contributes to the ongoing discussion on smart and integrated energy systems by demonstrating how data centres can act as decentralised thermal sources for heterogeneous urban buildings, supporting the transition towards low-carbon and Nearly Zero-Energy Buildings.

2. Materials and Methods

This chapter contains the description of the case studies and the functioning and modelling of the different scenarios.

2.1. General Considerations About the Study

The coupling potential between existing buildings and data centres was investigated through two case studies of real buildings: a public residence (case study B1) and a school (case study B2). Building B1 is located in Florence, while building B2 is located near Prato, both in Tuscany (Italy). The weather data of the two cities, which are only a few kilometres apart, are very similar. Both cities are in a Mediterranean climate zone, but with continental microclimatic characteristics, being approximately 100 km from the sea. Florence has 1821 Heating Degree Days (HDDs), while Prato has 1668 HDDs, pursuant to the Italian law D.P.R. 412/93 [34]. The climate dataset for the city of Florence is taken from the global database Energy Plus, compiled as Test Reference Year (TRY) in .epw format. The climate dataset for the city of Prato is instead taken from the national database Comitato Termotecnico Italiano (CTI), compiled as TRY in .xlsx format. Both the hourly climate data series are consistent with the hourly meteo data series taken from the meteorological station Consorzio Lamma, located midway between Florence and Prato, near the city of Sesto Fiorentino.

The choice of two buildings in very similar locations aims to highlight differences due to building use, while minimising the influence of climatic variability. The data centre is investigated as a theoretical case study located, depending on each time on the two real case studies, in the proximity of the building. Due to the similarity of the two climates, the annual hourly based climate data of Florence were considered to simulate the data centre.

The analysis aims to comparatively evaluate multiple system configurations suitable for heating existing buildings. Three configurations are considered in this study:

• Scenario S0: It is the baseline scenario, i.e., the actual configuration in which the data centre and buildings are totally independent. The data centre has its own cooling system, while the heating system of the two buildings is fuelled by a gas-fired condensing boiler.
• Scenario S1: The data centre and the buildings are still totally independent, but the heating system of the buildings is supplied by electric heat pumps.
• Scenario S2: It is the integrated scenario, which considers the coupling between the building and the data centre.

Operation is assumed to rely on a data centre air distribution system configured according to a “hot aisle–cold aisle” layout. This arrangement enables operation at higher temperatures, in accordance with the ASHRAE Thermal Guidelines for IT equipment cooling systems in A1-class data centres [35]. The temperature of the return air from the data centre is defined based on established technical and scientific literature to ensure the reliability of IT equipment [17,36].

A data centre with an installed electric capacity of 300 kW is considered. According to widely accepted industry classifications, this capacity falls within the category of edge data centres, which typically range from 50 kW to 2 MW. Such facilities are designed to provide distributed computing resources in proximity to end users and are increasingly deployed in urban contexts [37].

Two key assumptions are adopted for the analysis, in line with the scientific and technical literature of this sector:

• The cooling demand of the IT equipment is assumed to be equal to the installed electrical power, i.e., the entire electrical load is dissipated as heat [18,22].
• The electrical consumption of the IT equipment is assumed to be constant throughout the year, representing a typical base-load operation. This assumption is consistent with literature findings, which highlight that data centres operate under quasi-continuous load conditions due to their 24/7 operation [26,38,39].

The spatial separation between the data centre and the buildings is neglected; therefore, heat losses along the connection pipelines and the additional pumping energy required for fluid circulation are not considered. This assumption represents a best-case thermodynamic scenario and allows isolating the intrinsic performance of the proposed coupling strategy, focusing on the efficiency of heat recovery and utilisation within the DC–HP–building system.

In typical urban applications, the distance between edge data centres and potential waste heat users generally ranges from a few tens to several hundreds of metres. Within this range, transport-related losses and pumping requirements are expected to remain limited compared to the recovered thermal output and are unlikely to significantly affect the qualitative outcomes of the analysis. Nevertheless, in practical implementations, the physical distance would represent a critical parameter influencing both system efficiency and economic viability and should therefore be evaluated through a dedicated site-specific engineering and techno-economic assessment. Accordingly, the results presented herein should be interpreted as a best-case thermodynamic evaluation of the proposed integration concept.

Both buildings are in urban expansion areas dating back to the second half of the 20th century. At the time, as is the case today, urban planning regulations required a balanced ratio between built-up and open areas, including urban green spaces. The residential building is surrounded by a large plot of publicly owned land, which serves as the park for the overall social housing district. A similar situation applies to the garden surrounding the school. In a “best-case scenario” perspective, the underlying idea of this paper is to install the data centres in prefabricated containers in the immediate vicinity of the buildings. This makes it possible to bypass the cost and inefficiency issues that burden district heating networks. Indeed, an additional connection to the fibre-optic network for data transmission is far simpler and less expensive.

The thermodynamic simulations were carried out using TRNSYS v17 and Python v3.9.

2.2. General Description of the Two Buildings

2.2.1. Public Residence

The first case is a public housing multi-dwelling building, located near the city of Florence (Italy). The building consists of 33 apartments of varying sizes, distributed over four floors, for a total gross heated floor area of 3600.0 m². The geometry is regular, with an almost parallelepiped shape.

The building was constructed in the 1980s using reinforced concrete load-bearing wall technology, consisting of sandwich panels (concrete–polyurethane–concrete). Numerous unresolved thermal bridges are present. Windows have aluminium frames without a thermal break and single glazing. Overall, the building envelope performance is poor, and no energy retrofit interventions have been carried out over time.

