Data centers and air filtration: Why air quality determines your cooling system's reliability
For facility managers and IT leaders, air filtration has become far more than just a building code requirement. It directly impacts energy efficiency, the physical security of critical infrastructure, and the lifespan of expensive hardware – making it a strategic lever that many operations still underestimate. The cooling technologies available for data centers – from traditional air cooling and adiabatic systems to liquid cooling in the form of direct-to-chip, or immersion cooling – have fundamentally different requirements when it comes to air and water quality. But they all share one thing in common: without a well-designed filtration strategy, none of them can perform at the level that mission-critical infrastructure demands in the age of big data and AI. This puts us in a unique position: we're among the few providers worldwide delivering filtration solutions for both air cooling and liquid cooling under one roof – a one-stop-shop advantage that data center operators are actively capitalizing on as part of their supplier consolidation efforts.
Why data center cooling is really a filtration issue
Data centers rank among the most energy-intensive infrastructures in the modern economy. The International Energy Agency (IEA) estimated the industry's global electricity consumption at around 415 terawatt-hours for 20241 – with projections reaching up to 945 terawatt-hours by 2030. The main driver isn't the computing hardware itself: server cooling systems consume anywhere from 30 to 60 percent2 of a facility's total energy use, depending on the setup. The industry average Power Usage Effectiveness (PUE) sits at 1.571 – while leading operators achieve values below 1.21. Beyond energy availability and location, cooling has emerged as the third decisive success factor for data center construction and operations – elevating it from an infrastructure side issue to a strategic imperative.
At the same time, thermal demands are skyrocketing. Cooling high-density data centers is no longer the exception – it's the norm. While conventional racks operate with five to ten kilowatts of heat output, AI-accelerated systems now generate 30 to over 100 kilowatts per rack3 Average global rack density doubled from eight to 17 kilowatts3 between 2020 and 2023 – a pivotal trend in data center cooling that's reshaping infrastructure planning entirely.
What gets discussed far too rarely in this conversation: The quality of the air flowing through a data center's HVAC (heating, ventilation, air conditioning) systems is just as critical as their technical specifications. Dust buildup on heat exchangers and cooling coils reduces heat transfer and degrades PUE. Corrosive gases in data centers – such as sulfur compounds, nitrogen oxides, or chlorine compounds – attack electronic components. Hardware reliability through dust protection and corrosion control in server rooms aren't optional features – they're fundamental requirements for reliable uptime.
Here's why: When processors and GPUs exceed their maximum operating temperature, modern chips automatically resort to thermal throttling – they reduce their clock speed to prevent damage. That means measurably less computing power at the same operating costs. Sustained overheating from inadequate cooling or clogged filter media also accelerates the aging of electronic components and demonstrably shortens equipment lifespan by up to two to four years.
In data centers with raised floor systems, there's another often underestimated contamination risk: zinc-coated steel floor panels can release microscopically fine zinc whiskers – needle-shaped zinc fibers – under mechanical stress. These get picked up by airflow, make their way into server racks, and can cause electrical shorts on circuit boards. Zinc whisker prevention therefore starts with filtration: high-efficiency particulate filters in data centers capture even these fine metallic fibers before they reach sensitive electronics.
Air filtration is thus an integral part of any more environmentally friendly data center cooling approach and measurably reduces operating costs at the same time.
Compliance with the ISO 14644-1 cleanroom standard – Class 8, the minimum value recommended by the US guideline ASHRAE TC 9.9 for data centers – requires a functional, multi-stage filtration strategy.
HVAC in data centers: Cooling systems and the role of air filtration
Climate control in server rooms and data centers follows a proven multi-layered approach in professional environments: Precision cooling systems like CRAC units (Computer Room Air Conditioner) and CRAH units (Computer Room Air Handler) are frequently deployed, working in conjunction with central chillers or Air Handling Units (AHU) depending on the design. The AHU conditions outside or recirculated air, filters out contaminants, and supplies the data hall with cooled supply air. CRAC units use direct expansion cooling and regulate temperature and humidity locally. CRAH units rely on chilled water coils combined with variable-speed EC fan technology – a more scalable and energy-efficient solution for high-density environments.
In each of these systems, the filter chain is critical. Insufficiently filtered air leads to dust buildup on cooling coils – a gradual process that reduces heat transfer and forces fans to work against increased resistance. Corrosive gases form sulfide corrosion whiskers on circuit boards – microscopic metal growths that trigger short circuits. The consequences range from processor and hard drive failures and voided warranties to electrical fires in UPS (uninterruptible power supply) areas.
