Begin with meters and controls, then optimize airflow and supply temperature, and trim power-chain losses before investing capital in liquid cooling and heat reuse options. Establish PUE, ERE and WUE baselines, right-size IT and UPS capacity, seal containment, raise supply air as high as possible within ASHRAE guidelines, and calibrate economisers and VFDs. This sequence is recommended by DOE/FEMP for any energy efficiency upgrade project in a data center.
TL;DR:
- Monitor PUE, ERE, WUE, CUE at the rack, UPS, cooling and hall sensor level. DOE should define IT energy as the first item before reuse of heat, water, renewable sourcing.
- Perform leak tests on seals. Place copper and silver coupons inside for one month twice per year; good cleaning indicates that modest temperature increases of supply air are acceptable as long as they stay within ASHRAE guidelines.
- Measure UPS input and output prior to upsizing. Operating at low load reduces efficiency, while paralleling smaller modules can maintain redundancy and increase loading.
Reported 15% to 37% savings over static controls are possible with AI predictive controls. However, they require clean telemetry, staged pilots, safety limits, and operator supervision.
Pilot-scale liquid cooling first; provide capital funds for commissioning. Thermal reuse projects only pay off if there is warmer returning water that has some dependable, proximate demand and commercial terms have been negotiated.
Turn Data Center Telemetry Into Insight
PODTECH creates bespoke software for data centers, including telemetry and AI-based solutions, to assist in monitoring and decision-making.
Learn More >Table of Contents
- What can you start this quarter?
- Which metrics and telemetry do you need to measure first?
- How do you cut power conversion losses?
- How should you optimise airflow, containment and economisation?
- When do liquid cooling, rear-door exchangers and immersion make sense?
- How do controls and AI fit into energy optimization?
- Can you reuse waste heat from a data centre?
- What does a 30/90/365 day roadmap look like?
- Where do most programmes go wrong?
- How can PODTECH support your energy optimisation programme?
- FAQ
- Sources
What can you start this quarter?
The majority of operators will be able to achieve savings to bank ahead of investing in new plant. The following sequence is listed in order of speed of delivery relative to impact and time horizon:
- Pollutant coupons (low effort, weeks): put out copper and silver coupons to create supporting evidence for expanding the thermal envelope.
- Supply setpoints (high impact, weeks): increase in steps within the ASHRAE envelope as coupon results permit.
- Containment repair (medium to high impact, weeks): patch up gaps, missing blanking panels and cable cut-outs.
- VFD throttling (medium, weeks to months): slow pumps and fans to meet actual demand.
- AI controls pilot (medium to high impact, months): run it on one hall or one chilled-water loop.
- Small liquid-cooling rack (medium impact, months): use it to learn commissioning before scaling.
- UPS load reconfiguration (medium impact, months): consolidate load so units are not starving or running at very low load factor.
- Containment upgrades and CDU integration (high impact, months): budget these as planned projects.
- Immersion pilot and heat reuse infrastructure (high impact, 1+ year): manage these projects as capex with gates.
One thing leads to another. Improved metering tells you how high you can safely raise your setpoints. Knowing how high you can safely raise your setpoints tells you where containment leaks are costing you the most.
Which metrics and telemetry do you need to measure first?
You cannot manage what you only estimate. Four metrics cover the decisions that matter:
- PUE: total facility power divided by IT power. It informs cooling strategy and power-chain decisions.
- ERE: PUE adjusted for reused energy. It drives heat reuse decisions.
- WUE: water consumption per unit of IT power consumed. It is the basis for economiser and heat-rejection decisions.
- CUE: carbon emitted per unit of IT energy. It drives renewable sourcing.
DOE guidance prioritises actions: cut IT energy first, then maximise heat reuse, minimise water consumption and maximise renewables procurement. Metering, benchmarking and audit-driven improvements are also recommended by FEMP.
Build the metering hierarchy from the rack outward:
- Rack PDUs for IT load per rack.
- UPS input and output for conversion losses.
- Chilled-water flow and temperature for cooling plant performance.
- CDU supply and return where liquid cooling exists.
- Ambient, inlet and return-air sensors across each hall.
Sampling rates should be fast enough to observe control behavior, with intervals appropriate for the thermal and electrical points you are monitoring. Better granularity in sampling can improve controls and ML models. Most legacy enterprise facilities will operate around PUE 1.4 to 1.7. Advanced liquid-based architectures being demonstrated achieve around 1.04.
