Data Centre Cooling Demand Drives Growth in Butterfly & Ball Valve Applications
Data centre cooling loads have surged with AI infrastructure expansion, driving a step-change in flow control demands — butterfly and ball valves are now specified in the millions annually for hyperscale facilities.
Triple-offset butterfly valves and full-bore ball valves each have distinct strengths in cooling applications: understanding which to use, and when, directly impacts energy efficiency, maintenance intervals, and system uptime.
Correct valve selection for data centre HVAC and liquid cooling circuits requires detailed attention to Cv ratings, pressure drop, actuation speed, and material compatibility with glycol-based coolants.
Why Data Centres Have Become a Critical Market for Valve Engineers
If you’ve been paying attention to the flow control sector over the past five years, you’ll have noticed that data centres have quietly become one of the most technically demanding application environments valve engineers now work in. This isn’t the modest server room of the early 2000s — we’re talking about hyperscale facilities consuming 50–200 MW of power, with cooling systems that rival industrial process plants in complexity. And every single one of those systems runs on valves.
The explosion of AI computing infrastructure has been the primary accelerant. Training large language models and running inference at scale generates extraordinary heat densities — rack power densities have climbed from 5–10 kW per rack in traditional enterprise IT to 30–100 kW per rack in modern GPU clusters, with some liquid-cooled AI deployments exceeding 150 kW per rack. That heat has to go somewhere, and it has to be managed with precision. Flow control isn’t optional; it’s load-bearing infrastructure.
Global data centre construction is running at hundreds of billions of dollars annually, with the pipeline of committed projects stretching years into the future across Northern Europe, the US, Southeast Asia, and the Middle East. Each of those facilities represents a substantial valve procurement opportunity — and more importantly, a technically interesting engineering problem that rewards proper specification over commodity buying.
Walk into the plant room of any major hyperscale data centre and you’ll find two valve types doing the majority of the work: butterfly valves and ball valves. They’re not interchangeable — each has a specific role it performs exceptionally well — but together they cover virtually the entire valve demand within a typical cooling circuit.
Butterfly valves dominate large-bore isolation and control applications. When you need to isolate a 400 mm chilled water header or throttle flow through a plate heat exchanger circuit, a concentric or high-performance butterfly valve is almost always the right answer on cost, weight, and footprint grounds. They’re light, compact, quick to actuate, and available in sizes that make a comparable ball valve genuinely impractical.
Ball valves are the weapon of choice for smaller-bore service connections, metering, and any application where zero-leak shutoff and full-bore flow are non-negotiable. In direct liquid cooling (DLC) circuits that route coolant directly to server components, ball valves handle the branch connections, bypass loops, and isolation duties throughout the distribution manifold.
Understanding which to deploy where — and to what specification — is where the engineering value lies.
The butterfly valve family is not monolithic. In data centre applications, engineers typically encounter three main subtypes, and they are not equivalent.
Concentric (centric) butterfly valves are the workhorse of HVAC cooling systems. The disc is centred on the shaft, which runs through the centre of the pipe bore. Simple, cost-effective, and suitable for lower-pressure water and glycol service, they’re widely used on chilled water primary loops and cooling tower bypass circuits where tight shutoff isn’t critical. The seat is typically EPDM rubber, which offers good compatibility with treated cooling water and glycol mixtures up to around 50% concentration. Pressure ratings are modest — typically PN10 or PN16 — which is entirely adequate for most data centre HVAC pressure regimes. Leakage to atmosphere at shutoff is acceptable to Class IV (EN 12266-1) rather than Class VI, which is fine for isolation but inadequate for tight throttling.
Double-offset (high-performance) butterfly valves introduce two offsets to the shaft geometry that pull the disc away from the seat on opening, dramatically reducing wear. They’re the standard choice for modulating control applications on chilled water variable flow systems, where the valve cycles continuously in response to differential pressure controllers and flow setpoints. The tighter shutoff characteristics — typically Class IV to Class VI depending on seat material — make them suitable for pump isolation and bypass duties where bubble-tight sealing matters. You’ll find them on cooling distribution units (CDUs), adiabatic coolers, and the interconnecting pipework of precision air conditioning (PAC) units.
Triple-offset butterfly valves (TOVs) are the high end of the butterfly family: metal-seated, cam-geometry closure, genuinely bubble-tight shutoff to ASME Class VI / EN 12266-1 Class A. In data centre applications their use is more selective — you’ll find them on high-pressure chiller bypass circuits, emergency shutoff positions on critical cooling headers, and any application where fire-safe performance is required by the specification. The torque characteristics of a TOV are fundamentally different from a concentric valve: because the disc engages its seat in the final 2–3 degrees of travel via a cam action rather than a compressive force, actuation torque is lower and more predictable, which matters for sizing electric actuators correctly.