The ISA standard 71.04-2013 classifies corrosive environments into four levels: from G1 (mild) to GX (extreme). ASHRAE TC 9.9 recommends ISO 14644-1 Class 8 as the minimum standard3. For filter selection, the guideline is: MERV 8 filters for recirculated air, MERV 11 to 13 for outside air without economizers. With free cooling in data centers using air-side economizers, requirements increase significantly – since unconditioned outside air enters the facility directly, dedicated particulate filters and site-specific configurations become mandatory.
Differential pressure monitoring of the HVAC system enables a proactive maintenance strategy: Rising pressure differential across filter stages signals clogging early and prevents uncontrolled bypass flows before they impact cooling performance.
Data center solutions from MANN+HUMMEL Air Filtration
Our portfolio provides air filtration for secure data centers. With MANN+HUMMEL Air Filtration, you get comprehensive filter solutions from a single source for every application:
Pre-filters are the first active filter stage within the AHU and serve a dual function: They capture coarse particles, heavy dust, and fibers from the airstream while simultaneously protecting all downstream filter stages from premature loading. By intercepting the bulk of the particulate load in the first step, they significantly extend the service life of the more expensive fine and final filters – a key lever for reducing operating costs. Pre-filters deployed in data centers must withstand special mechanical demands: Steel wire grids as media support ensure dimensional stability even at high air velocities and pressure spikes, while moisture-resistant frames enable use in outside air intakes and adiabatic cooling systems. Efficiency ratings from MERV 9A to MERV 10A ensure that pre-filtration already supports the requirements of ASHRAE TC 9.9 for basic supply air cleaning. The self-sealing construction is particularly important: It prevents bypass airflows around the filter media, which can significantly reduce actual separation performance even when nominal filter efficiency is correct.2
- Prime9 XT
- Prime10 WL
- Tri-Dek® Cubes
While particulate filters remove visible and measurable airborne particles, molecular pre-filters combat the invisible threat: gaseous contaminants like sulfur compounds, nitrogen oxides, chlorine compounds, and organic vapors that pass through conventional filter media unimpeded and cause corrosion on metallic connections on circuit boards. This filter class combines both particulate and molecular filtration in a single filter media – through a hybrid media construction of impregnated activated carbon layers and fiber filtration that captures both contaminant groups in one step. Deployment is strategic within the AHU filter chain, typically as an upstream protective layer before high-efficiency filters or as a standalone stage in systems with documented gas contamination in the supply air – classified according to ISA-71.04-2013 starting at corrosivity class G2. Different carbon loadings, carbon blends, and housing geometries – from compact panel filters and pleated configurations to V-bank designs with up to eleven kilograms of activated carbon per filter – allow precise matching to the site-specific gaseous contaminant profile. The fact that this filter stage is still missing in many data centers represents an underestimated risk: even at supposedly clean urban locations, seasonal gas contaminant spikes or industrial emissions from the surrounding area can build up measurable corrosion loads inside the facility within just a few months.2
- MC Prime Serie
- MC TL Serie
- MC VB4 15
- MC VB4 HC
In data centers, fine particulate filtration is the critical barrier between outside air and sensitive server hardware. High-efficiency particulate filters perform this key function within the AHU: They capture fine aerosols, dust particles, and airborne microbiological matter that would otherwise settle on cooling coils or circuit boards and permanently impair performance. The filter media deployed in this category – from microglass to high-performance synthetic fibers – are designed for low pressure drop at high separation efficiency, which directly impacts fan energy consumption and thus the PUE of the entire facility. Efficiency ratings from MERV 11 to MERV 16A cover different requirement profiles – from standard office data centers to critical hyperscale and AI infrastructures. Especially in high-density environments where any restriction of airflow immediately leads to hotspots, these filters offer maximum dust-holding capacity at consistently low resistance throughout their service life through their tapered pocket or mini-pleat design.2
- ProPocket DC Elite
- ProPocket Hybrid
- ProPocket Nano+
- Thin Line Serie
- Thin Line Ultra Max Frameless
Final filters form the last filter stage before supply air enters the data hall and thus the final protective barrier for server racks and IT hardware. They typically come in V-bank construction, whose large filter media surface area enables particularly low pressure drop – a decisive advantage in variable air volume (VAV) systems where pressure fluctuations can compromise the filtration effectiveness of other filter types. The frames in this filter class are made from high-impact polystyrene (HIPS) without metallic components, making them fully incinerable at the end of their service life and thereby contributing measurably to landfill waste reduction. Available in efficiency ratings from MERV 11 to MERV 16, final filters combine high dust-holding capacity with long service intervals – both essential factors for low total cost of ownership in mission-critical operating environments. Aerodynamically optimized vertical supports in the filter frame additionally minimize air inlet turbulence, which improves airflow uniformity across the entire filter pack.2
- VB4 Elite
- VB4 Ultra
- VB2 Elite
- VB2 Ultra
Liquid cooling in data centers: What’s changing
Air cooling will continue to dominate data centers in the near term – however, liquid cooling is poised to become the leading cooling technology over the medium to long term. This transformation is already taking shape in hyperscale and AI data centers around the globe. The shift to AI-intensive workloads is making conventional air cooling increasingly uneconomical in high-density environments. Liquid-cooled servers with direct-to-chip cooling and rear-door heat exchangers are gaining importance, as are immersion cooling systems where hardware is fully submerged in dielectric fluids.
