Smarter Data Centre Cooling: Why Energy Efficiency Must Be Measured, Verified and Monitored
Smarter Data Centre Cooling: Why Energy Efficiency Must Be Measured, Verified and Monitored
Artificial intelligence may be contributing to the rapid expansion of data centres, but researchers are also exploring how AI can reduce the environmental burden of operating them.
Researchers at Penn State have developed physics-informed artificial intelligence software designed to optimise data centre cooling according to changing weather, humidity, electricity prices and operating conditions.
In simulated testing, the system reduced cooling energy consumption by more than 24% over a 24-hour period while maintaining the temperature and operational limits required to protect computing equipment. The researchers believe the same approach could eventually be applied to other commercial buildings and industrial facilities.
This development illustrates an important principle for the emerging data centre industry:
Environmental performance is not determined only by how much infrastructure is built. It is also shaped by how intelligently that infrastructure is operated.
For the Caribbean, where high temperatures, humidity, electricity costs, grid constraints and water availability can all affect data centre performance, cooling efficiency should form a central part of project design, environmental assessment and long-term operational monitoring.
Why data centres require so much cooling
Servers, processors, storage systems and networking equipment convert much of the electricity they consume into heat.
That heat must be continuously removed. Excessive temperatures can reduce equipment performance, shorten component life and lead to system failures.
Data centres therefore depend on cooling infrastructure that may include:
- Air-conditioning systems
- Chillers
- Cooling towers
- Pumps and fans
- Air-handling equipment
- Liquid-cooling systems
- Heat exchangers
- Humidity controls
- Temperature and airflow sensors
Penn State researchers estimate that cooling can account for approximately 40% of a data centre’s total electricity consumption. The US Department of Energy has similarly reported that cooling may represent up to 40% of overall data centre energy use.
This means that cooling is not a minor operational cost. It is one of the largest opportunities for reducing electricity demand, emissions and operating expenditure.
The limitations of static cooling
Traditional cooling systems are often managed using fixed temperature and humidity targets.
This conservative approach is intended to protect equipment. However, it may also cause facilities to cool server rooms more aggressively than necessary, even when external conditions, computing demand or electricity prices change.
A static operating strategy may fail to account for:
- Cooler periods during the day or night
- Changes in outdoor temperature and humidity
- Variations in computing workloads
- Different thermal conditions across server racks
- Changes in electricity prices
- Available renewable-energy generation
- Equipment-specific operating tolerances
- Differences between local climates
The Penn State research team sought to address this problem by developing a system that learns how to adjust cooling without exceeding accepted hardware safety limits.
The objective is not to reduce cooling indiscriminately. It is to provide the required cooling more precisely.
How the new system works
The software uses a physics-informed reinforcement learning model.
Researchers first create a digital twin—a virtual representation of a data centre and its cooling systems. Within this simulated environment, variables such as temperature, humidity, equipment limits, electricity prices and facility operating conditions can be adjusted.
An AI agent is then trained inside the digital environment to recommend cooling actions that improve energy efficiency while respecting the operational limits of the physical equipment.
For the reported study, researchers simulated a data centre in Houston, Texas, where high temperatures and humidity create demanding cooling conditions.
The resulting system was designed to make recommendations based on real-time conditions while balancing:
- Equipment safety
- Cooling performance
- Electricity consumption
- Operating cost
- External weather
- Humidity
- Facility workload
- Economic conditions
Because the physical operating constraints are built into the model, the system can explore more efficient cooling strategies without instructing equipment to operate outside accepted temperature or humidity ranges.
Why a 25% reduction matters
A reduction of approximately 25% in cooling energy does not necessarily mean that the entire facility’s electricity consumption falls by 25%.
The saving applies to the cooling component.
However, because cooling may represent a substantial share of a data centre’s electricity use, the overall effect can still be significant.
Lower cooling-energy demand may result in:
- Reduced electricity consumption
- Lower operating expenditure
- Lower indirect greenhouse gas emissions
- Reduced peak electricity demand
- Less pressure on generation and grid infrastructure
- Improved power usage effectiveness
- Greater capacity to support additional computing equipment
- Reduced demand on backup power systems
- Better alignment with renewable-energy availability
These improvements become increasingly important as global data centre demand expands.
