Data center chillers are the refrigeration machines that pull heat out of a facility’s closed chilled water loop, after CRAH coils, fan walls and coolant distribution units have collected it from the IT equipment, and push that heat outdoors through cooling towers or air-cooled condenser coils, every hour of the year. Paul Industries works on them from selection through decades of service: selection support, rigging and setting, chilled and condenser water piping, flushing, startup, maintenance, repair, retrofit and capacity expansion, nationwide. This page goes into the machines themselves and the plant around them; the wider picture of air, liquid and heat rejection is on our data center cooling systems page.

What is a data center chiller?

A data center chiller is a vapor-compression refrigeration machine that cools the water serving the white space, built and controlled for continuous duty, tight supply temperature control and fast recovery after a power event. Inside are four parts: an evaporator where refrigerant boils and absorbs heat from the chilled water, a compressor that raises the refrigerant pressure, a condenser where that heat is given up to condenser water or outdoor air, and an expansion device that drops the pressure again. Everything else in the data center chiller system, the pumps, headers, valves, towers or coils and controls, exists to keep those four parts working at the right temperatures.

The machines sit either indoors in a plant room, beside pumps and plate heat exchangers, or outdoors as packaged air-cooled units lined up in a data center chiller yard. They serve three kinds of load: chilled water coils in CRAH units and fan walls, in-row and rear-door coils, and the primary side of liquid-to-liquid coolant distribution units, where the plant’s water takes heat from the server loop. That last load is growing quickly, and how it changes plant temperatures is covered on our data center liquid cooling page.

Sizing starts from heat. One ton of refrigeration is 12,000 Btu per hour, or about 3.52 kilowatts, so a hall carrying 10 megawatts of IT and supporting heat needs roughly 2,840 tons of net cooling before any spare machine is added. Water flow follows from the temperature difference: flow in gallons per minute equals 24 times tons divided by the delta T in degrees Fahrenheit, so a 1,000-ton machine running a 12 degree difference moves about 2,000 gallons per minute through its evaporator. Those two numbers drive pipe sizes, pump selections and how many machines the plant needs.

How does a data center chiller work?

The compressor does the work of moving heat uphill, from water at perhaps 65 degrees Fahrenheit to a tower or the outdoor air at a much higher temperature. The gap between the refrigerant’s evaporating and condensing temperatures is called lift, and compressor power rises with lift. That single idea explains most data center chiller design choices: warmer chilled water raises the evaporating temperature, colder condenser water or cooler outdoor air lowers the condensing temperature, and either one shrinks lift and cuts energy. A common rule of thumb is that each degree Fahrenheit of higher chilled water temperature improves efficiency by roughly 1 to 2 percent on a constant-speed machine, and more on a variable-speed one.

Types of data center chillers

The compressor defines the machine. All three common types appear in data centers; the right one depends on capacity, part-load profile, heat rejection method and the space available.

Compressor typeWhere it fitsStrengthsPoints to plan for
Centrifugal, oil-lubricatedLarge water-cooled plantsLowest kW per ton at full load in big sizes; long service lifeSurge at low load with high lift; oil system upkeep; heavy rigging
Centrifugal, oil-free magnetic bearingWater-cooled and some air-cooled plants with long hours at part loadExcellent part-load efficiency, no oil management, low vibrationDrive and bearing controls need trained service; power quality matters
ScrewMid-size water-cooled and most large air-cooled machinesHandles high lift well; wide capacity range; robustOil separation and filtration; part-load efficiency depends on capacity control
ScrollSmaller modular air-cooled units and edge sitesSimple, multiple compressors give step redundancy inside one unitLower efficiency than large machines; many small circuits to service

An oil free chiller removes a whole class of maintenance and a common efficiency loss: in an oil-lubricated machine, some oil always travels with the refrigerant and coats the heat exchanger tubes, which hurts heat transfer as the years go by. The oil free magnetic bearing chiller levitates the compressor shaft in a magnetic field, so there is no metal contact and no oil pump, filter or heater, and it pairs that with an integrated variable-speed drive. An oil free centrifugal chiller of this kind is at its best exactly where many data center plants live, running for long periods at partial load with relatively low lift. Large oil-lubricated centrifugal machines remain common for the biggest water-cooled plants, and many owners mix machine types, with a variable-speed unit as the swing machine that trims the load.

Water-cooled vs air-cooled chillers for data centers

The biggest selection decision is how the condenser rejects heat: to water that a cooling tower cools by evaporation, or directly to outdoor air through finned coils and fans. Choosing a chiller for data center use is mostly a trade between energy, water and site layout.