Further insights from our CEO:
In the DeepDive CleanTech podcast, MANN+HUMMEL President and CEO Kurk Wilks discusses how advanced filtration technologies can help data centers reduce energy and water consumption while supporting reliable and efficient cooling operations.
Water cooling and hybrid systems in data centers: Higher density, higher demand
Hybrid data centers combine air cooling for moderate loads with water-cooled servers for high-performance components. Adiabatic cooling and water quality are inseparably linked: Adiabatic systems significantly reduce chiller energy consumption through water evaporation, but require carefully treated water – scale deposits and microbial growth compromise wetting surfaces and increase Legionella risk. Makeup water treatment in data centers isn't an optional step here – it's an operational requirement.
Adiabatic systems are energetically attractive but increasingly viewed as problematic in water-scarce regions: depending on climate zone and utilization, they consume several million liters of water per year. Given growing ESG requirements and regulatory reporting obligations, this water consumption – expressed as Water Usage Effectiveness (WUE) – is increasingly factored into data center sustainability assessments.
Sidestream filtration at the cooling tower keeps suspended solids and corrosion products out of the loop and measurably extends heat exchanger service life. To prevent heat exchanger fouling, multi-stage water treatment is also recommended: from mechanical pre-filtration to chemical conditioning of cooling circuits. A complete water quality management approach for data centers also includes treating incoming water through water treatment systems and handling wastewater – both standard applications supported by dedicated MANN+HUMMEL Water & Membrane Solutions components.
Here's what matters: Even in fully liquid-cooled facilities, air filtration remains relevant – for equipment areas, maintenance zones, and staff indoor air quality. The combination of rigorous air filtration and fluid management enables operators to run higher rack densities, meet sustainability goals, and reliably document compliance with regulatory requirements under ANSI/TIA-942, and IEC-60721-3-3-2019.
Stationary energy storage in data centers: BESS, Redox Flow, and fuel cells
Power supply reliability in a data center doesn't depend solely on cooling – it begins with uninterruptible delivery of electrical energy. Beyond traditional UPS systems, operators are increasingly adopting Long Duration Energy Storage (LDES), Battery Energy Storage Systems (BESS): large-scale stationary energy storage that acts as a buffer between the power grid and IT infrastructure, seamlessly taking over during grid outages.
These systems also impose specific filtration and protection requirements. From our proven portfolio – originally developed for eMobility applications, now equally suited for stationary applications – we offer targeted solutions:
- ProVent battery vents protect BESS enclosures and battery modules from dangerous pressure buildup due to outgassing while simultaneously controlling moisture ingress.
- Cathode water separators and cathode air filters protect fuel cell stacks in backup systems from contaminants, moisture, and particulate pollutants in the intake air.
- Membrane humidifiers for fuel cells maintain membrane ionic conductivity, reliably extending the service life of fuel cell systems.
Redox Flow Batteries (RFBs) represent a highly promising technology in stationary energy storage focusing on Long Duration Energy Storage (LDES): unlike conventional lithium-ion batteries, they store energy in liquid electrolytes that circulate in external tanks – scalable, durable, and non-flammable. MANN+HUMMEL Water & Membrane Solutions offers MIONEX, a proprietary ion exchange technology deployed in Redox Flow Batteries that contributes to electrolyte system stability and efficiency.