Sizing butterfly valves for cooling applications requires careful attention to Cv (flow coefficient) and rangeability. A butterfly valve at 90° open has a high Cv — essentially full bore flow — but the relationship between angle and Cv is highly non-linear. Butterfly valves have a usable control range of roughly 20°–70° (in some designs up to 80°). Outside this range, you’re either barely cracking the disc open (unstable, cavitation risk, poor control resolution) or you’re wide open with no authority left. For variable flow chilled water systems, this means properly matched actuator travel and positioner calibration is essential. An improperly calibrated butterfly control valve hunting between 5° and 15° open is destroying its seat and providing appalling control — I’ve seen this more than once on commissioning visits.
Material selection for the cooling medium matters. Most data centre chilled water systems operate with a glycol inhibitor package — typically 20–35% ethylene or propylene glycol with a corrosion inhibitor package designed for aluminium, copper, and mild steel compatibility. EPDM seats handle this well at temperatures between -20°C and +120°C. PTFE-lined seats offer enhanced chemical resistance for more aggressive inhibitor formulations. The body should be ductile iron as a minimum; specify LGC (low zinc) brass or stainless steel trim for systems where dezincification is a concern.
Ball valves bring a different set of properties to the table. Where butterfly valves win on large-bore economics and compactness, ball valves win on shutoff quality, full-bore flow with zero restriction, and precise metering capability at smaller diameters.
Full-bore (full-port) ball valves are the specification standard for direct liquid cooling circuits. When coolant is running directly to server chassis or rear-door heat exchangers, you cannot afford the flow restriction or turbulence of a reduced-bore valve. Full-bore ball valves maintain the same internal diameter as the connecting pipe, contributing essentially zero pressure drop in the fully open position. For cooling distribution manifolds feeding multiple server racks, this translates directly to pump energy savings — smaller pump motors, lower kW consumption, better PUE (Power Usage Effectiveness) metrics.
Reduced-bore ball valves have their place in general service connections, drain points, vents, instrumentation isolation, and lower-priority branch circuits where pressure drop at that specific point is not a system-level concern. They’re lighter and less expensive than full-bore equivalents and perfectly adequate for the duty.
Three-way ball valves are increasingly specified in liquid cooling distribution circuits to manage bypass flow and temperature mixing. A three-way ball valve with a T-port or L-port configuration enables diversion of warm return flow around a heat exchanger without needing two separate two-way valves and a tee piece — saving installation space in densely packed plant rooms. The actuation requirement is straightforward: a quarter-turn electric actuator with a 3-position controller or a modulating positioner.
Sizing ball valves correctly for cooling circuits comes down to the Cv table and the acceptable pressure drop budget. A ball valve at full open has a very high Cv relative to its bore — the Cv of a full-bore ball valve is approximately 29 × d² (where d is bore diameter in inches), compared to approximately 17–22 × d² for a reduced-bore equivalent. For isolation-only duties, simply match the nominal bore to the pipe bore and move on. For throttling duties, however, ball valves are not ideal — they have a highly non-linear characteristic curve with very sensitive control at small openings and very little differential control at large openings. Use them for throttling only where a globe or butterfly valve is genuinely impractical.
Seat materials in ball valves for cooling applications deserve attention. PTFE seats are the standard choice: chemically inert, low friction, and compatible with virtually all glycol formulations and inhibitor packages. RPTFE (reinforced PTFE with glass or carbon filler) offers improved dimensional stability at elevated temperatures and higher cycle life — a sensible upgrade on modulating or frequently operated valves. For high-cycle applications (automated circuit switching, pump balancing), consider valves rated to at least 100,000 cycles. In server-side liquid cooling applications where valves may actuate tens of thousands of times per year, cycle life specification is not academic — it directly predicts maintenance intervals.
The shift from constant-flow to variable-flow chilled water distribution in data centre design has been one of the most significant changes in cooling system engineering over the past decade, and it has material implications for valve specification.
In a variable primary flow (VPF) system, the primary chilled water pumps modulate speed via variable speed drives (VSDs), maintaining a fixed differential pressure setpoint measured across the most remote AHU or CDU. Flow control valves on each cooling unit — typically two-port modulating butterfly or globe valves — throttle in response to cooling demand, and the differential pressure controller trims pump speed to maintain the setpoint. This is a beautiful, energy-efficient system when it works. When the control valves are improperly sized or the ΔP setpoint is misconfigured, you get hunting, poor temperature control, and substantially elevated pump energy consumption.
The key valve specification parameter for VPF systems is the authority of the control valve — defined as the ratio of the pressure drop across the valve at full open to the total pressure drop in the circuit including the valve. High authority (0.5 or above, ideally) means the valve exerts real control influence; low authority (below 0.3) means you’re chasing shadows. Achieving adequate authority in large-bore cooling circuits often forces engineers toward pressure-independent control valves (PICVs), which combine a differential pressure controller with a two-port control valve in a single body. PICVs maintain constant flow regardless of system pressure variation and deliver inherently high valve authority — they’ve become the default for AHU coil connections in data centre cooling systems above about 20 mm bore.
The actuator on a modulating butterfly or ball valve is not an afterthought. It’s half the system. And it’s a half that gets underspecified repeatedly, with consequences that show up months later during commissioning or fault-finding.