FactorWater-cooledAir-cooled
Typical full-load efficiencyAbout 0.5 to 0.6 kW per ton for large centrifugal machinesAbout 1.0 to 1.2 kW per ton for screw machines, scroll often higher
Heat sinkWet-bulb temperature, through the towerDry-bulb temperature, through the condenser coil
Site waterContinuous tower makeup for evaporation and blowdownNone for heat rejection; adiabatic options use some in peak hours
LocationIndoor plant room plus towers on the roof or in a yardOutdoor chiller yard or roof, no plant room for the chillers
Footprint and structureCompact machines, but towers, condenser pumps and piping add spaceLarge outdoor area with airflow clearances between units
MaintenanceCondenser tube cleaning and tower water treatmentCoil cleaning, fan and motor upkeep
Hot-weather behaviorHolds capacity well on hot, dry daysCapacity and efficiency fall as ambient climbs; select for site design dry-bulb

A water cooled chiller for data center campuses is usually a centrifugal machine in a plant room with towers outside, and it wins on energy wherever water is available and affordable. An air cooled chiller for data center sites wins where water is scarce or expensive, where the site is small or phased in modules, or where the owner wants to avoid tower water treatment altogether; it also pairs naturally with integrated free cooling. The water side of the trade is set out in our data center water usage guide, and the dry heat rejection option without compressors in our dry cooler guide. The generic comparison for process plants lives on our process chillers page.

Efficiency is also quoted as a coefficient of performance, the ratio of cooling delivered to power drawn. A data center chiller COP of about 6.4 corresponds to 0.55 kW per ton, and an air-cooled machine at 1.1 kW per ton works out to a COP near 3.2. Real performance depends on the temperatures the plant actually runs.

Chiller ratings: kW per ton, IPLV and minimum efficiency

Chillers are tested and rated under AHRI Standard 550/590, which fixes full-load conditions and a part-load metric called IPLV, the integrated part load value. IPLV blends efficiency at four load points with fixed weights: 1 percent at full load, 42 percent at 75 percent load, 45 percent at half load and 12 percent at 25 percent load, each at progressively cooler condenser conditions. AHRI itself states that IPLV represents a single chiller and is not intended to predict the annual energy of a particular installation. That caveat matters in a data center, where machines run around the clock at temperatures far from the rating point and loads are shared across several units.

Minimum efficiencies come from ASHRAE Standard 90.1, which offers Path A for machines optimized at full load and Path B for machines optimized at part load, and data center facilities must also meet ASHRAE Standard 90.4, the energy standard written for them. Federal buyers follow the Federal Energy Management Program purchasing levels, which for a centrifugal machine of 600 tons or more call for about 0.50 kW per ton at both full load and IPLV. A good selection also asks for performance at the site’s own temperatures and several loads.

Free cooling chillers and warmer chilled water

A free cooling chiller lets outdoor air or tower water do some or all of the cooling with the compressors throttled back or off, and warmer chilled water is what makes that pay. The ASHRAE TC 9.9 recommended range for server inlet air runs up to 27 degrees Celsius, and coils selected for warm water let the plant supply 18 to 20 degrees Celsius or more instead of the 6 to 7 degrees of an office plant. Liquid-cooled halls push further: the ASHRAE facility water classes W17, W27, W32, W40, W45 and W+ name the upper supply temperature in degrees Celsius, and the warmer the class the equipment accepts, the more hours of the year the plant can run with no compressor at all.

On air-cooled machines the option is built in. A chiller with integrated free cooling has extra finned coils ahead of the condenser that the return water, usually glycol, passes through before reaching the evaporator. When the outdoor air is a few degrees colder than the return water, the coils start pre-cooling it and the compressors unload; as the air gets colder the coils take more of the load, and well below the supply setpoint the compressors stop. One manufacturer’s published data, for example, starts free cooling at about 2 degrees Celsius below return temperature and reaches full free cooling at about 13 degrees below. Where the building loop must stay glycol-free, a plate heat exchanger separates a glycol free cooling circuit from the plain-water plant.

On water-cooled plants the same job is done by a plate-and-frame exchanger between condenser water and chilled water, piped either in series ahead of the chillers, so it can pre-cool return water on mild days, or in parallel for full free cooling in cold weather. Series piping captures far more partial hours, which is why most new data center plants use it. The control transitions between mechanical, mixed and free cooling modes are where these systems succeed or fail, so they get tested deliberately before handover.

