Skip to content
Koidra
Book a demo

Cooling and power control

Data Centers

Cooling that adjusts itself every minute.

Less power spent on cooling, and more of your grid connection left for IT.

Book a demo

What changes in your facility

57%
of their grid connection that smaller colocation sites use, averaged over a year International Energy Agency, 2026, citing EPRI
1.54
average power usage effectiveness in 2025, the sixth year it has barely moved Uptime Institute, 2025
1 in 5
serious outages cost more than $1 million Uptime Institute, 2025
11×
rise in the power density of AI servers, 2020 to 2025 International Energy Agency, 2026
More IT load on the grid connection you have
Power, not floor space, is what limits growth now. Every kilowatt the cooling plant no longer needs is one your IT can use.
Less power spent on cooling
Setpoints tuned once at commissioning cool for the worst day, every day. Moving them with the load and the weather keeps servers inside their limits for less.
Fewer outages that start in the building
Power problems remain the top cause of serious outages. A chiller that starts short-cycling, or a setpoint left behind after a service visit, shows up the same day instead of at the next audit.
Control that keeps up with AI loads
AI racks draw far more power, and their heat changes faster. Cooling has to answer in minutes, not at the next shift change.

Where the cooling energy goes

The IT load is given. Everything spent moving its heat out of the building is overhead, and most of that overhead sits in setpoints that were tuned once, at commissioning, with a margin to be safe.

Heat leaves a data hall through a chain: server fans into the cold aisle, CRAH or CRAC units, the chilled water loop, the chillers, and the cooling towers or dry coolers outside. Each stage has its own setpoint: supply air temperature and fan speed, chilled water supply temperature, condenser water approach, and the point at which the economizer takes over. Each was set with a margin, and the margins add up to air delivered colder than any server needs, water colder than the coils need, and free-cooling hours left unused on days the weather would have given them away.

The load underneath those setpoints moves every hour and the weather moves every day. Holding the setpoints still is what turns a safety margin into a permanent energy cost. What actually constrains you is the air the servers will accept, temperature and humidity together, and that is a range rather than a single number.

Where the cooling energy goes, and where the levers areThe IT load is given. Its heat leaves through the CRAH units, the chilled water plant and the cooling towers or dry coolers, each stage with a setpoint tuned at commissioning. KoPilot moves those setpoints as the load and the weather change, keeping the air arriving at the servers within the temperature and humidity your hardware warranty allows.IT loadgiven, and movingevery hourCRAH and CRAC unitssupply air temperature,fan speedChilled water plantsupply temperature, chiller stagingTowers and dry coolersapproach temperature,economizer hoursAir the servers acceptthe temperature and humidityyour warranty allowsKoPilotmoves every setpoint as theload and the weather changeWeathernow and ahead
Every stage of the chain has a setpoint, and each one was set for the worst day. KoPilot sets them for today, inside the range the servers accept.

Where the margin sits

The saving is the distance between the warmest air the servers accept and where the setpoints sit today, and it is paid every hour the two are apart.

Where cooling energy is decided A temperature scale running from cool on the left to the equipment limit on the right. A recommended band sits inside a wider allowable band, which ends at the hard equipment limit. A setpoint set by hand sits low in the recommended band. The distance from there to the top of the same band is the margin continuous control can recover without touching the limits. cooler, more energy warmer, less energy equipment limit recommended allowable set by hand and rarely revisited held continuously inside the same limits the margin worth recovering
The equipment limit is fixed. The range you choose to run inside it is yours to set. The setpoint inside that range is where the energy is decided, and on most sites it was chosen once at commissioning. Closing part of that distance, continuously, without moving what you set, is the whole job of the control layer.

How KoPilot runs the cooling plant

KoPilot is built around the facility's own equipment, operating limits and control architecture, and it leaves the building management system in charge of the equipment.

Physics-informed models
Of cooling and airflow, built for this facility rather than a facility of this type.
Continuous optimization
Of operating setpoints, as IT load and outdoor conditions change.
Equipment and thermal constraints
The limits the optimization is not allowed to cross.
Supervisory controls
That define what the system can and cannot change.
Monitoring and fallback
So the facility keeps running safely if anything upstream stops.
What a data center deployment is built fromPhysics-informed models of cooling and airflow and the equipment limits feed continuous setpoint optimization. Supervisory controls decide what may change, and the facility's existing control system still runs the equipment, with monitoring and fallback beside it.Models of cooling and airflowphysics-informed, for this facilityContinuous setpoint optimizationEquipment and thermal limitsSupervisory controlswhat the system may and may not changeYour existing control systemstill runs the equipmentMonitoring and fallbackfor safe operation
The BMS still runs the plant. KoPilot decides what it should be holding, from what the hall is doing now.

Start with the operational data

Closed-loop optimization does not need to be the first step.

DataPilot can bring thermal, power, equipment, operator and maintenance data into unified operational data. That alone makes it easier to understand where energy is being used, identify changes in equipment performance, and investigate shifts in PUE.

Maintenance Management can connect those signals directly to maintenance work. Predix can then surface changes, recurring issues and opportunities that would otherwise require someone to find them manually.

Once the data and operating behavior are understood, KoPilot closes the loop through a supervised deployment: one hall or one zone first, with people watching, then the rest.