The building is representative of the social housing stock constructed between the 1970s and the 1990s, when construction costs were often prioritised over energy performance. In the social housing district Le Piagge, near the city of Florence, where the building is located, many buildings had the same shape, size and construction techniques. The same scenario is repeated in other neighbourhoods on the outskirts of Florence, and in other neighbourhoods on the outskirts of major Italian cities. Therefore, the building is highly representative of the Italian social housing stock.

In the model, the building is divided into 33 thermally coupled heated zones, corresponding to the 33 apartments. Additional unheated thermal zones were defined to represent stairwells and basements. In a thermally coupled configuration, each heated zone exchanges heat not only with the outside and with unheated zones, but also with adjacent heated zones if there is a temperature difference across the partition walls. The user pattern is typical for a social housing multi-dwelling building. The heating system is continuously activated during the winter season, and it is managed in a two-temperature mode. During the day, the temperature setpoint is 20 °C, instead during the night the setpoint is lower, about 16–18 °C, due to energy saving purposes. The effective indoor temperature is managed at the single room level through thermostatic valves that modulate the hot water flow across the radiators. The conventional air change rate, in natural ventilation mode, is equal to 0.5 vol/h, in compliance with the standard UNI/TS 11300-1, which applies in Italy the international standard UNI EN ISO 13790. Also, the indoor energy gain, both sensible and latent, is defined in the same standard as a conventional value for a square metre of dwelling floor. The value is an average, taking into account users, kitchens and bathrooms.

In the current state, the building is served by a centralised gas-fired condensing boiler. The distribution network is a high-temperature hydronic system, with supply temperature of 70 °C and return temperature of 60 °C, serving steel panel radiators. The supply temperature is controlled through a weather-compensated control curve based on an outdoor temperature sensor. Winter design conditions for the city of Florence are set at 0 °C outdoor temperature (according to UNI/TR 10349-2 [40]) and 20 °C indoor temperature (according to UNI EN 12831 [41]). The flow rate at each terminal unit is modulated through thermostatic valves. Overall system balancing is achieved through a variable-speed circulation pump, capable of continuously adapting the flow rate to the opening degree of the thermostatic valves. Ventilation is natural, achieved through manual window opening. The reference ventilation rate for residential buildings is 0.50 air changes per hour. Domestic hot water (DHW) is produced by individual electric resistance boilers and is therefore not relevant for the purposes of this analysis. No cooling systems are installed, except for a few individual split units.

In summary, the multi-dwelling building represents a seasonal user, where coupling with the data centre can be implemented from November to April (conventional heating season for Florence). During this period, the building is continuously occupied, as it serves elderly or socially vulnerable residents.

2.2.2. School

The second case analysed is a primary school (children aged 6–10 years). It is located near the city of Prato (Italy), not far from Florence. The school season extends approximately from mid-September to mid-June. The building is open from Monday to Friday, from 7.00 AM to 5.00 PM. During the opening hours, the indoor temperature setpoint is 20 °C. Instead, at night, as well as during the weekends and winter holidays, the central heating is off. The school user profile is taken from the national annex UNI/TS 11300-1 to the global standard UNI EN ISO 13790, applying the A3 assessment mode “tailored fit”. It consistently represents the actual school energy management. In particular, the sensible and latent thermal gains due to the students and the teachers follow UNI EN ISO 7730, differentiating the main thermal zones (classrooms, gym, hallways and bathrooms).

The building has a regular layout and consists of three adjacent blocks, for a total heated floor area of 3300 m². The first is the classroom block, developed over two above-ground floors, each with an internal height of 3.5 m. The second is the entrance block, developed over a single floor and acting as a connection to the third block, consisting of a double-height gymnasium (internal height of 7 m). The total heated volume is 15,000 m³, with a heat loss surface of 6600 m², resulting in a shape factor (S/V) of 0.45, indicating a moderately compact shape. The building was constructed in the 1980s using a reinforced concrete frame structure, with heavy sandwich panel infill walls (concrete–mineral wool–concrete). Numerous unresolved thermal bridges are present. Windows consist of aluminium frames without a thermal break and single glazing.

The building has been selected because it is representative of standard school buildings constructed between the 1970s and the 1990s to remedy the lack of common services in urban growth. During the second half of the 20th century, the main Italian cities were subject to a phenomenon of urbanisation, which led to the construction of large suburbs for almost exclusively residential use. At the same time, it became necessary to quickly build schools that were simple from an architectural point of view and economical from a construction point of view. Such schools are generally characterised by poor energy performance.

The average overall heat transfer coefficient is approximately 1.00 W/(m²·K), with a ratio between equivalent summer solar area and useful floor area equal to 0.5. These parameters indicate moderate heat losses and a significant contribution of solar gains through glazing. Overall, the building envelope performance is poor, and no energy retrofit interventions have been carried out, except for the installation of a photovoltaic system on the gym roof, with an estimated electrical capacity of 40 kW. Since the school building is equipped with a photovoltaic solar system, while the residential building is not, it was decided not to account for on-site electricity generation to make the two analyses more comparable. The solar system would be able to produce 41,300 kWh annually, of which 30% during the heating season.

In the current state, the building is served by a centralised gas-fired condensing boiler. The distribution network is a medium-temperature hydronic system, with supply temperature of 50 °C and return temperature of 40 °C, serving fan-coil units. Ventilation is natural, through manual window opening. The reference ventilation rate for school buildings is 2.00 air changes per hour, derived as the average between classrooms (3.00 vol/h) and circulation areas (1.00 vol/h). These values are based on the Italian national annex of UNI EN 16798-1, assuming indoor environmental quality class II and very low indoor pollution. Domestic hot water (DHW) is produced by individual electric resistance boilers and is therefore not relevant for the purposes of this analysis. No cooling systems are installed.