Power supply reliability in a data center starts with the uninterrupted delivery of electrical energy. Beyond traditional UPS systems, operators are increasingly adopting Battery Energy Storage Systems (BESS): large-scale, stationary storage that acts as a dynamic buffer between the grid and IT infrastructure, bridging outages seamlessly and stabilizing short-term fluctuations.
As power demand from data centers continues to surge, driven by hyperscale and AI workloads, grid infrastructure is often unable to keep pace. This is where Long Duration Energy Storage (LDES) becomes a strategic enabler. Unlike conventional BESS, LDES solutions provide extended discharge capabilities, aligning with data center load profiles over hours or even days. They ensure operational continuity not only during brief interruptions but also through prolonged grid constraints or instability.
Within this landscape, Redox Flow Batteries stand out as a highly flexible LDES technology. In addition to delivering long-duration energy >4hours, RFBs can also respond rapidly to short-term load peaks, such as those generated by AI-driven data center workloads. Their ability to decouple power and energy capacity enables efficient handling of both sustained demand and sudden spikes, providing a dual benefit of peak shaving and extended backup in a single system.
Crucially, LDES solutions - including RFBs - can be deployed significantly faster than grid expansion projects, offering a near term solution to capacity bottlenecks. By decoupling data center growth from grid readiness, operators gain the flexibility to scale infrastructure without compromising reliability. In combination with BESS and UPS, LDES forms a resilient, multi-layered energy architecture - ensuring consistent power delivery even when the grid cannot meet demand.
Design and TCO: Optimizing PUE through filtration and the right strategy
Filtration isn't a one-time investment decision but an ongoing strategy. For facility managers, it's worth examining three dimensions: protective effectiveness, energy efficiency, and Total Cost of Ownership (TCO).
MANN+HUMMEL Air Filtration for cleaner air in data centers
The physical security of a data center doesn't start at the server rack – it starts with the air flowing through your facility. As one of the few global filtration specialists delivering both air cooling and liquid cooling solutions from a single source, we are happy to develop the right filtration-integrated cooling solution for your data center: from site analysis and corrosivity assessment through media selection to ongoing quality monitoring.
For more than 80 years, MANN+HUMMEL has been recognized worldwide for innovative, highly effective filtration solutions. We ensure cleaner mobility, cleaner air, cleaner water, and cleaner industry. With our concepts and solutions, data center cooling is secure, high-performing, and future-ready. Contact us directly for a personalized initial consultation.
FAQs on filtration in data centers
Servers, storage systems, and network components convert electrical energy into computing power – inevitably generating waste heat as a physical byproduct of this process. Without continuous heat removal, processors and other components exceed their maximum allowable operating temperatures within a short time, initially leading to performance throttling (thermal throttling) and, with sustained overheating, to permanent hardware damage. Modern AI accelerator cards reach power dissipation levels of several hundred watts per chip – at the rack level, this adds up to 30 to over 100 kilowatts. Additionally, elevated operating temperatures increase the error rate of electronic components and shorten equipment lifespan – by up to two to four years when cooling and filtration requirements aren't met. Reliable cooling is therefore not a comfort feature but the mission-critical foundation for availability, hardware protection, and economic viability of any data center infrastructure.
Liquid cooling refers to the use of cooling media – typically water, water mixtures, or dielectric fluids – for direct heat removal from IT components or server racks. Unlike conventional air cooling, which transports heat via conditioned supply air, liquid cooling exploits the significantly higher heat capacity of liquid media. Water can absorb many times the amount of heat per unit volume compared to air, enabling more compact and efficient cooling. Liquid cooling is primarily deployed where very high rack densities – such as AI servers with 30 to over 100 kilowatts per rack – make adequate heat removal through air challenging. Common implementations include direct-to-chip cooling (DLC), rear-door heat exchangers, and immersion cooling.
A blanket assessment isn't possible here – the optimal cooling strategy depends on rack density, workload type, and operational constraints. Air cooling remains cost-effective and well-established for moderate heat loads up to approximately 15 to 20 kilowatts per rack. However, for AI-accelerated high-performance applications with 30 kilowatts and above per rack, it reaches physical limits as fan energy demand increases exponentially and hotspots emerge. Liquid cooling enables lower PUE values and more precise temperature control in such scenarios. In practice, modern data centers therefore often adopt hybrid approaches that strategically combine both technologies.