The International Energy Agency reported that worldwide data centre electricity consumption grew by approximately 17% in 2025. It projects that total data centre electricity use could rise from approximately 485 terawatt-hours in 2025 to around 950 terawatt-hours by 2030. Electricity consumption from AI-focused facilities is projected to grow even faster.
Efficiency improvements can reduce the energy needed to deliver a given level of digital service. However, rapidly rising demand means efficiency alone may not prevent total electricity consumption from increasing.
Both expansion and efficiency must therefore be monitored.
Caribbean climate conditions make cooling critical
Data centre cooling requirements depend heavily on location.
Facilities in cooler climates may be able to use outside air or other low-energy cooling methods during parts of the year. Tropical facilities operate under very different conditions.
Caribbean data centres may face:
- High year-round temperatures
- High humidity
- Salt-laden coastal air
- Intense rainfall
- Flood and storm risks
- Water-supply constraints
- High electricity costs
- Limited reserve generation capacity
- Greater dependence on mechanical cooling
- Vulnerability to grid interruptions
A cooling strategy developed for a temperate location cannot automatically be assumed to perform equally well in Trinidad and Tobago or elsewhere in the Caribbean.
Local climate data, equipment specifications, utility conditions and environmental constraints must be built into the facility design and operating model.
The location-specific structure of the Penn State system is therefore especially relevant. Its digital twin can be trained using the conditions affecting a particular facility rather than relying exclusively on a generic cooling schedule.
Energy efficiency claims must be verified
A projected or simulated reduction is an important research result, but it is not the same as verified operational performance.
Once an optimisation system is deployed in a working data centre, its performance should be measured over an appropriate period and across a variety of conditions.
Verification should examine:
- Total facility electricity use
- Cooling-system electricity use
- Server and information-technology electricity use
- Peak electricity demand
- Indoor temperature and humidity
- Outdoor temperature and humidity
- Computing workload
- Equipment uptime and reliability
- Water consumption
- Backup-generator operation
- Seasonal performance
- Greenhouse gas emissions
- Performance during unusually hot periods
Measurements should also establish what would have occurred without the new system.
Without a reliable baseline, it may be difficult to determine whether reduced electricity use resulted from the optimisation software, lower computing demand, cooler weather, equipment replacement or another operational change.
Useful performance indicators
Data centres can use several indicators to assess efficiency and environmental performance.
Power usage effectiveness
Power usage effectiveness compares total facility electricity consumption with the electricity used by the information-technology equipment.
A lower value generally indicates that less energy is being consumed by supporting infrastructure such as cooling, lighting and power conversion.
However, this indicator should not be considered alone. It does not fully describe water use, electricity sources, greenhouse gas emissions or the useful computing work delivered.
Cooling-system energy consumption
Cooling energy should be measured separately where possible.
This allows operators to determine whether improvements come from:
- Better controls
- More efficient chillers
- Improved airflow
- Higher allowable operating temperatures
- Liquid cooling
- Reduced recirculation
- Better server-room layout
- Weather-responsive operation
Water usage
Some cooling systems save electricity by using more water.
An efficiency programme should therefore ensure that a reduction in electricity consumption does not create an unacceptable increase in water demand.
Water monitoring should identify:
- Source-water volumes
- Potable and non-potable water use
- Cooling-tower make-up water
- Evaporation losses
- Blowdown water
- Recycled water
- Wastewater quality
- Discharge volumes
Carbon intensity
The environmental value of an electricity saving depends partly on the source of the electricity.
Reducing consumption during a period supplied by high-emission generation may produce a different carbon benefit from reducing the same amount when low-carbon generation is abundant.
Advanced operating systems could potentially incorporate grid carbon intensity, renewable-energy availability and peak-demand conditions into future optimisation decisions.
Energy efficiency does not replace environmental assessment
Efficient cooling is a positive design feature, but it does not resolve every environmental issue associated with data centres.