For butterfly valves, the actuator torque must overcome three contributions: bearing friction torque, hydrodynamic torque (from flow acting on the disc), and seat sealing torque. The hydrodynamic torque component is often overlooked — it reaches a maximum around 60–80° open and can exceed static breakaway torque under some flow conditions. Sizing actuator torque at 1.3× the valve manufacturer’s published torque at maximum differential pressure is a minimum safety factor; for critical service positions, 1.5× is more appropriate.
Electric actuators for modulating duty should specify modulating duty cycle — not just on/off rating. An actuator rated for on/off duty at 500 Nm will overheat if run continuously for modulating control. Specify actuators with a duty cycle appropriate to the application: 25% duty for slow-cycling temperature control, up to continuous duty for fast-responding pressure control loops. Integral positioners with 4–20 mA input and 4–20 mA position feedback are standard in BMS-controlled cooling systems. Ensure the actuator and positioner communicate correctly with the BMS protocol in use — BACnet and Modbus are the dominant protocols in data centre BMS architecture; proprietary protocols create integration headaches.
Fail-safe positioning is a critical consideration. For cooling systems, the standard fail-safe philosophy is fail open on cooling supply valves (to maintain cooling if control power is lost) and fail closed on isolation valves (to prevent uncontrolled flooding). Spring-return electric actuators and fail-safe pneumatic actuators both achieve this; the choice depends on available utilities (compressed air availability, power supply redundancy).
The shift toward direct liquid cooling (DLC) and two-phase immersion cooling creates genuinely new challenges for valve specification that go beyond adapting existing HVAC valve practice.
Single-phase liquid cooling using dielectric fluids or water/glycol circulated to cold plates mounted on processors creates high flow velocity in very small-bore pipework — often 10–25 mm in the secondary loop. Ball valves in these applications must handle turbulent, high-velocity flow with minimal pressure loss, and must be built to handle the thermal cycling inherent in a system that powers up and down with computing load. Stainless steel body construction is increasingly specified to avoid long-term corrosion issues with deionised water circuits.
Two-phase immersion cooling uses dielectric fluids that boil at relatively low temperatures (40–60°C) to absorb heat from submerged server components. Valves in these systems must handle vapour-liquid mixtures with potential for flash vapour formation — sizing must account for two-phase flow conditions that can be dramatically different from single-phase Cv calculations. Standard single-phase Cv sizing gives non-conservative answers; two-phase valve sizing requires application of correction factors derived from the Fanning friction factor and quality (vapour fraction) at the valve inlet conditions.
Leak detection and self-sealing fittings are becoming a standard spec requirement on server-side liquid cooling circuits, and this is influencing valve design. Dry-break or dry-disconnect couplings that self-seal when disconnected are being integrated into cooling distribution manifolds to allow server swaps without coolant spills. These impose specification requirements on upstream isolation valves — specifically, zero dead-legs and full-bore flow paths to allow proper drainage before disconnection.
Several technical standards are directly relevant to valve specification for data centre cooling applications, and procurement teams and specifying engineers should be familiar with them.
EN 12266-1 (Testing of valves — pressure tests) defines the acceptance criteria for seat leakage in water service valves. Class A (zero leakage) to Class D ratings are relevant depending on the service position. Isolation valves on primary chilled water headers are typically specified to Class D; control valves to Class B or C.
BS EN ISO 5211 governs mounting flanges and drive connections between valves and actuators. Ensuring actuator and valve use the same ISO 5211 flange size and drive square/coupling avoids the costly and time-wasting exercise of specifying bespoke adaptor plates.
ASHRAE 90.4 (Energy Standard for Data Centers) defines energy performance requirements for data centre mechanical and electrical systems. Valve selection — specifically the pressure drop contribution of control valves and PICVs — has a direct bearing on pump energy consumption and ASHRAE 90.4 compliance calculations.
Uptime Institute Tier Standards (Tier I–IV) drive redundancy and maintainability requirements that directly affect valve specification: number of isolation points, valve accessibility, provision of bypass loops, and whether on-line maintenance is required.
What This Means for Valve Procurement Strategy
If you’re specifying or procuring valves for data centre cooling projects, the practical implications of everything above can be distilled into a few clear directives.
Butterfly valves on large-bore HVAC circuits should be high-performance (double-offset) as standard, not concentric, even at additional cost — the modulating performance and longer seat life justifies the premium over a typical data centre asset life.
Ball valves on DLC circuits should be specified full-bore, stainless body (or at minimum, nickel-plated ductile iron), rated for the appropriate cycle count, with RPTFE seats for high-cycle modulating applications.
Control valves — whether butterfly, ball, or globe — should have valve authority formally calculated at design stage. PICVs should be the default on variable flow AHU/CDU connections unless there is a specific engineering reason to use a standard two-port valve with external DP control.
Actuators should be specified with appropriate duty cycle, fail-safe position, and BMS protocol compatibility confirmed before procurement, not during installation.
Finally: pressure test and leak test every valve before installation. The cost of a post-installation coolant leak in a live data centre — system downtime, hardware damage, remediation — dwarfs the cost of a pre-installation test rig. Data centre operators do not tolerate unplanned downtime, and neither should you.
