Publications – Heating Bits

Ravi, S. S., Löffler, T. S., Pina, E. A., Sharma, S., Lepour, D., Terrier, C., & Maréchal, F. (2026). From servers to services: Modeling data centers as heat-active urban energy prosumers. Applied Energy, 327, 127049.

Abstract:

The rapid expansion of cloud services has significantly increased the global energy footprint of data centers, which now account for approximately 2–3 % of global electricity consumption, a figure projected to rise to somewhere between 10 % and 51 % by 2030. While technological advances such as liquid cooling and high-temperature waste heat streams offer opportunities for improved energy efficiency, the integration of data centers into broader urban energy systems remains limited. This study investigates how data centers can transition from passive energy consumers to active prosumers through advanced heat recovery and flexible demand strategies. In this study, five system-level scenarios are modelled, varying by grid connectivity, renewable energy integration, and workload flexibility. Further, two distinct heat recovery approaches are compared: a legacy strategy that dynamically chooses between direct thermal reuse and electricity generation via an Organic Rankine Cycle (ORC), and an “exergy-aware” strategy that enforces continuous ORC operation and cascades the rejected heat from the condenser into a secondary heat pump. Using a multi-objective Mixed-Integer Linear Programming framework, the study reveals the trade-offs between data self-sufficiency and renewable energy utilization in an urban district case study comprising the EPFL campus in Lausanne. The results show that flexible computing workloads and integration with district heating networks can significantly enhance the buffering potential of data centers for variable renewable energy by up to 28 % in certain cases thereby reducing renewable curtailment, and support more efficient heat-electricity coupling by showing potential to supply up to 40 % of the heat demand of the campus. This work positions data centers as critical enablers of sustainable urban energy systems and offers a transferable modeling framework for their systemic integration.

DOI: https://doi.org/10.1016/j.apenergy.2025.127049

Figini, E., & Paolone, M. (2025). Achieving dispatchability in data centers: Carbon and cost-aware sizing of energy storage and local photovoltaic generation. Sustainable Energy, Grids and Networks, 43, 101920.

Abstract:

Data centers are large electricity consumers due to the high consumption needs of servers and their cooling systems. With the rapid growth of crypto-currency and artificial intelligence, their electricity consumption is expected to increase substantially. With the electricity sector being responsible for a large share of global greenhouse gas (GHG) emissions, it is important to lower the carbon footprint of data centers to meet GHG emissions targets set by international agreements. Moreover, uncontrolled data center integration into power distribution grids increases the stochasticity of electricity demand, thus increasing the need for reserve capacity and leading to operational inefficiencies and higher emissions.

This work provides a method to size a PhotoVoltaic (PV) system and an Energy Storage System (ESS) for an existing data center looking to reduce both its carbon footprint and demand stochasticity through day-ahead dispatching. A scenario-based optimization framework is developed to jointly size the PV and ESS, minimizing the expected operational and capital expenditures and the carbon footprint of the data center complex. The model considers the life cycle assessments (LCA) of the systems and the dynamic carbon intensity of the upstream electricity supply. Case studies in different Swiss cantons and regions of Germany emphasize the need for location-aware sizing processes since the obtained optimal solutions strongly depend on the local electricity carbon footprint and on the irradiance conditions. The maximum carbon footprint reduction reaches approximately 50 % in Germany and 4 % in Switzerland. Installed power generation and energy storage capacities vary by up to 36 times across regions.

Access to the pre-print

Ravi, S. S., Loeffler, T., Giacomini, G., Pina, E. A., Sharma, S., Terrier, C., & Marechal, F. (2025, June 29–July 4). Exploring data centers as prosumers in urban energy systems – A scenario analysis study. In Proceedings of the 38th International Conference on Efficiency, Cost, Optimization, Simulation and Environmental Impact of Energy Systems (ECOS 2025), Paris, France.

Abstract:

The growth of information services has led to a sharp increase in the number and size of data centers,
making them vital to today’s economy. Currently, they account for 2-3% of global energy use, projected to reach 10% by 2030. Many still rely on outdated cooling systems, which waste energy. However, newer data centers are adopting efficient technologies like micro-fluidic chip cooling. A key issue remains: the underutilization of waste heat, which is often not recovered. This paper discusses strategies to repurpose this waste heat, by using Organic Rankine Cycles (ORC) to generate electricity and incorporating waste heat into district heating systems with heat pumps, presenting sustainable solutions for urban energy needs while reducing the environmental impact of data centers by exploring their role as potential prosumers in the energy system.