Data center chiller plant design: redundancy and concurrent maintenance

A data center chiller plant is designed so that losing any one machine, pump or valve never costs the IT load its cooling, and so that every component can be serviced without a shutdown. Chiller plant redundancy is expressed against N, the number of machines needed to carry the design load on the hottest design day.

SchemeWhat it meansPiping and valving consequence
NExactly enough capacity; no spareAny chiller outage or maintenance reduces cooling
N+1One spare chiller, pump and tower cell beyond the design needCommon headers with isolation valves on every machine
N+2Two spares, so one can be under maintenance while another failsSame as N+1 with more valving and larger headers
2NTwo complete, independent plants, each able to carry the full loadTwo sets of headers and distribution paths, often looped to the hall
2(N+1)Two independent plants, each with its own spareHighest cost; used where fault tolerance is the stated requirement

N+1 chiller redundancy is the most common arrangement, and it is only real if the spare can actually take over: same supply temperature capability, valves that open on command, and controls that start it automatically. The Uptime Institute’s Tier III calls for concurrent maintainability, so that any component or path can come out of service for planned work; Tier IV adds fault tolerance, so an unplanned failure of any single component or path does not affect the load. Mechanical contractors turn those words into ring headers or dual mains, isolation valves on both sides of every chiller and pump, bypasses around exchangers, and enough valves on the distribution that a header section can be isolated while the rest stays live. Our data center mechanical contractors page covers how that work is organized on a project.

Data center chiller capacity is set at the site’s design wet-bulb or dry-bulb temperature, with allowances for tube fouling, glycol, altitude on air-cooled units and the efficiency loss as a machine ages. Phased campuses add another layer of data center chiller plant design: headers, pads and tower cells sized for the final build, with chillers added as halls fill.

Pumping arrangements

Two pumping layouts dominate. Primary-secondary systems give each chiller a constant-flow primary pump and use separate variable-speed secondary pumps for the hall, joined by a decoupler. Variable primary systems vary flow through the chillers themselves, which saves pump energy and space, but need a bypass valve controlled from a flow meter or differential pressure sensor to keep each evaporator above its minimum flow, and controls that limit how fast flow changes, since a sudden drop can freeze an evaporator. Packaged machines often have a minimum around 60 percent of design flow, and custom-configured ones can go lower. The general principles are on our chilled water systems page; in a data center, the extra question is how each scheme behaves when a chiller trips and the others must pick up its flow within seconds.

Chilled water thermal storage tank for ride-through

A chilled water thermal storage tank keeps cold water flowing to the hall while chillers restart after a power interruption, bridging the minutes between the utility dropping out, the generators taking load and the compressors coming back to full capacity. A conventional chiller that loses power for more than a moment has traditionally needed 10 to 15 minutes to restart and recover, and quick-restart designs cut that to a few minutes. During that gap the heat keeps coming, so the plant draws on stored cold water, with secondary pumps and CRAH fans kept on uninterruptible power so the stored water actually reaches the coils.

Tanks are sized for the minutes of full load the owner wants to cover, plus margin, and are usually stratified: warm return water floats on top of cold supply water, separated by a thermocline that diffusers protect by keeping inlet velocities low. Smaller sites use pressurized buffer tanks in the loop, which also add the system volume that stops chillers short-cycling at low load. Paul Industries pipes tanks into the plant, including diffuser connections, isolation and bypass valves, and the charging and discharging controls interface on the piping side.

Chiller plant controls and sequencing

Data center chiller controls decide which machines run, at what setpoint, and what happens when something fails. A written chiller plant sequence of operation typically covers staging up and down on load, flow and supply temperature; lead and lag rotation to even out run hours; failure response, so a tripped machine is replaced automatically and its isolation valves close; transitions into and out of free cooling; condenser water and tower fan control; and restart after a power event. ASHRAE Guideline 36 added standardized chilled water plant sequences in its 2021 edition, and many engineers now start from them. Good sequences also protect equipment: minimum run and off times, surge avoidance on centrifugal machines at low load and high lift, and limits on how fast flow and setpoints may change.

Every one of those behaviors is tested during functional testing by actually failing machines, pumps and valves under load. Our data center commissioning page describes how those tests are run and documented, following the ASHRAE Guideline 0 commissioning process.

