The motor-driven cooling behind AI uptime
Most AI infrastructure discussions move quickly to grid capacity, power purchase agreements, backup generation, and connection queues. Those issues matter, but they can push another part of the problem into the background: the motor-driven cooling plant inside the building.
AI data centers are built around compute, but they only stay online if heat can be removed continuously. Almost every kilowatt drawn by the IT load eventually has to leave as heat, through pumps, fans, compressors, chillers, towers, drives, starters, and controls.
That makes cooling more than a support service. If those systems start badly, trip at the wrong time, or recover unpredictably after a disturbance, the risk moves quickly from the plant room to the server hall.
The International Energy Agency (IEA) expects data center electricity consumption to rise from 485 TWh in 2025 to roughly 950 TWh by 2030. That growth is usually discussed in terms of utility supply and server efficiency, but it also raises a practical question for the mechanical plant: can the cooling system start, run, and recover reliably as the compute load grows?
Data center infrastructure is usually discussed in digital terms: chip architecture, rack density, storage, network latency, redundancy, cybersecurity, and workload growth. Inside the site boundary, the limits are also physical. Heat has to move. Air and fluids have to circulate. Compressors, pumps, fans, chillers, towers, drives, starters, and controls all have to keep doing their job.
Electrical energy consumed by the IT load ultimately leaves as heat. To keep equipment within its operating limits, that heat has to be captured, transferred, and rejected through the mechanical cooling plant.
The U.S. Department of Energy has indicated that cooling can account for up to 40% of total data center energy use. The exact percentage depends on climate, rack density, fluid temperature, cooling architecture, and operating strategy, but the underlying requirement does not change. The heat still has to leave the rack, the room, the loop, and the site. That puts cooling directly on the critical path of uptime.
Follow the heat out of the building, and the motor load becomes obvious. Large rotating equipment appears at every stage of the cooling loop:
Air movement: moving air through containment systems, CRAH units, CRAC units, and auxiliary ventilation.
Fluid circulation: circulating chilled water, condenser water, glycol mixtures, and dielectric coolants through primary, secondary, and direct-to-chip loops.
Heat rejection and compression: operating chiller compressors, cooling tower fans, and condenser systems to reject heat outside the facility.
These are not incidental building-service loads. If a chilled-water pump, tower fan, compressor, or circulation loop fails to start, trips under load, or recovers badly after a disturbance, the server hall feels the result.
The issue is not simply whether the motor turns. It is how the motor starts, how torque is applied, how protection logic responds, and how the system restarts after a disturbance. Pumps, fans, and compressors place current, torque, and protection demands on the upstream supply, the mechanical system, and the process around them.
In a data center, that process is compute availability.
In hyperscale cooling plants, large chilled-water pumps and chiller-related loads can move into medium-voltage territory, depending on plant scale, pump selection, and distribution design. At that point, the starting and control method becomes part of the electrical and mechanical design of the plant.
An uncontrolled start can draw several times full-load current, depress the local bus, and reduce generator headroom during recovery. The mechanical side is just as important. A hard start can shock the fluid loop and accelerate wear across couplings, bearings, valves, and pipework.
Choosing between a soft starter and a VFD is therefore not just a package decision. It affects how the cooling plant starts, loads, modulates, and recovers after a disturbance.
A poor control strategy can create problems well beyond the motor terminals. Voltage sag, nuisance trips, hydraulic shock, harmonic distortion, and poor restart behavior all become facility-level risks when the load is tied to cooling availability.
Soft starters reduce starting current and mechanical shock where continuous speed control is not required. VFDs can match pump or fan output to changing heat load, but they also bring design obligations around harmonics, bearing currents, grounding, cable length, motor insulation, protection coordination, and system integration.
Uptime is physical, not only electrical. Power, cooling, maintainability, and fault response all sit inside the same availability model. The motor-control system has to work during starting, changeover, isolation, trips, and restarts, not only when the plant is running steadily.
A cooling motor that starts cleanly on normal utility supply may behave differently on generator supply, after fault clearance, or during partial plant operation. Protection settings have to allow for normal starting and recovery transients without ignoring a real electrical or mechanical fault. Overload curves, thermal models, fault thresholds, and restart logic all matter here. A system that trips unnecessarily under peak cooling demand can create its own uptime problem.
This is becoming more important as AI data center buildout puts pressure on grid connections, backup generation, switchgear availability, cooling equipment, construction capacity, and server supply. At the same time, higher rack densities are pushing more heat into liquid-cooled and hybrid architectures. The plant has to control operating cost without giving away uptime margin.
For owners, consulting engineers, contractors, and operations teams, the questions are practical:
Once the plant is live, these decisions appear in power consumption, generator headroom, nuisance trips, bearing life, pipe stress, recovery time, maintenance access, and lifecycle cost.
At Amconex Group, we look at the data center cooling plant as an industrial motor-control system. The facility may support digital infrastructure, but much of the cooling load is familiar industrial equipment: pumps maintaining flow, fans moving air, compressors operating under load, and starters and drives coordinating with the wider electrical system.
Hardware selection is only one part of the work. The starting or speed-control method has to match the duty of the machine, the strength of the supply, the mechanical limits of the system, and the way the plant is expected to recover after a disturbance.
A starter selected to limit inrush current does not provide operating speed control. A VFD selected for variable-speed operation still has to start the motor reliably, manage acceleration, and behave predictably during transfer, fault clearance, or generator-backed operation.
For fixed-speed loads, controlled starting may be the better answer where continuous speed variation is not required. Where pump or fan output needs to be continuously matched to cooling demand, a VFD may provide a strong operating case, particularly on variable-torque loads. In either case, the control method has to be engineered into the power system, motor circuit, and plant sequence from the beginning.
The sequence is as important as the equipment. What starts, what locks out, and what restarts during generator transfer, fault clearance, or maintenance isolation should be defined in the control strategy, not discovered during commissioning.
Through Benshaw and AuCom, Amconex brings decades of low- and medium-voltage motor-control experience to the cooling loops that keep high-density computing online. Soft starters, variable frequency drives, protection systems, and engineered control panels are part of that work, but the engineering decision is still made load by load.
| Cooling load application | Control solution | Primary benefit | Integration focus |
| Fixed-speed high-tonnage chiller compressors | Soft starter | Limits inrush current and mechanical stress during start, without adding speed control the load does not require during normal operation. | Overload coordination, thermal protection, and restart logic to avoid nuisance trips during peak cooling demand. |
| Primary chilled-water pumps | VFD | Harmonic mitigation, grounding, shaft current management, motor protection, and restart behavior. | Harmonic mitigation, grounding, shaft current management, motor protection, and restart behavior. |
| Condenser and cooling tower fans | VFD | Allows airflow to follow heat-rejection demand, with energy savings where fan speed can be continuously modulated. | Harmonic control, EMC management, cable routing, grounding, and protection coordination across the speed range. |
These should be treated as starting points, not rules. Staged tower-cell operation, redundancy strategy, supply strength, and harmonic limits can all shift a given load from one column to the other.
Amconex Group combines the motor-control engineering strengths of Benshaw and AuCom across global low- and medium-voltage applications. Through soft starters, variable frequency drives, protection systems, and engineered control solutions, Amconex supports the reliability of critical motor-driven infrastructure where power and process performance are directly connected.