2.3. Modelling and Energy Analysis of the Two Buildings

The buildings were fully simulated under transient conditions using TRNSYS v.17.

Concerning the residential building, in a first phase, the geometries were modelled using SketchUp v8, while metadata were defined through the TRNSYS3D plugin. This enabled not only a complete three-dimensional representation, but also the automatic calculation of the shading matrix and view factor matrix. The former is used to determine, at each timestep, the external surfaces effectively exposed to direct solar radiation, accounting for shading from overhangs, recesses, and potential surrounding obstructions (not present in this case). The latter is used to calculate, at each timestep, the effective fraction of radiative heat exchange towards the sky vault, also known as extra heat loss. The TRNSYS3D models were subsequently imported into TRNBuild for the thermal characterisation of the building envelope components.

Concerning the school, the building was fully simulated under transient conditions using TRNSYS v.17. In this case, the model was directly implemented in TRNBuild due to the simpler geometry compared to the previous case. The main shading effects due to building geometry were calculated in Simulation Studio using multiple Type 34 components, such as “overhang and wingwall shading” from the “loads and structures” library. Within TRNBuild, the building was divided into 10 thermally coupled zones, considering use, floor level, and orientation.

Running the TRNBuild model, the heating demand profiles of the two buildings were defined. These were calculated on an hourly basis using standard weather data for the locations of Florence and Prato, available in EPW format from the global Energy Plus database. Through Type 15-3, solar radiation incident on surfaces with arbitrary tilt and orientation can be calculated, starting from the values of direct and diffuse solar radiation measured on the horizontal plane. The sky temperature was determined using Type 69. Ground temperature at a given depth below ground level was calculated using Type 77, which correlates outdoor air temperature with ground temperature at a specified depth based on the geophysical properties of the soil.

The Simulation Studio environment in TRNSYS is used to assemble the various system components and generate the input file for simulation. Although the hourly heating demand profile was already sufficiently representative, it did not allow numerical convergence of the associated HVAC components. Component libraries and control routines require smaller time steps to iteratively converge to a stable solution. Therefore, both building models were recalculated with a 10 min timestep. This resulted in a significant computational burden but fixed TRNSYS-solver convergence issues.

The validation of both models was carried out through comparison with previously validated TRNSYS models. The TRNSYS model framework, which served as the basis for the workflow related to the specific energy systems analysed, was validated with reference to two similar models. The first relates to a school building, published in [42], while the second relates to a sports facility, published in [43]. Both the reference models are validated through a numerical-experimental analysis, on the basis of the effective energy consumption during a winter season, and the climate data corresponding to the same winter season.

2.4. Heat Pump Performance Modelling and Performance Map Generation

The performance of the heat pumps adopted in this study was implemented in TRNSYS through Type 917. However, the direct application of available TRNSYS heat pump components is not always straightforward, particularly for water-to-water and high-temperature configurations. Most TRNSYS heat pump models require extensive performance maps describing heating capacity and electrical consumption over a wide range of operating conditions. Such datasets are often not available from manufacturers, who typically provide performance information only for standard rating conditions.

Several studies have addressed this limitation by generating performance maps through dedicated thermodynamic models and subsequently implementing them within TRNSYS simulations [44,45,46]. Following a similar approach, a dedicated heat pump model was developed in Python to generate the performance datasets required by Type 917.

The model is based on fundamental mass and energy balance equations and allows the simulation of different heat pump configurations under varying source and sink conditions. Compressor performance is described through polynomial correlations, while thermodynamic properties are evaluated using REFPROP. This approach allows the adoption of different working fluids, including high-temperature refrigerants.

For each heat pump configuration considered in this study, the Python model was used to generate performance maps expressing heating capacity, electrical consumption, and coefficient of performance (COP) as functions of the operating conditions. The resulting datasets were then implemented within TRNSYS Type 917 to simulate both the residential and school heating systems under off-design conditions throughout the annual simulations.

The proposed approach combines the flexibility of a physically based thermodynamic model with the computational efficiency of TRNSYS performance-map simulations, providing a consistent representation of heat pump behaviour across all the scenarios investigated.

2.5. Energy and Environmental Assessment Criteria for the Two Buildings

The UNI EN ISO 52000-1 standard defines primary energy as “energy that has not been subjected to any conversion or transformation process”. The entire energy chain is considered, from natural resources to final delivery to the end user, including intermediate transformations and/or conversions for each energy carrier [47].

To convert energy carriers into non-renewable primary energy, non-renewable conversion factors are used, defined as “non-renewable primary energy for a given energy carrier, including the delivered energy and the considered energy overheads of delivery to the points of use, divided by the delivered energy.” Similarly, renewable conversion factors can be defined. The sum of renewable and non-renewable factors represents the ratio between total primary energy and total delivered energy for a given carrier. UNI/TS 11300-5 provides the values of

Table 1. Factors of renewable and non-renewable energy vectors from UNI/TS 11300-5.

Energy Vector	Total Factor	Ren. Factor	Non-Ren. Factor
Gas	1.05	0	1.05
Electricity from the grid	2.42	0.47	1.95
Sun/Wind	1.00	1.00	0
Air/Water/Ground	1.00	1.00	0

[48].

