AI in action: Saving energy, cutting costs and keeping data centers cool

AI and energy efficiency are major focus areas in the data center industry. At our Rhine-Ruhr 1 Data Center in Bonn, Germany, an NTT Global Data Centers pilot project has brought them together.

Rather than operating the data center’s cooling systems with static controls, the site now uses dynamic, AI-driven optimization through the etaONE® platform from etalytics, which develops software for smart industrial energy systems.

This has led to a remarkable reduction of nearly 20% in the cooling system’s energy consumption.

Innovative technology for data center sustainability

As one of the world’s largest data center colocation providers, we’re always exploring innovative ways of optimizing our infrastructure. Our ambitious sustainability goal is to achieve net-zero emissions across our data centers by 2030, our offices by 2035 and our value chain by 2040 — and energy efficiency is crucial to these efforts.

This commitment is shared by one of our public sector data center clients who requested and supported the rollout of this pioneering pilot project at Rhine-Ruhr 1.

A feasibility study based on historical data revealed significant potential to save energy by focusing on the primary water–glycol circuit — the liquid cooling system that uses a mixture of water and glycol to absorb heat directly from technical components.

With the etalytics software, we could go even further, combining physical observations and data-driven AI models to optimize the operation of cooling systems. This hybrid approach is achieving substantial energy savings across industries, including data centers.

• ALSO READ → Water use in data centers is not a black-and-white issue

How the hybrid AI system works

To improve control of the data center’s refrigeration systems, etalytics creates a digital twin that includes all relevant cooling-circuit components, such as plate heat exchangers, chillers and pumps. Trained on historical operational data, the digital twin also incorporates physical models — thermodynamic, physical and electrical factors, like the electrical consumption of a fan increasing when the air flow doubles — for added accuracy.

It simulates optimization scenarios that account for factors such as the ambient temperature, varying setpoints (target values for key control parameters in the system) and changes in the flow rate of the chilled water circuit. It can even model an entire year’s operation in five-minute intervals, offering a detailed view of how the AI-based control behaves in real-world conditions.

The etalytics software can also extrapolate beyond an original data range. For example, even if a pump had operated at 100% capacity, the system can accurately predict performance at 80% capacity. This is where the hybrid approach excels: The data-driven model delivers precision, while the physical model ensures robustness and reliability.

A collaborative approach to integration

After the offline optimization phase — designed to prevent failures or disruptions in the production environment — the next step was integrating the etalytics software into our existing data center.

Collaborating with our building management and heating, ventilation and air-conditioning (HVAC) partners, we defined clear responsibilities for all stakeholders and developed a detailed roadmap. Our partners supported the process by updating the functional descriptions of all components and adapting the data center’s chiller control software.

The server running the etalytics platform is located in the data center, so the entire AI-based optimization process is hosted on-site.

Proactive measures to protect data center infrastructure

Because the optimization process affects critical infrastructure in real time, we implemented proactive safety measures to address potential risks and comply with our operational safety regulations.

These measures include a robust fallback solution. Optimization suggestions from the etalytics platform are not executed if:

• The optimizer is set to manual mode (for example, during an update or maintenance period)
• Communication between systems fails
• The defined setpoint ranges in the cooling system are exceeded

Our control-center team continuously monitors the process and can switch back to conventional controls at any time if a fault occurs.

From one pilot project to data centers everywhere

The first few months of AI-based optimization have already demonstrated the project’s success: Within the ambient temperature range covered during this time, our chillers have consumed 19.1% less energy. We are currently not expecting a significant deviation once the system completes a full year of operation.

Based on these results, we expect savings could reach up to 25% over a full year of operation — an outcome that has inspired us to scale up the use of AI in our data centers.

This pilot project has laid a strong foundation for implementing the etalytics software at other sites. Each site and system component will have to be evaluated individually, and the AI must be trained on site-specific operating conditions.

More innovation on the horizon

NTT Global Data Centers and etalytics are also jointly advancing a research initiative — funded by the German Federal Ministry for Economic Affairs and Climate Action — that extends the scope of AI optimization to include waste-heat recovery, further improving the energy efficiency of data centers.

• ALSO READ → Data centers’ waste heat holds untapped potential

As advanced AI systems increasingly rely on high-performance computing supported by direct liquid cooling, this technology will keep delivering significant value.

As Dr Thomas Weber, Co-Founder and CSO at etalytics, explains, liquid cooling systems already consume less energy by design. However, the optimization software achieves additional savings because its physical models are particularly well-suited to fluid-driven environments.

We’re excited to continue this journey and make our data centers around the world as energy efficient as possible.