In liquid cooling, a liquid medium absorbs waste heat from IT components (Technology Cooling System = TCS loop) and transports it through a circuit to a heat exchanger or chiller, where the energy is released and the medium is cooled again (Facility Water System = FWS loop). In direct-to-chip cooling, cold plates are mounted directly on processors or GPUs with cooling water flowing through them. Rear-door heat exchangers capture the hot exhaust air from a rack at the outlet and cool it before it flows back into the data hall. In immersion cooling, servers or components are fully submerged in dielectric fluids that absorb heat directly from component surfaces. The heated fluid then passes through an external heat exchanger, where it's cooled and recirculated back into the loop.
Power Usage Effectiveness (PUE) measures a facility's total energy consumption relative to the energy actually used for the IT load – a PUE of 1.0 would be theoretically ideal and mean zero overhead. In practice, the global industry average sits around 1.57; state-of-the-art data centers achieve values below 1.2. Hyperscale operators with optimized cooling systems regularly report PUE values between 1.1 and 1.2. Cooling is the biggest lever here: HVAC systems account for 30 to 60 percent of total energy consumption, which is why every efficiency improvement in the cooling system directly impacts PUE. Filter optimization also makes a measurable contribution: reducing filter pressure drop from 0.72 in. w.g. to 0.3 in. w.g. noticeably lowers fan operating power and thereby improves PUE.
In immersion cooling, servers or individual hardware components are fully submerged in an electrically non-conductive (dielectric) fluid that absorbs heat directly from component surfaces. Since the cooling medium is in direct contact with the electronics, heat transfer is more efficient than any indirect method. Single-phase immersion is distinguished from two-phase immersion: in single-phase, the liquid remains liquid and is cooled via an external heat exchanger, while in two-phase, the fluid evaporates at hot surfaces, recondenses at a condenser, and flows back. Immersion cooling enables very compact rack designs with high heat densities and can achieve PUE values near 1.0. The purity and quality of the dielectric medium must be continuously monitored, as contaminants can gradually compromise the fluid's electrical insulation properties.
In water-cooled systems – whether chilled water circuits, adiabatic cooling, or cooling tower loops – water quality directly affects the efficiency and operational reliability of the entire cooling system. Scale deposits on heat exchangers increase thermal resistance and gradually reduce cooling capacity without being immediately apparent. Microbial growth – including Legionella bacteria – poses a serious health risk to operating personnel and leads to biofouling in pipes and cooling heads. Corrosion products from untreated water settle in microchannels of cold plates and heat exchangers and can permanently damage them. Regular makeup water treatment, sidestream filtration, and chemical circuit conditioning are therefore integral components of professional data center operation.
Air cooling uses conditioned air as the heat transfer medium: CRAC or CRAH units cool the warm air from the data hall, which is then returned to the servers to absorb their waste heat. Liquid cooling, in contrast, uses water or dielectric fluids that absorb heat either indirectly via cold plates on components or directly through hardware immersion. The fundamental physical difference lies in heat capacity: liquids transport many times the amount of heat per unit volume compared to air. Air cooling is proven and cost-effective for moderate rack densities but reaches its limits with high-density AI workloads. Liquid cooling is more energy-efficient at high heat loads but requires additional infrastructure, rigorous fluid management, and higher capital investment.
Evaporative cooling – also called adiabatic cooling – exploits the physical effect that evaporating water extracts energy from ambient air, thereby cooling it. In data centers, this principle is used to pre-cool outside air before it enters the chillers or data hall – reducing the operating hours of energy-intensive compression refrigeration equipment. Adiabatic cooling can substantially lower cooling energy demand in temperate climates and contribute to significantly lower PUE values. The drawback is water consumption: adiabatic systems require continuously treated water to prevent scale deposits, contamination, and Legionella risks in humidification systems. In water-scarce regions or under strict ESG mandates, the water consumption of these systems is increasingly viewed critically.
CRAC (Computer Room Air Conditioner) and CRAH (Computer Room Air Handler) are both precision air conditioning units for server rooms, but they differ fundamentally in their cooling technology. A CRAC unit has its own integrated refrigeration circuit using direct expansion – it cools warm air directly via an internal evaporator and returns conditioned air to the room, similar to a standalone air conditioner. A CRAH unit has no refrigeration circuit of its own but instead uses externally supplied chilled water that flows through internal coils to cool the warm air. CRAH units are more scalable, typically more energy-efficient, and better suited for large or high-density data centers since they benefit from centralized chilling. CRAC units are more commonly found in smaller server rooms or as decentralized supplements in existing infrastructures.