CO₂ is the main greenhouse gas. Thermal generators with fossil fuels produce on-site CO₂ emissions, whereas emissions of the grid electricity are generated remotely at power plants. CO₂ emissions for a given energy carrier are estimated using specific emission factors. The UNI EN ISO 52000-1 standard defines the emission factor as a “coefficient that describes the amount of CO₂ released from a given activity, such as burning one tonne of fuel in a furnace” [47]. UNI/TS 11300-5 provides the values of

Table 2. Factors of CO₂ emission from UNI/TS 11300-5.

Energy Vector	Emission Factor kgCO2/kWh
Gas	0.21
Electricity from the grid	0.47
Sun/Wind	0
Air/Water/Ground	0

[48].

2.6. Scenarios S0 and S1: No Integration Between DC and Buildings

In Scenario S0, the DC is cooled by its own cooling system, and the buildings are served by their own heating system (Figure 1).

It is useful to visualise the different thermal levels of the systems (

Table 3. Temperature levels of the different systems.

Value °C	Description
70.0	Supply of water loop, residential building
60.0	Return of water loop, residential building
50.0	Supply of water loop, school
40.0	Return of water loop, school
40.0	Return of air loop, data centre
35.0	Return of water loop, data centre
25.0	Supply of air loop, data centre
20.0	Supply of water loop, data centre
15.0	Outdoor air limit for free cooling, theoretical
10.0	Outdoor air limit for free cooling, calculation

).

2.6.1. DC with Non-Integrated Cooling System

A 300 kW electric load data centre with an air cooling system is considered.

The system is made by:

-

The air loop, which cools the data centre, and is cooled at the Air-to-Water Heat Exchanger (AWHE).

-

The cold-water loop, which is cooled at the Free Cooling Heat Exchanger (FCHE) by outdoor air or at the evaporator of Vapour Compression Cycle 1 (VCC 1).

-

The refrigerant loop of Vapour Compression Cycle 1 (VCC 1).

Outdoor air is used for the free cooling operations and for the condensation of the refrigerant of VCC 1. The system operates in indirect water-side free cooling (FC mode) when the outdoor air temperature is below a predefined threshold (T_{lim FC}); when the external temperature exceeds this limit, a vapour-compression cooling system VCC 1 (AC mode) operates; hybrid operation of the two modes is not considered. The model uses a modulating water-to-air unit operating with refrigerant R515B.

The system adopts a hot-aisle/cold-aisle air distribution configuration, which allows higher operating temperatures than a fully mixed airflow arrangement, in line with the ASHRAE Thermal Guidelines for IT equipment cooling systems [35]. The cooling system activates AC mode only when outdoor conditions exceed the threshold and free cooling is ineffective. The return air temperature from the data centre follows values recommended in the technical and scientific literature [17,36].

The cooling power required by the IT equipment is assumed to be equal to the generated heat load. Accordingly, the cooling airflow rate supplied to the data centre (m_DC) in both operating modes is given by:

$${\mathrm{Q}}_{\mathrm{DC}, \mathrm{cold}} = {\mathrm{c}}_{\mathrm{p}} · {\mathrm{m}}_{\mathrm{DC}} · \left({\mathrm{T}}_{\mathrm{DC}, \mathrm{out}}-{\mathrm{T}}_{\mathrm{DC}, \mathrm{in}}\right)$$

(1)

where ${\mathbf{Q}}_{\mathbf{DC}\mathbf{,} \mathbf{cold}}$ is the cooling demand of the data centre, ${\mathbf{m}}_{\mathbf{DC}}$ is the cooling air mass flow rate supplied to the data centre, c_p is the specific heat capacity at constant pressure of air, and ${\mathbf{T}}_{\mathbf{DC}, \mathbf{in}}$ and ${\mathbf{T}}_{\mathbf{DC}, \mathbf{out}}$ are the inlet and outlet air temperatures, respectively.

In FC mode, the external airflow rate required to deliver the target cooling capacity is determined by imposing a constant heat exchanger effectiveness. The exchanged heat is assumed to be equal to the required cooling capacity, which in FC mode coincides with the data centre thermal load. The outdoor airflow in FC mode is modulated in order to achieve the proper supply temperature, also when the outdoor temperatures are low.

Given the known airflow rates on both the data centre side and the ambient side, the fan power and the corresponding annual energy consumption can be calculated by assuming appropriate pressure drops across the heat exchangers and fan efficiency:

$${\mathrm{W}}_{\mathrm{f}} = {\displaystyle \frac{\mathrm{m} · \mathrm{DP}}{\mathrm{d} ·{ \mathrm{eff}}_{\mathrm{f}}}}$$

(2)

where ${\mathbf{W}}_{\mathbf{f}}$ is the fan power consumption, m is the air mass flow rate, DP is the pressure drop across the system, d is the air density, and eff_f is the fan efficiency.

When the outdoor temperature exceeds the limit for free cooling operation, a vapour compression cycle (VCC 1) provides the required cooling capacity. In air-conditioning (AC) mode, the total electrical demand includes both the fan power and the compressor power of the VCC (W_comp). Estimating the AC power consumption, therefore, requires the evaluation of the cycle Energy Efficiency Ratio (EER).

The EER is modelled using a polynomial correlation as a function of the refrigerant saturation temperature levels, based on data from commercially available units [49]. Validation of the adopted polynomial formulation is reported in the Supplementary Material of [50]. By imposing fixed pinch-point temperature differences at the evaporator and condenser (DT_PP) and a fixed sink temperature difference (DT_sink), the refrigerant saturation temperatures can be determined. An additional constraint ensures a minimum allowable temperature lift between the refrigerant saturation levels (DT_comp).