Rigging, setting and piping a data center chiller

Installation quality decides whether a chiller delivers its rated performance for its whole life. A large water-cooled machine can weigh tens of thousands of pounds, arrives with water boxes and sometimes ships in sections, and has to be moved into a plant room with its tube-pull clearance intact. Rooftop and yard units arrive by crane. Our millwright and rigging services crews plan the lift, set the machine on its pad or isolators, level it and confirm clearances before any pipe is connected.

StepWhat gets doneWhy it matters
Pad and settingHousekeeping pad checked level, isolators or pads installed, machine set and leveledAn out-of-level shell affects refrigerant and oil distribution and vibration
Service clearancesTube-pull space at one end, water box swing room, electrical panel access kept clearWithout them, tube cleaning and replacement become a rebuild
Evaporator and condenser pipingFlexible connectors, isolation valves, strainers, thermometers, gauges, flow meters, vents and drainsLets each machine be isolated, measured and serviced without draining the plant
Pipe supportPiping carried on its own supports, never hung on the water box nozzlesNozzle loads distort tube sheets and cause leaks
Machinery room itemsRelief valve vent piping to outdoors, refrigerant detector placement, ventilation interface per ASHRAE Standard 15Life safety and code compliance in the plant room
FlushingLoops flushed with the chiller bypassed until strainers stay cleanConstruction debris plugs evaporator and condenser tubes
Pressure testHydrostatic test of new piping to the governing code, joints inspectedFinds leaks before insulation and before load
StartupFlow proven, factory startup performed, readings logged against the selectionSets the baseline every later service visit is compared with

Piping is usually welded or grooved carbon steel for chilled and condenser water, sized from the flow arithmetic above and insulated with a vapor retarder wherever it runs below the dew point. Our process piping crews fabricate spools in the shop so the field work is mostly fit-up and closing welds, and pressure testing follows ASME B31.9 or the code the engineer specifies, as described on our hydrostatic testing page.

Data center chiller maintenance schedule

Data center chiller maintenance is planned around redundancy: one machine is taken out at a time while the spare carries its share, so no task ever reduces cooling below N. Intervals in this table are common practice for large machines, adjusted to each manufacturer’s instructions and to how dirty the condenser water runs.

IntervalTasks
Daily or continuousTrend supply and return temperatures, evaporator and condenser approach, refrigerant pressures, amps or kW, run hours and alarms
MonthlyReview approach trends, check oil level and pressure on lubricated machines, inspect for leaks, clean strainers if pressure drop has risen
QuarterlyRefrigerant leak check, vibration readings, condenser water chemistry review with the tower program, controls and sensor checks
AnnuallyCondenser tube cleaning, oil analysis for acid, moisture and wear metals, oil and filter change where needed, relief valve and safety control checks, coil cleaning on air-cooled units
Every few yearsEddy current testing of evaporator and condenser tubes, evaporator tube cleaning, compressor inspection per the manufacturer

Approach temperature, the difference between leaving water and saturated refrigerant temperature, is the most useful single trend: a condenser approach that creeps up means fouled tubes or poor tower water, and an evaporator approach that rises can mean low charge or oil in the bundle. Condenser water quality is controlled at the tower, covered on our cooling tower water treatment page. Paul Industries runs these tasks under preventive maintenance agreements built around the site’s redundancy and change windows.

Common data center chiller faults

Most trips and capacity losses trace to a short list of causes, and the plant data usually points to which one.

SymptomProbable causeCorrective work
High head pressure, rising condenser approachFouled or scaled condenser tubes, poor tower performance, low condenser flowTube cleaning, tower and water treatment review, flow verification
Low evaporator pressure, freeze-protection tripLow chilled water flow, low refrigerant charge, sensor driftStrainer and valve checks, leak repair and recharge, sensor calibration
Surge on a centrifugal machineLow load combined with high lift, often warm condenser waterCondenser water reset, staging changes, hot gas bypass or drive tuning
Frequent starts and stopsSmall loop volume, oversized machine for early-phase load, poor stagingBuffer tank, sequence changes, smaller swing machine
Low-flow trips during stagingIsolation valves opening too fast, minimum flow bypass mis-setValve stroke times, bypass control tuning
Oil loss or high oil temperatureOil migration at low load, failing heater or separatorOil recovery checks, component repair
Capacity lower than the selectionTube fouling, non-condensables, wrong setpoints, coil fouling on air-cooled unitsCleaning, purge service, setpoint and sequence review

When a machine goes down outside a planned window, our emergency repair service covers the piping, valves, pumps and heat exchangers around it so the spare can carry the load while the fault is cleared.