A comprehensive environmental assessment may still need to consider:
- Total electricity demand
- Source of power
- Greenhouse gas emissions
- Water abstraction
- Wastewater and thermal discharge
- Backup-generator emissions
- Fuel storage
- Noise
- Electronic waste
- Batteries and hazardous materials
- Stormwater
- Flood risk
- Land use
- Traffic
- Community impacts
- Cumulative effects from multiple facilities
A facility may use advanced cooling controls and still create significant impacts if it is poorly located, depends on carbon-intensive generation, consumes scarce water resources or lacks an effective waste-management plan.
Efficiency should therefore be treated as one component of responsible development—not as a substitute for environmental safeguards.
The importance of continuous monitoring
Optimisation software depends on data.
Temperature sensors, humidity sensors, electricity meters, water meters, equipment controls and external weather information all help the system understand current conditions and recommend appropriate actions.
Environmental monitoring serves a related role at the facility and community level.
A robust monitoring programme can help determine whether the project is meeting its commitments for:
- Electricity efficiency
- Water consumption
- Air emissions
- Noise
- Waste management
- Wastewater quality
- Thermal discharge
- Equipment reliability
- Environmental compliance
Monitoring also makes adaptive management possible.
When conditions move outside an approved limit, operators can investigate the cause, adjust equipment and verify whether corrective action was effective.
From smart software to accountable performance
The Penn State research demonstrates how artificial intelligence and digital-twin technology may help reduce the cooling burden of data centres.
It also highlights a wider environmental lesson.
Technology can support more efficient operations, but environmental benefit should be demonstrated through measurable results.
For proposed data centres, developers should establish clear performance commitments before construction. These may include:
- Maximum cooling-energy intensity.
- Water-consumption targets.
- Temperature and humidity operating limits.
- Renewable or lower-carbon electricity targets.
- Peak-demand management plans.
- Minimum monitoring and reporting requirements.
- Independent performance verification.
- Procedures for investigating deviations.
- Periodic review as equipment and workloads change.
- Public reporting of relevant environmental indicators.
This turns sustainability from a general promise into an auditable management system.
How Ecotox can support responsible data centre development
Ecotox Environmental Services can support infrastructure developers, regulators and facility operators through services such as:
- Environmental baseline studies
- Ambient air-quality monitoring
- Noise monitoring
- Surface-water and groundwater testing
- Wastewater characterisation
- Soil and sediment sampling
- Waste characterisation
- Environmental compliance monitoring
- Ecological assessment
- Environmental impact assessment support
- Sampling-plan development
- Independent environmental performance verification
For a data centre, these services can complement internal energy-management systems by examining environmental conditions beyond the server room.
Learn more about Ecotox Environmental Monitoring and Assessment Services.
Efficiency must be designed for local conditions
The environmental implications of expanding data centres cannot be addressed through a single technology.
More efficient processors, improved software, renewable energy, better cooling systems, water reuse, appropriate site selection and credible environmental monitoring will all have roles to play.
The reported potential to cut cooling energy by approximately 25% is encouraging. It shows that smarter operation may reduce part of the energy burden associated with digital infrastructure.
The next challenge is real-world verification.
As data centre investment expands, operators must demonstrate that projected savings are achieved under actual workloads, weather conditions and operating constraints.
For the Caribbean, this means designing and testing cooling strategies against tropical heat, humidity, electricity conditions, water availability and climate risk.
The most credible data centre will not simply claim to be efficient.
It will measure its performance, disclose its results and continuously improve them.
Sources
Penn State — New software could cut cooling energy use by 25% in data centers
https://www.psu.edu/news/research/story/new-software-could-cut-cooling-energy-use-25-data-centers
Environmental News Network — New Software Could Cut Cooling Energy Use by 25% in Data Centers
https://www.enn.com/articles/77892-new-software-could-cut-cooling-energy-use-by-25-in-data-centers
International Energy Agency — Key Questions on Energy and AI
https://www.iea.org/reports/key-questions-on-energy-and-ai