Given the resulting relationship between EER and ambient temperature, the compressor power consumption and the heat rejected to the environment (Q_sink) can be calculated.

$${\mathrm{W}}_{\mathrm{comp}}={\displaystyle \frac{{\mathrm{Q}}_{\mathrm{cold}}}{\mathrm{EER}}}$$

(3)

$${\mathrm{Q}}_{\mathrm{sink}}={\mathrm{Q}}_{\mathrm{cold}}+{\mathrm{W}}_{\mathrm{comp}}$$

(4)

where W_comp is the compressor power consumption, Q_cold is the cooling capacity provided by the vapour compression cycle, Q_sink is the heat rejected to the environment, and EER is the Energy Efficiency Ratio.

The electrical demand of the evaporator and condenser fans is computed using Equation (2), following the same approach adopted for free cooling mode.

All the boundary conditions and constant parameters are summarised in

Table 4. Regulation of FC and AC modes.

FC Operations, Air and Water Loops			VCC Cycle (AC Mode)
effhe		0.8	${\mathrm{D}\mathrm{T}}_{\mathrm{p}\mathrm{p}}$	5.0	°C
DP exch. [51]	Pa	150.0	${\mathrm{D}\mathrm{T}}_{\mathrm{c}\mathrm{o}\mathrm{m}\mathrm{p},\mathrm{m}\mathrm{i}\mathrm{n}}$	10.0	°C
efff		0.6	${\mathrm{D}\mathrm{T}}_{\mathrm{s}\mathrm{i}\mathrm{n}\mathrm{k}}$	10.0	°C
Thermophysical properties
cp	kJ/kg/K	1.0	d	kg/m3	1.2

.

2.6.2. Buildings with Non-Integrated Heating Systems, Boilers

The current (actual) HVAC configuration of the residential building was modelled using a gas boiler (Type 700) with a nominal capacity of 300 kW, supplying the distribution network without intermediate storage. The boiler produces water at a setpoint temperature regulated by a weather-compensation control curve, and a maximum value equal to 70 °C (minimum value equal to 50 °C). Two circulation pumps are present: one in the primary loop between the boiler and the heat exchanger, and one in the secondary loop between the heat exchanger and the radiators. In the absence of a buffer tank, the boiler operates at every timestep in which the building heating demand is non-zero, i.e., for approximately 2000 h per heating season. Using UNI/TS 11300-2, the seasonal efficiency values were estimated as follows: distribution = 0.93, emission = 0.95, and control = 0.98. The generation efficiency is estimated at 0.90, as the return water temperature to the boiler is almost always above the flue gas dew point, except during the milder periods of the heating season. The interaction between the HVAC system and the building represents the most complex aspect of the model. Since it was not feasible to simulate a detailed distribution network across 33 thermal zones, the heating demands of all zones were aggregated into a single value at each timestep. The thermal energy in the return line of the hydronic network was calculated as the difference between the thermal energy in the supply line and the building heating demand.

The current HVAC configuration of the school building was modelled in a similar way, with boiler operation exceeding 900 h per season (effectively operating whenever the school is open). In this case, however, since the terminals are fan coil units, the operating temperatures of the hydronic circuit always allow flue gas condensation, resulting in a higher generation efficiency compared to the previous case.

Both models use TESS Type 515 to define the conventional heating season, identical for both cases, extending from 1 November to 15 April. The results presented in this study refer to this period. Both models also use TESS Type 514 to define system operation schedules. In the residential building, an attenuated regime is adopted, with a setpoint temperature of 20 °C from 08:00 to 20:00 and a setback temperature of 16 °C from 20:00 to 08:00. In the school building, an intermittent regime is adopted, with system operation at 20 °C from Monday to Friday, between 08:00 and 16:00, and free-floating conditions during the remaining time. Weekend shutdowns are considered, whereas winter holidays and occasional school closures are neglected.

Operating schedules also affect ventilation and internal gains. In the residential building, a constant ventilation rate of 0.5 vol/h is assumed over 24 h. In the school building, ventilation rates from the Italian national annex to UNI EN 16798-1 are applied only during occupancy. Similarly, internal gains are assumed to be continuous at 6 W/m² for the residential building and intermittent at 16 W/m² for the school building.

2.6.3. Buildings with Non-Integrated Heating Systems, Heat Pumps

Under the same operating conditions and user profiles, this scenario considers a standard retrofit intervention limited to the replacement of the heating plant. In line with current European energy policies promoting electrification of end uses, the gas boiler is replaced with an air-to-water inverter-driven heat pump.

In the residential case, a high-temperature (two-stage compression) heat pump is adopted, whereas in the school case, a single-stage unit is used. The decision not to modify the distribution system or terminal units in the residential building is based on practical considerations. Given that the building hosts elderly and vulnerable occupants, invasive retrofit interventions are not considered feasible. This results in reduced efficiency of the two-stage heat pump, which must supply water at temperatures up to 70 °C. The model employs an air-to-water heat pump operating with refrigerant R1233zd(E).

Within Simulation Studio, the heat pump is modelled using TESS Type 917. Performance maps and correction factors for off-design conditions were generated through a Python-based routine, once the refrigerant had been selected. The heat pump and primary circulation pump operate to maintain a buffer tank, sized at 10 L/kW_thermal, at a setpoint temperature of 70 °C for the residential case and 50 °C for the school case, with a 5 °C temperature difference on the heat exchanger. The system is controlled via a thermostat located at the buffer tank. The secondary circulation pump and terminal units operate whenever the building heating demand is greater than zero. The buffer tank also acts as a hydraulic separator, decoupling the plant-side operation from the user-side demand, resulting in improved hydronic stability.