The order a data center takes the products inDataPilot first, bringing thermal, power, equipment, operator and maintenance data into unified operational data, with Maintenance Management beside it. Predix on both. KoPilot last, through a supervised deployment.DataPilotthermal, power, equipment,operator and maintenance dataPredixchanges, recurring issues,opportunitiesMaintenance Managementthe signals tied to the workKoPilotclosed loop, after asupervised deployment
Each product pays for itself before the next exists, and the data the first ones build is what closed-loop control is tested against.

Is your facility a good fit?

Five questions decide where you start.

Do you have power and thermal data at sufficient resolution?
Minute-level readings from the cooling plant and the IT load, not a monthly PUE.
Can cooling equipment setpoints be adjusted through the existing control system?
Chilled water and airflow setpoints have to be writable for the loop to close.
Are airflow and containment issues understood well enough to separate equipment problems from control problems?
A hot spot caused by a missing blanking plate is not a setpoint problem.
Are operating constraints and safe limits clearly defined?
Server inlet limits, equipment limits and the margin the site wants to keep.
Is there an environment where new control strategies can be tested under supervision?
A hall or a zone where a strategy can run with people watching before it runs alone.
Where a data center startsFive questions decide the starting point. With data at resolution, adjustable setpoints, understood airflow, defined limits and a place to test, KoPilot can be evaluated. Without them, DataPilot, Maintenance Management and Predix build the foundation first.yes to all fivenot yetFive questionsdata at resolution?setpoints adjustable?airflow understood?limits defined?a place to test?Evaluate KoPilotclosed-loop cooling optimizationBuild the foundationDataPilot, Maintenance Management,Predix
With the foundation in place, KoPilot goes straight to closed-loop cooling optimization. Without it, the other three products build the foundation first.

What each product gives a facility like yours

Start with Maintenance Management or DataPilot. Predix needs both, and KoPilot adds control on top. Each product pays for itself on its own.

  1. Start here

    DataPilot

    Inlet air temperature per rack, chilled water setpoints and CRAH performance in one place, so PUE stops being a number that arrives monthly and becomes something you can watch drift.

    How a reading becomes an action A sensor reading and an operator's hand-entered reading both enter unified operational data. From there an alarm reaches a phone, a work order opens against the right asset, reports are written, and KoPilot adjusts setpoints. Process sensor every few seconds Operator's reading typed on a phone Unified operational data same equipment and batch, same clock Alarm on a phone Work order opens Report is written KoPilot adjusts setpoints both are the same shape to the platform
    Your existing data logger never sees the operator's reading, and a maintenance system never sees the trend. Holding both makes everything to the right of the record possible.

  2. Start here

    Maintenance Management

    A chiller that begins short-cycling raises the job itself, carrying its own service history, and setpoint drift after a maintenance visit shows up the same week rather than next quarter.

    How a signal becomes a cost line A condition being met on live signals, or a preventive schedule falling due, opens a work order against the right asset with its checklist and an assignee. Completing it consumes parts from stock, logs labour hours and attaches photo evidence, and the record of work done properly is what raises mean time between failures, lowers mean time to repair, and gives a defensible cost per asset. A condition is met not just a limit on one tag Schedule falls due preventive, on a cycle Work order opens right asset, right checklist assigned to who is on shift Parts consumed stock level moves Labour hours logged against the job Photo evidence attached on completion MTBF up, MTTR down and the cost per asset behind it no one has to notice first
    The value is not the ticket, it is the work being done right: raised on real evidence, tracked, and closed against the machine that needed it, so time between failures rises and time to repair falls. A condition is more than a limit on one tag, and the examples on this page show what that buys.

  3. AI insights

    Predix

    A weekly account of where the controllable part of PUE went, including free cooling hours you did not take and the approach temperature you paid to hold, and what to change to take them.

    What Predix produces, and what you put in From unified operational data, Predix produces daily and weekly reports that say what to adjust or keep, and maintenance work orders raised when an asset is due. Your team acts on them and sets the targets, which shape what is measured. Unified operational data Daily and weekly reports what to adjust, or keep Maintenance work orders raised when an asset is due Your team acts, and sets the targets arrives on its own your targets
    Predix works on the unified operational data from DataPilot and Maintenance Management, measured against the targets you set.

  4. Autonomous control

    KoPilot

    Chilled water and airflow setpoints moved against real inlet temperatures rather than a commissioning margin nobody has revisited, inside limits written to protect the hardware.

    The closed loop, and the limits it obeys The process is read by a physics-aware model, whose proposed setpoints pass through an operating box that states the limits it may not cross, then reach the controllers, which write them back to the process. The loop runs every minute. The process dryer, boiler, hall, crop Physics-aware model knows what the equipment can and cannot do Operating limits set by your engineers Controllers setpoints written back every minute, not every shift
    Every setpoint passes through limits your engineers set for your site. The model cannot learn or change them, and your safety systems and interlocks stay in charge.

Why the control transfers

37%
less air-handler fan energy from automatic fan-speed control, in a working colocation data center Lawrence Berkeley National Laboratory, 2012
2×
winner of the Autonomous Greenhouse Challenge Greenhouse control, a different domain from this page

KoPilot’s control was proven in greenhouses, where the same physics of heat, air and water sets the limits, against expert growers on a live crop.

Every deployment starts read-only, on the data your facility already has. One hall or one zone goes first, with people watching, inside limits your team sets, and the building management system stays in charge of the equipment.

What we measure here

See Koidra on your facility data

Bring a week of readings from one hall. We will show you what the platform sees, what it would have flagged, and what it would have scheduled.

Book a demo