Retrofitting chillers for AI and liquid cooling loads

Many existing plants were sized for air-cooled halls at modest density and are now being asked to serve racks many times denser. AI data center chillers are being selected, or reset, for warmer supply water, because the liquid-cooled share of the load accepts it and the remaining air-cooled share can often be served by reselected coils. Retrofit options include raising chilled water setpoints in stages with coil checks, adding a heat exchanger and dedicated warm-water loop for the CDUs, replacing constant-speed machines with variable-speed or oil-free units, and adding chillers to existing headers. Each one changes flow, temperature and redundancy, so the plant is re-evaluated as a whole rather than one machine at a time.

Capacity expansion in a live facility is planned around the existing redundancy: new machines are set and piped to capped header connections, tie-ins are made during an approved window or by hot tap, and the new equipment is flushed and tested off-line before it sees plant water. Chillers for data centers are long-lived assets, and a well-planned retrofit often reuses pads, headers and towers while replacing the machines and controls.

How much does a data center chiller cost?

The installed cost of a chiller is the machine plus everything needed to set it, connect it and prove it works, and on data center projects the second part is often comparable to the first. Main cost drivers are compressor type, capacity and efficiency, refrigerant, integrated free cooling, crane access, pads and steel, piping size and length, controls integration, flushing and testing, and whether the work happens beside live load. Lead times are long, so procurement timing affects schedule as much as price, and lifetime running cost usually outweighs first cost.

How to choose a contractor for data center chillers

Ask any contractor bidding chiller installation or service for evidence on these points:

  • A rigging plan with crane or skate method, weights, routes and clearances for the specific machines.
  • Welding procedures and welder qualifications for the chilled and condenser water piping, and a weld map at turnover.
  • A written flushing procedure that keeps chillers and heat exchangers bypassed until the loop is clean.
  • Pressure test records to the governing code and a leak inspection of every mechanical joint.
  • A method of procedure for every tie-in to a live plant, with rollback steps.
  • Maintenance capability after handover: logs, tube cleaning, oil analysis and eddy current testing arranged on a redundancy-aware schedule.

The data center hub page, data center water and piping, shows how this chiller work fits with water treatment and the rest of the mechanical scope.

Paul Industries on data center chiller projects

Paul Industries supports selection by reviewing loads, temperatures, redundancy and space with the owner’s engineer; rigs, sets and levels chillers in plant rooms, yards and on roofs; fabricates and installs chilled water, condenser water and glycol piping, plate heat exchanger free cooling packages and thermal storage connections; flushes, pressure tests and starts up the plant alongside the manufacturer’s technicians; and maintains, repairs, retrofits and expands chiller plants for the life of the facility. Crews travel nationwide for new builds and for work beside live load. Request a quote for data center chiller installation, service or a capacity expansion.

Service needed *

Standards referenced: AHRI 550/590 · ASHRAE Standard 90.1 · ASHRAE Standard 90.4 · ASHRAE TC 9.9 Thermal Guidelines for Data Processing Environments · ASHRAE Standard 15 · ASHRAE Guideline 36 · ASHRAE Guideline 0 · ASME B31.9 · Uptime Institute Tier Standard · EPA AIM Act Technology Transitions Rule

Frequently asked questions

Do data center chillers use water?

The chiller itself circulates the same closed-loop water continuously. Water is consumed by the heat rejection side: a water-cooled chiller depends on a cooling tower that evaporates water and needs blowdown, while an air-cooled chiller rejects heat to outdoor air and consumes none for heat rejection unless it has adiabatic pre-cooling for the hottest hours.

How many chillers does a data center have?

It depends on load and redundancy. A small site may have two or three packaged air-cooled units; a large campus building often has four to eight large water-cooled machines per plant, and multi-building campuses have several plants. The count is the number needed for the design load plus the spares the reliability target calls for.

Why are water cooled chillers more efficient than air-cooled machines?

Because the tower cools condenser water toward the outdoor wet-bulb temperature, which is lower than the dry-bulb temperature an air-cooled coil works against, and water transfers heat far better than air. Lower condensing temperature means less lift for the compressor, so water-cooled centrifugal machines typically draw around half the power per ton of air-cooled units.

How does a free cooling chiller work?

Return water passes through extra coils or a plate heat exchanger that uses cold outdoor air or tower water to cool it before it reaches the evaporator. In mild weather this pre-cooling lets the compressors unload; in cold weather the free cooling side carries the whole load and the compressors stop. Controls move between mechanical, mixed and full free cooling automatically.