From an environmental perspective, the fuel switch from gas to electricity leads to immediate local decarbonization of heating plants, improving air quality in densely populated urban areas. In addition to CO₂, emissions of NO_x, particulate matter, and fine dust are eliminated locally. However, since the Italian electricity mix is still largely dependent on thermoelectric generation, emissions are effectively shifted rather than eliminated [52]. Nevertheless, the increasing share of renewable energy sources is expected to progressively reduce the emission factor, currently estimated at 0.46 kgCO₂/kWh.

Also in this scenario, the data centre operates independently and is not coupled with the heating systems. It has a constant electrical load of 300 kW, corresponding to an equivalent thermal load dissipated to the environment throughout the year.

2.7. Scenario S2: Integration Between the Data Centre and the Buildings

The third scenario involves recovering waste heat from the data centre to supply the heating systems of the buildings. From the data centre perspective, the coupling affects only the cooling system. Between May and October, the system operates in stand-alone mode. From November to April, part of the waste heat is diverted to the heat recovery system.

In Scenario S2, the data centre is cooled by the same cooling system adopted in the previous scenarios, while waste heat is recovered from the return of the cold-water loop. A second vapour compression cycle (VCC 2) is employed to upgrade the recovered thermal energy and supply the building heating system when required (Figure 2).

2.7.1. DC Integrated Cooling

In this scenario, the air loop still works in the same way as scenarios S0 and S1. Generally, the modalities free cooling—active cooling are maintained: VCC 1 switches on when the outdoor air is too hot to operate the indirect water-side free cooling. The logic of the waste heat recovery is based on diverting part of the return water flow from the AWHE towards the VCC 2 system (heat pump of a specific building). In the water loop of the data centre, a motorised modulating diverter valve is installed, which diverts a portion of the fluid at 35 °C exiting from the AWHE coil toward the heat recovery heat pump. If, at a given timestep, the thermostatic control of the buffer tank is not satisfied, the heat pump is activated, and during that timestep, a share of the heat generated by the data centre is used to increase the water temperature in the buffer tank. Otherwise, during that timestep, the heat is rejected to the external environment. The opening percentage of the diverter valve is regulated through a proportional control based on the deviation between the temperature measured in the buffer tank and the desired setpoint temperature. This control strategy is not optimised, but it is acceptable given the short timestep of 10 min and the thermostat deadband of the buffer tank equal to 1 °C.

During free cooling operating hours, the data centre cooling system consumption is limited to system auxiliaries. The water-side flow rate is constant, and, therefore, the electric pumps’ power consumption is constant. Conversely, the air-side flow rate on the dry cooler is variable via an inverter, depending on the heat load to be dissipated. This allows the unit to operate at nominal capacity in the absence of heat recovery, i.e., when the entire data centre load needs to be dissipated. Conversely, it operates at partial capacity (i.e., with fans at reduced speed) when part of the load to be dissipated is used by the heat recovery heat pump.

2.7.2. Buildings Integrated Heating Systems, Heat Pumps

From the user-side perspective, the heating system is activated at each timestep in which the building’s net heating energy demand is non-zero. In such cases, the primary circulation pump of the distribution network is started, assuming that the water in the buffer tank is already at a suitable temperature. The control on the user side and the control on the source side are independent, as the buffer tank operates as both a hydraulic and functional decoupler.

The recovered heat, at approximately 35 °C, is not at a sufficiently high temperature to directly supply the hydronic systems, which require average supply temperatures of 70 °C for radiators (residential case) and 45 °C for fan coils (school case). In the residential case, a double-stage heat pump is required. In the school case, a single-stage heat pump is sufficient to bridge the temperature gap between the recovery source and the buffer tank setpoint. The units are modelled using TESS library models. Performance datasets under off-design and part-load conditions were generated using Python routines.

Starting from the transient analysis of the building’s thermal performance, the energy demands for each considered HVAC configuration were initially assessed by energy carrier, distinguishing between gas and electricity. In a second phase, by applying the conversion and emission factors defined in UNI/TS 11300-5, the non-renewable primary energy demand and CO₂ emissions were calculated. These demands were then normalised with respect to the heated floor area for each building, to express the energy performance according to the same metric used in energy certification schemes. In all the scenarios considered, both individual and combined, the energy consumption of the residential building or school was added to that of the data centre, and the overall value was compared.

2.7.3. Control Strategy of the Integrated System

In the residential case, heat recovery is controlled through a thermostatic regulation of an intermediate buffer tank between the two stages of the heat pump, set at 50 °C. When this threshold is reached, both stages are activated: the first recovers heat from the data centre and raises the temperature to the intermediate level, while the second increases it up to 70 °C to meet the requirements of the high-temperature distribution system.

In the school case, a single-stage heat pump is sufficient to maintain the buffer tank at 50 °C, which directly supplies the heating system. Therefore, the temperature lift at the recovery stage is identical in both cases, but only the residential configuration requires an additional stage to reach the final supply temperature.

The activation of the heat recovery system is driven by the thermal demand of the end user and is not based on an optimisation of device efficiency. This approach reflects a simplified configuration in which no auxiliary or backup generators are installed. The control logic can be interpreted from the user side to the source side: when the building heating demand increases, thermal energy is drawn from the buffer tank, causing its temperature to drop below the setpoint value (70 °C for the residential building and 50 °C for the school). This triggers the activation of the heat pump(s), diverting the available thermal energy towards heat recovery.