A DCIM layer transforms data streams into trends. See our DCIM case study for a real-world example of dashboards and ops data working together.
How do you cut power conversion losses?
Every watt wasted by the UPS and distribution is heat that your cooling plant has to remove. Electrical efficiency thus benefits you twice. A small increase in efficiency leverages throughout the year. If a site has an average IT load of 1 MW and increases UPS efficiency by one percentage point, that is approximately 10 kW saved continuously, or about 87,600 kWh per year before factoring in the savings from reduced cooling. The numbers above are examples, but the math applies to any facility.
Size is as important as what unit you pick. Because UPS efficiency is poor at low load factors, an oversized unit operating at a tiny percentage of its rated capacity will waste energy. Try these strategies:
- Scale out smaller UPS modules horizontally to maintain high unit load and redundancy.
- Right-size to measured IT load, not nameplate figures from commissioning day.
- Schedule maintenance carefully to avoid unnecessary transfers and cycling.
- Check distribution routinely: power factor, harmonics and transformer loading all leak energy quietly.
According to DOE guidance, improving conversion loss and matching supply to operational load are at the heart of data centre efficiency. You will find benchmarking tools in the FEMP toolkit to assist with this. Check out our explainer for more information on architecture and protocol-level specifics: power management system.
Think of the electrical chain as a measurement problem. Measure the UPS input and output, and conversion loss becomes a live number instead of a datasheet boast.
How should you optimise airflow, containment and economisation?
Efficiencies in air management represent the low-hanging fruit for the biggest savings, typically. The reason is that bypass and recirculation push the cooling infrastructure’s fans and chillers well beyond what is required for the IT load. ASHRAE recommends a move away from fixed set-points to measurement-led operation which allows operators to safely increase their temperature and humidity envelopes while depending less on compressors.
- Contain seals first. Eliminate gaps around racks, floor tiles, cable penetrations and open unused U-spaces so supply air can reach IT equipment inlets without leaking back.
- Prove increased supply-air temperatures with test data. ASHRAE has suggested running a simple program of copper and silver coupons for one month, twice annually, to evaluate your facility’s performance on gaseous contaminants. If your tests come back clean, you can justify increasing set-points toward the outside of the ASHRAE envelope.
- Size the economiser for the climate it serves. Airside economisers work well in climates with cool, clean air. Waterside economisers work best when outside air poses a contamination or humidity risk. Refrigerant-based economisers work well in retrofit situations where space constraints limit installation of the other two types. Calibrate changeover points to actual wet-bulb and dry-bulb conditions, not calendar dates.
- Vary fan and pump control. Variable frequency drives can be used to drive fans from differential pressure and pumps from load. Fan power drops off rapidly with decreased speed, so even slight slowing can be beneficial.
- Retire fixed set-points. Float supply temperature and airflow with load. Thermal-envelope review each coupon cycle.
ASHRAE’s template for AI halls takes the form of a “cooling-first” approach involving liquid cooling together with airflow optimisation, economisers and full monitoring.
When do liquid cooling, rear-door exchangers and immersion make sense?
Air cooling becomes insufficient when rack density exceeds what containment, and even larger fans, can exhaust. This is becoming more frequent: according to IEA, increased server acceleration is driving power density higher while making thermal efficiency a key feasibility factor. Options, listed in ascending order of intervention:
- Rear-door heat exchangers: the easiest retrofit, taking heat at the rack exhaust.
- Direct-to-chip: liquid plates on processors, with air still handling residual heat.
- Single-phase and two-phase immersion: the deepest change, suited to the highest densities.
Promising signs exist. Hyperefficient designs with PUEs nearing 1.04 are detailed in ARPA-E. Google has published its work showing CDU architectures tailored for dense racks.
Budget for commissioning as seriously as for hardware:
- CDU integration with the facility water loop.
- Flow and pressure management across manifolds.
- ASHRAE-focused commissioning verification. Commissioning agents caution that insufficient commissioning of liquid systems creates energy and dependability issues.
- Dielectric fluid choice, including global warming potential, water use and leak management.
How do controls and AI fit into energy optimization?
Controls are where measurement becomes continuous savings. According to one review from Springer, AI powered predictive control and reinforcement learning applied to HVAC and electrical systems can lead to meaningful reductions versus static control. At the same time, the IEA states AI can optimise complex energy systems, but notes barriers such as data availability, interoperability, and skills gaps.