What is chiller lift?

Lift is the difference between the refrigerant's condensing and evaporating temperatures. It sets how hard the compressor works. Warmer chilled water and colder condenser water both reduce lift, which is why data center plants that run warm supply water use much less compressor energy than comfort plants.

How long does a chiller take to restart after a power interruption?

Conventional machines have traditionally needed 10 to 15 minutes to restart and return to setpoint after losing power. Quick-restart controls and drives cut that to a few minutes. The plant has to bridge that gap with stored chilled water and pumps on uninterruptible power.

What is surge in a centrifugal chiller?

Surge is a flow reversal through the compressor that happens when it cannot produce enough pressure to overcome the lift, typically at low load with warm condenser water. It is noisy and damaging if repeated. Condenser water reset, variable-speed drives, correct staging and hot gas bypass on some machines prevent it.

Do oil-free chillers need less maintenance?

They remove all oil-related tasks: no oil changes, oil analysis, filters, heaters or oil recovery problems. Tube cleaning, refrigerant leak checks, eddy current testing and drive and bearing controller checks remain, and the electronics need technicians trained on that design.

Can an existing chiller run at a warmer chilled water setpoint?

Usually yes, and it typically gains efficiency, but the manufacturer should confirm the machine's limits, and the hall's coils must still meet the load at the warmer water. Setpoints are normally raised in steps while server inlet temperatures and coil valve positions are watched.

How does the HFC phasedown affect chiller selection for a data center?

The EPA's Technology Transitions rule under the AIM Act limits the global warming potential of refrigerants in newly manufactured chillers. Selections now center on lower-GWP refrigerants, which can change machine sizes, efficiencies and safety classifications, and ASHRAE Standard 15 requirements for the plant room follow the refrigerant chosen.

What is the minimum flow through a chiller evaporator?

Each chiller has a minimum evaporator flow set by the manufacturer to keep heat transfer stable and prevent freezing; packaged machines are often around 60 percent of design flow. Variable primary plants protect it with a bypass valve controlled from a flow meter or differential pressure sensor.

What is eddy current testing on a chiller?

A probe is pushed through each evaporator and condenser tube and measures changes in an induced electrical field to find wall thinning, pitting and cracks before a tube leaks. It is usually done every few years, and the results guide plugging or retubing decisions.

How long do data center chillers last?

A well-maintained centrifugal or screw machine commonly runs around two decades or longer. Life is shortened by poor condenser water treatment, repeated surge, refrigerant leaks and running without oil analysis or tube inspection. Many plants replace machines for efficiency or refrigerant reasons before they wear out.

Should an aging chiller be overhauled or replaced?

Compare the overhaul cost, the efficiency of the old machine at the site's real temperatures against a new one, refrigerant availability, tube condition from eddy current results, and how the plant's load and temperatures have changed. A machine that no longer matches the plant's supply temperature or capacity needs is usually replaced.

Can chillers be added to a running data center plant?

Yes. New machines are set and piped to valved or capped header connections, flushed and tested off-line, then tied in during an approved window or by hot tap while redundancy covers the load. Controls are updated so the new machine joins staging and failover.

Is IPLV a good measure for comparing data center chillers?

It is a useful standard comparison, but its weighting assumes a building load profile and cooler condenser conditions at part load. Data center machines often run long hours at partial load and warm supply water, so ask for performance at the site's own conditions as well.

Do air-cooled chillers lose capacity in hot weather?

Yes. As outdoor dry-bulb temperature rises, condensing pressure rises, so capacity falls and power per ton increases. Air-cooled machines are selected at the site's design dry-bulb with margin, kept clear of recirculating hot discharge air, and sometimes given adiabatic pre-cooling for peak days.

What is the difference between a chiller yard and a chiller plant room?

A yard holds packaged outdoor units, usually air-cooled, with piping running to the building; a plant room houses water-cooled machines indoors with pumps and exchangers, with towers outside. Yards need airflow spacing and weather protection; plant rooms need tube-pull space and refrigerant safety measures.

Related: Data Centers · Data Center Cooling Systems · Chilled Water Systems · Process Chiller Installation & Service · Industrial Process Chillers: How They Work & How to Choose · Data Center Liquid Cooling · Request a quote · Server Immersion Cooling: Tanks, Piping and Facility Water for Data Centers · Chiller Repair, Overhaul and Replacement for Process and HVAC Chillers