As a result, heat recovery mainly occurs during the colder periods of the heating season, when transmission and ventilation losses exceed internal and solar gains. During these periods, free cooling conditions may also occur, as outdoor air temperatures fall below the threshold of 10 °C. The overlap between free cooling and heat recovery amounts to 2506 h/year in the residential case and 1914 h/year in the school case.

3. Results and Discussion

In this chapter, the results of the energy analysis of the different scenarios are reported and compared in terms of the performance of data centre cooling systems and waste heat recovery and reutilisation. The results are organised into three sections: data centre performance, residential building performance, and school building performance.

3.1. Data Centre

3.1.1. Non-Integrated Scenarios

For scenarios S0 and S1, the combination of water-side indirect free cooling and VCC 1 active cooling in the data centre results in an electricity consumption and waste heat as reported in

Table 5. Annual electricity consumption and waste heat production of data centre cooling system, Scenarios S0–S1.

Heat IT	Electricity FC	Electricity AC	Electricity Tot	Waste Heat FC	Waste Heat AC	Waste Heat Tot
MWh	MWh	MWh	MWh	MWh	MWh	MWh
2628.0	12.4	495.7	508.2	846.0	2290.2	3136.2

. In this table, it is important to underline that the term “waste heat” refers to the heat discharged into the outdoor environment. The heat production of IT equipment is equal to the installed electric capacity (300 kW) multiplied by the total hours of functioning (8760 h/year), and it is discharged into the water loop through the exchange air-water at the AWHE.

3.1.2. Integrated Scenario

For scenario S2, the combination of water-side indirect free cooling and VCC 1 active cooling in the data centre results in an electricity consumption and waste heat as reported in

Table 6. Annual electricity consumption and waste heat production of data centre cooling system—Scenario S2.

Building	Electricity FC	Electricity AC	Electricity Tot	Waste Heat FC	Waste Heat AC	Waste Heat Tot
	MWh	MWh	MWh	MWh	MWh	MWh
Residential	12.4	475.0	487.4	469.3	2195.0	2664.3
School	13.4	466.3	479.7	663.8	2152.8	2816.6

, differentiated between the two integrations: DC-residential building and DC-school. For scenario S2, it is useful to note that there is a slight reduction in electricity consumption because part of the heat produced by the data centre and discharged into the water loop is taken by the recovery system and directed towards the heat pump (VCC 2) of a specific building, reducing the cooling load required at the evaporator of VCC 1. Therefore, the reduction in electricity consumption observed in Scenario S2 is directly linked to the partial recovery of waste heat, which reduces the thermal load to be rejected by the cooling system.

The “chain” of waste heat recovery and valorisation is indicated in

Table 7. Chain of waste recovery and valorisation—Scenario S2.

Building	Waste Heat Cooling DC→Environment MWh	Waste Heat DC→VCC 2 MWh	Building Heating Demand MWh	Additional Heat from VCC 2 MWh
Residential	2664.3	451.2	580.1	128.9
School	2816.6	291.2	332.8	41.6

, where there are the results for: waste heat discharged from the cooling system towards the outdoor environment, the waste heat diverted from the water loop towards the VCC 2 (heat pump for each building), the thermal request of each building, the additional heat generated by VCC 2 (heat pump) for each building.

3.2. Residential Building

3.2.1. Non-Integrated Scenarios

Assuming a seasonal average efficiency of 0.87 for the generation subsystems and 0.90 for the gas-fired condensing boiler, the baseline scenario (Scenario S0) yields a total gas consumption for space heating of 71,951.0 m³/year.

In the all-electric scenario (Scenario S1), in a stand-alone configuration, the high-temperature two-stage heat pump (replacing the gas-fired condensing boiler) exhibits an average seasonal COP of 2.15. The relatively low efficiency of the system is due to the large temperature lift between evaporation and condensation temperatures. The minimum and maximum COP values are 1.64 and 3.19, respectively. The seasonal electricity consumption for heating amounts to 289,537.0 kWh.

3.2.2. Integrated Scenario

In the third scenario, which combines the data centre with the building heating plant through heat recovery via an electric heat pump, the COP of the unit serving the heating system increases to 4.50, resulting in a reduction in the building electricity consumption to 128,910 kWh. This improvement is primarily due to the higher source temperature provided by the data centre (approximately 35 °C), which significantly reduces the temperature lift required by the heat pump compared to ambient air conditions. However, the overall performance remains limited by the high supply temperature of the radiator-based system (up to 70 °C), which constrains the achievable efficiency.

3.3. School

3.3.1. Non-Integrated Scenarios

Assuming a seasonal average efficiency of 0.88 for the generation subsystems and 0.95 for the gas-fired condensing boiler, the baseline scenario (Scenario S0) yields a total gas consumption for space heating of 48,700 m³/year.

In the all-electric scenario (Scenario S1), in a stand-alone configuration, the heat pump (replacing the gas-fired condensing boiler) exhibits an average seasonal COP of 2.51. A significant increase in efficiency is observed compared to the previous case, due to the transition from a high-temperature hydronic system to a medium-temperature hydronic system (i.e., from radiators to fan-coil units). The minimum and maximum COP values are 1.69 and 4.07, respectively. The seasonal electricity consumption for heating amounts to 161,290 kWh.

3.3.2. Integrated Scenario

In the third scenario (Scenario S2), which combines the data centre with the building heating plant through heat recovery via an electric heat pump, the COP of the unit serving the heating system increases to 8.00, resulting in a reduction in the building electricity consumption to 49,080 kWh.