Figini, E., & Paolone, M. (2025). Planning energy resources for data centers: Accounting for economic and carbon costs of power imbalance. In Proceedings of the 2025 IEEE Kiel PowerTech (PowerTech 2025). IEEE.

Abstract:

Data centers are critical assets in today's digital world, supporting applications ranging from cloud computing to artificial intelligence. Their substantial energy consumption and stochasticity present challenges for sustainability and power grid operation. Long-term planning of data centers with co-located electricity generation and energy storage offers a fundamental step toward addressing these challenges, but current planning methods often overlook the influence of power imbalance on the data center's carbon footprint and operational costs. In this work, we propose a scenario-based mixed-integer stochastic optimization problem that considers intraday power imbalances to size co-located photovoltaic generation and energy storage. The framework also accounts for the life cycle assessment of the resources, their aging, the local power grid carbon intensity and the solar irradiance conditions. Using a lower bound of the intraday imbalance cost to analyze the case of three Swiss cantons, we observe significant differences in carbon and cost reduction opportunities.

Access to the pre-print

Shahbazinia, A., Huang, D., Costero, L., & Atienza, D. (2025, September 3). CloudFormer: An attention-based performance prediction for public clouds with unknown workload. arXiv.

Abstract:

Cloud platforms are increasingly relied upon to host diverse, resource-intensive workloads due to their scalability, flexibility, and cost-efficiency. In multi-tenant cloud environments, virtual machines are consolidated on shared physical servers to improve resource utilization. While virtualization guarantees resource partitioning for CPU, memory, and storage, it cannot ensure performance isolation. Competition for shared resources such as last-level cache, memory bandwidth, and network interfaces often leads to severe performance degradation. Existing management techniques, including VM scheduling and resource provisioning, require accurate performance prediction to mitigate interference. However, this remains challenging in public clouds due to the black-box nature of VMs and the highly dynamic nature of workloads. To address these limitations, we propose CloudFormer, a dual-branch Transformer-based model designed to predict VM performance degradation in black-box environments. CloudFormer jointly models temporal dynamics and system-level interactions, leveraging 206 system metrics at one-second resolution across both static and dynamic scenarios. This design enables the model to capture transient interference effects and adapt to varying workload conditions without scenario-specific tuning. Complementing the methodology, we provide a fine-grained dataset that significantly expands the temporal resolution and metric diversity compared to existing benchmarks. Experimental results demonstrate that CloudFormer consistently outperforms state-of-the-art baselines across multiple evaluation metrics, achieving robust generalization across diverse and previously unseen workloads. Notably, CloudFormer attains a mean absolute error (MAE) of just 7.8%, representing a substantial improvement in predictive accuracy and outperforming existing methods at least by 28%.

Access to the pre-print

Sudharshan, R. S., Orrego Florez, D. A., Sharma, S., & Maréchal, F. (2024, June 2–6). A comparative study of ORC working fluids performance in ultra–low-grade waste heat recovery from data centres. In Proceedings of the ESCAPE34–PSE24 Symposium, Florence, Italy.

Abstract:

The increasing growth of data centers worldwide has ushered in an era of unprecedented data processing and storage capabilities. As data centers play a pivotal role in the ever-increasing use of cloud computing, social media, and online services in general, their energy consumption continues to rise, accounting up to 1-1.13% of the global electricity demand. Data centers produce low-grade heat. The energy-intensive operations of data centers have spurred a growing interest in waste heat recovery technologies as a means to enhance energy efficiency by usage of this waste heat for electricity generation and district heating. This study examines the waste heat potential of a 5MW data center that primarily employs liquid chip-level cooling to convert waste heat into electricity using an organic Rankine cycle (ORC). By comparing the performances of different working fluids such as R245fa (dry), R134a (wet) and R1234zeE (isentropic) in ORC systems optimized to have the lowest operating cost, this study attempts to understand the operating conditions that maximize the energy savings and improve the Energy Reuse Efficiency (ERE) indicator in datacenters, aiming to reduce the environmental impact.