The Hidden Power Drain Behind AI Innovation
Artificial intelligence (AI) is transforming everything from personalized recommendations to autonomous vehicles, but this digital revolution runs on hardware. We’re not just talking about the high-profile GPUs and processors. Behind the scenes, a vast amount of energy is quietly consumed by a different kind of workhorse: the HVAC systems responsible for keeping AI data centers cool and operational. As AI models grow in complexity and hardware becomes denser and hotter, cooling has shifted from an infrastructure afterthought to a strategic design challenge. Efficiency isn’t just nice to have; it’s absolutely necessary.
Updated October 8, 2026 to address direct-to-chip liquid cooling, coolant distribution units, and motor efficiency across the cooling system.
AI Data Centers: A New Breed of Energy Demands
AI data centers are fundamentally different from traditional server farms. Housing racks of high-performance graphics processing units (GPUs), application-specific integrated circuits (ASICs), and specialized accelerators, these centers are designed for massive parallel computation. That level of performance requires a lot of energy, which generates a great deal of heat.
The thermal load in modern AI data centers can be more than double what was typical in server centers a decade ago. That heat must be removed continuously, reliably, and efficiently. Enter the HVAC system, the unsung hero responsible for thermal stability and uptime.
The Cooling Challenge: Air, Liquid, and the Energy Needed to Move Heat
Cooling can account for a substantial share of a data center’s energy use. As rack power density rises, thermal design extends beyond moving conditioned air through the data hall. Direct-to-chip liquid cooling moves heat from processors into a circulating coolant, which must then carry that heat to the facility’s cooling infrastructure.
In a typical liquid-cooled architecture, a coolant distribution unit (CDU) separates the facility water system (FWS) from the technology cooling system (TCS) serving the IT equipment. A heat exchanger transfers heat between the circuits; the CDU can also include pumps, valves, temperature control, and monitoring. The facility-water temperature and the coolant temperature supplied to servers are therefore distinct design values.
For illustration, facility water might enter a CDU at 32°C while coolant leaves it for the servers at 36°C. The 4°C difference is a heat-exchanger approach temperature. It is not the temperature rise through the server loop used to estimate coolant flow.
A Simple Coolant-Flow Example
Consider a 600 kW heat load with a water-based TCS at 36°C supply and 46°C return. The loop temperature rise is 10°C. Using Q = ṁ × Cp × ΔT and a water heat capacity of approximately 4.18 kJ/(kg·°C), the estimated flow is 14.4 kg/s, or about 227 US gallons per minute (assuming water near 1 kg/L). This is an illustrative heat balance, not a pump or piping specification; actual coolant properties, pressure losses, redundancy, and equipment limits govern design.
As compute density increases, the energy needed to move that coolant becomes part of the cooling-efficiency equation. ASHRAE’s AI data center energy and thermal guidance considers pumps, CDUs, distribution networks, and heat rejection together with the IT thermal load.
Motors at the Core: What Drives the Cooling of AI Infrastructure
Whether heat leaves a rack through air, liquid, or both, it still has to be transported. Electric motors drive the fans, blowers, pumps, compressors, and heat-rejection equipment that perform that work.
In a liquid-cooling system, pumps may respond to changing rack loads, flow requirements, differential pressure, and coolant temperatures. ASHRAE identifies variable-speed pumping as a way to match pumping energy to real-time thermal demand. Motor evaluation should therefore consider efficiency across the actual speed and load profile, along with package size, heat, reliability, controls, and manufacturing requirements—not only the rated-load efficiency point.
Conventional laminated-steel motors remain effective in many cooling applications. Where a compact, variable-speed, or tightly integrated system calls for a different topology, engineers can also evaluate alternative magnetic materials and motor architectures. That is where soft magnetic composites (SMCs) may offer design flexibility.
Where Soft Magnetic Composites May Help
Soft magnetic composites are made from individually insulated ferromagnetic particles. Their magnetically isotropic behavior can support three-dimensional flux paths and geometries that are difficult to realize with stacked electrical steel. Their value is strongest when those properties enable a better motor architecture for the application, rather than a direct material swap in an unchanged design.
- Topology flexibility: Three-dimensional flux paths can support axial-flux, transverse-flux, and hybrid motor architectures where the system benefits from a different shape or winding arrangement.
- Loss tradeoffs: Particle insulation can limit eddy-current losses, especially as electrical frequency rises. Total motor efficiency still depends on hysteresis loss, winding loss, geometry, drive strategy, and operating speed.
- System integration: A motor’s package, thermal path, manufacturability, and performance across the duty cycle can matter as much as its peak efficiency.
SMCs are one option in a broader engineering trade study. A well-designed laminated motor may remain the best choice when its performance, cost, and production requirements fit the cooling system.
Evaluating Energy Savings Across the Duty Cycle
The savings from a more efficient motor should be calculated from the electrical power drawn by the specific cooling equipment over its real operating profile. For example, a 3% reduction in a pump motor system’s 100 kW electrical input would save about 3 kW while operating at that point, or about 26,000 kWh over 8,760 hours at that same load. Actual annual savings depend on speed, load, controls, run hours, and the complete pump-and-drive system.
That makes system-level testing more useful than applying a percentage improvement to a facility’s entire cooling load. Engineers should compare candidate designs at representative flow and pressure points and account for both motor and drive losses.
Looking Ahead: What About the Rotor?
SMC discussions often focus on the stator, but rotor design also deserves attention. Sintered soft magnetic materials may offer another design route when density, shape, strength, or production needs support their use. Any benefit to loss, inertia, or heat management needs to be demonstrated in the complete motor and cooling-system duty cycle.
Unlocking Innovation
As AI data centers move toward higher compute density, thermal management becomes a system-design challenge. Heat may travel through air, direct-to-chip coolant, facility water, refrigerant, or a combination of them. Moving it consumes energy at every stage.
That puts the motors driving pumps, fans, blowers, compressors, and heat-rejection equipment directly into the efficiency conversation. SMCs give engineers another design option when topology, electrical frequency, winding strategy, package size, variable-speed performance, or manufacturing constraints call for a different architecture.
If the cooling architecture is changing, should the motors driving it remain unchanged?