In summary, the comparison between the residential building and the school shows that the heat recovery strategy is more effective in the latter. This might seem counterintuitive, as the residential building is continuously heated, whereas the school alternates between opening and closing hours. Over the course of the winter season, the school therefore has fewer total hours to exploit heat recovery. However, the key factor is not the operating hours, but the temperature levels. The residential building operates with radiator terminals designed for a flow temperature of 70 °C, while the school operates with fan coil units designed for a flow temperature of 50 °C. Given that the recovery heat source is available at the same temperature of 35 °C, it is evident that the COP of the heat pump is heavily penalised in the first case compared to the second. Furthermore, from a construction standpoint, the first case requires two heat pumps in cascade with an intermediate thermal buffer, whereas a single heat pump is sufficient in the second. This circumstance further widens the performance gap in terms of energy consumption and CO₂ emissions.

3.4. Final Summary on Building Heating Systems: Energy and Environmental Performances

3.4.1. Residential Building

Considering a heated floor area of 3600.0 m², the normalised energy performance indices for the three scenarios are reported in

Table 8. Summary of energy and environmental assessment for the residential building.

Scenario	EP nren	Differ.	CO2	Differ.
Non-integrated, boiler (S0)	188.0	/	38	/
Non-integrated, heat pump (S1)	157.0	−16%	37	0%
Integrated (S2)	70.0	−63%	16	−58%

and clearly shown in Figure 3.

Compared to Scenario S1, the integration with the data centre significantly improves the heat pump performance by reducing the temperature lift required, which is particularly relevant in high-temperature systems such as radiator-based heating.

Considering the combined system of the residential building and the data centre, the total non-renewable primary energy demand in the three scenarios is as follows:

-

Baseline scenario (S0), EP = 1,667,687.0 kWh;

-

Stand-alone heat pump scenario (S1), EP = 1,555,505.0 kWh;

-

Heat pump with waste heat recovery scenario (S2), EP = 1,201,792.0 kWh.

3.4.2. School

Considering a heated floor area of 3300 m², the normalised energy performance indices for the three scenarios are reported in

Table 9. Summary of energy and environmental assessment of the school.

Scenario	EP nren	Differ.	CO2	Differ.
Non-integrated, boiler (S0)	139	/	28	/
Non-integrated, heat pump (S1)	95	−32%	21	−20%
Integrated (S2)	29	−79%	7	−75%

and clearly shown in Figure 4.

The higher performance observed in the school building is mainly due to the lower operating temperatures of the heating system, which are more compatible with the temperature level of the recovered waste heat, thus allowing a more efficient heat pump operation.

Considering the combined system of the school and the data centre, the total non-renewable primary energy demand in the three scenarios is as follows:

-

Baseline scenario (S0), EP = 1,433,402 kWh;

-

Stand-alone heat pump scenario (S1), EP = 1,289,840 kWh;

-

Heat pump with waste heat recovery scenario (S2), EP = 1,031,108 kWh.

3.5. Final Considerations on Building Typology

A comparison between the two case studies highlights the strong influence of building typology on the effectiveness of waste heat recovery systems. While the residential building shows significant improvements, the school building achieves higher performance gains due to its lower supply temperature and more favourable operating conditions. This confirms that the compatibility between waste heat temperature levels and building heating systems is a key factor in determining the overall system efficiency.

4. Conclusions

The present study analyses the potential integration of data centre waste heat for building heating through a comparative annual energy assessment applied to two building typologies in a Mediterranean climate (a residential building and a school). Three scenarios were considered: non-integrated scenario S0 (data centre with its own cooling system and buildings with gas-fired boilers), non-integrated scenario S1 (data centre with its own cooling system and buildings with air-to-water heat pumps), and integrated scenario S2 (data centre cooling system coupled with the buildings through waste heat recovery and heat pump technology). A theoretical 300 kW data centre is modelled as a constant waste heat source. The analysis is based on dynamic simulations carried out in TRNSYS and Python. The main findings of the study are as follows:

-

Data centre waste heat can be effectively reused as a local thermal source, significantly improving the overall efficiency of building heating systems when coupled with heat pump technologies.

-

The integrated configuration S2 leads to a substantial increase in system performance with respect to the non-integrated configuration S1, with the seasonal COP rising from 2.15 to 4.50 in the residential case and from 2.51 to 8.00 in the school case, due to the reduced temperature lift.

-

Energy consumption is markedly reduced in the integrated scenario S2 with respect to the non-integrated scenario S1, with electricity demand decreasing by more than 50% in the residential building and by approximately 70% in the school building.

-

The environmental benefits are significant, with reductions in non-renewable primary energy demand of up to 63% and 79% for the residential and school buildings, respectively, compared to the baseline boiler scenario S0, confirming the strong decarbonisation potential of the approach.

-

Building typology and load profiles strongly influence system effectiveness, with lower-temperature systems and more favourable demand patterns enabling higher efficiency gains, as observed in the school case.

The present study is based on some simplified assumptions, which may lead to an overestimation of the achievable system performance under real operating conditions. Future research should therefore investigate the impact of dynamic data centre loads, distribution losses, advanced control logics, and economic feasibility. Further developments could also include the integration of thermal storage systems, renewable energy sources, and additional building typologies in order to assess the scalability and practical applicability of the proposed approach.

These results demonstrate that data centres can act as decentralised thermal sources within urban environments, supporting the transition towards low-carbon and Nearly Zero-Energy Buildings (NZEBs).