“All models are wrong, but some are useful.” Start with useful inputs. Label your sensors and log what you are doing to the process, not just readings. Sample your fast loops at one minute or better. Next steps:
- Baseline current behaviour with clean telemetry.
- Automate rule-based controls such as setpoint resets and VFD curves.
- Move to model predictive control on one subsystem.
- Pilot reinforcement learning safely with a bounding box, hard safety constraints and immediate fallback to a rule-based layer.
Pro tip: Always run every model in shadow mode for several weeks, comparing suggested actions to operator decisions before it turns anything on or off.
Stay in the loop with operators. More important than algorithm complexity are staged validation, qualify-before-overwrite workflows, and defined authority to override.
Can you reuse waste heat from a data centre?
It somewhat depends on having a customer close by. District heating, greenhouses, office or campus heating, and domestic hot water are all fairly typical sinks. The primary technical consideration is temperature: air-cooled exhaust is very low-temperature, dirty heat, while liquid-cooled return water is hotter and vastly more usable.
Technical enablers consist of warm-water loops, heat pumps to boost temperature, organic Rankine cycle units to recover power, and controls integration so plant controls ensure reuse never jeopardizes cooling. We provide definitions and options in our glossary entry on waste heat reused.
Business model terms determine viability. Contracts for difference, seasonal variation in demand, and ownership of heat exchangers are examples of questions that need resolution at the front end. Quantify the output using ERE so that waste heat is accounted for. Rank waste heat recovery higher if you operate at a large scale, your primary cooling is liquid based, and there is steady demand nearby. Otherwise, earlier optimizations provide a quicker return on investment.
What does a 30/90/365 day roadmap look like?
Gate each phase on measured results, not calendar dates alone.
- Starting point: fill in all metering holes, install pollutant coupons, plug leaks and slowly raise setpoints. Establish a PUE baseline.
- Pilot phase: start pilots with specific KPIs, including an AI controls pilot on one loop, one liquid-cooled rack, and UPS load reconfiguration. Benchmark against each other and against baseline energy and temperature stability.
- Investment phase: invest in the pilots that justified it, whether immersion, heat reuse infrastructure, or a major UPS refresh, with ROI milestones along the way.
- Scale only when four criteria hold: proven energy savings, reliability not degraded, operational complexity is acceptable, and total cost of ownership makes sense over the asset life.
Where do most programmes go wrong?
Every single time this sad scenario appears: teams purchase equipment before they know how to measure the problem they are trying to fix. Monitoring, analytics and controls should get implemented first, as they determine where capex should actually be spent. Typical mistakes involve sparse telemetry, skipping vendor commissioning, and overlooking ERE and WUE while optimizing for a low PUE number. An integrated DCIM and telemetry layer is what guarantees savings, as it holds setpoints, containment and controls accountable well after the project team has disbanded.
— Harry
How can PODTECH support your energy optimisation programme?
Our implementation teams deliver against this roadmap, starting with baseline assessment and leading through telemetry deployment, pilot and managed rollout. Services that fall under this umbrella include:
- DCIM consultancy for metrics, dashboards and operating procedures.
- Datacenter telemetry covering rack, power and cooling points.
- Machine learning development for predictive control pilots.
- Master systems integration across BMS, PMS and DCIM.
Explore our datacentre services or contact us to book a baseline assessment.
FAQ
What is the fastest way to reduce data centre energy use?
Begin with metering, fixing leaks, and cautiously increasing supply-air setpoints inside the ASHRAE envelope. These require minimal investment and can show results in a matter of weeks. Rather than maintaining fixed cooling set-points, ASHRAE advises measurement-based operation.
Which metrics should I track for energy optimization?
Measure PUE for general efficiency, ERE for heat reuse, WUE for water and CUE for carbon. The order of priority in DOE guidance is IT energy, heat reuse, water, then renewables. Measure at rack, UPS and chiller plant level.
Can AI really cut cooling and power consumption?
Yes, if it is based on good telemetry and deployed gradually. Springer journal reviews show savings of 15% to 37% from AI predictive control of HVAC and electrical systems versus static control for the sites studied.
When should I move from air to liquid cooling?
Move when rack density exceeds containment and airflow removal capacity. IEA states AI servers are increasing power density and cooling requirements. Pilot the first rack to understand the commissioning process.
Sources
The article’s primary focus is on the utilization of artificial intelligence technologies to enhance building performance and optimize energy consumption.