Data Center & AI Infrastructure Glossary
Confused by terms like liquid cooling, direct-to-chip, CDU, PUE, GPU density or immersion cooling?
This glossary explains the most important terms used in AI infrastructure, data center cooling, GPU infrastructure, liquid cooling and modular data centers — in simple language.
Whether you're planning an AI data center or simply trying to understand the technology, use this guide to quickly find the terminology you need.
AI & Compute
What is AI infrastructure?
AI infrastructure is the combination of computing, power, cooling, networking, storage and data center systems required to deploy and operate artificial intelligence workloads.
What is an AI data center?
An AI data center is a facility designed to support high-performance AI computing, typically using large numbers of GPUs or other AI accelerators with significant power and cooling requirements.
What is AI compute?
AI compute refers to the computing resources used to train, fine-tune and run artificial intelligence models. GPUs and other specialized accelerators are commonly used.
What is HPC?
HPC stands for High-Performance Computing. It refers to computing systems designed to process complex workloads at very high speeds, including scientific research, simulations, engineering and AI.
What is an AI accelerator?
An AI accelerator is a processor specifically designed to efficiently perform the calculations required by artificial intelligence and machine learning workloads.
What is a GPU?
A GPU, or Graphics Processing Unit, is a processor capable of performing many calculations simultaneously. GPUs are widely used for AI and HPC because they can efficiently handle parallel workloads.
What is a GPU cluster?
A GPU cluster is a group of interconnected GPU servers working together as a larger computing system.
What is GPU infrastructure?
GPU infrastructure includes the servers, racks, power, cooling, networking and facility systems required to deploy GPUs at scale.
What is GPU density?
GPU density describes how many GPUs are concentrated within a server, rack, data center or physical footprint.
What is rack density?
Rack density refers to the amount of electrical power or computing capacity concentrated within a single data center rack, typically measured in kilowatts (kW).
What is a high-density rack?
A high-density rack contains significantly more compute power and therefore generates more heat than a conventional data center rack.
What is a GPU server?
A GPU server is a server designed to contain one or more GPUs for AI, HPC, graphics, simulation or other high-performance workloads.
Data Center Cooling
What is data center cooling?
Data center cooling is the infrastructure used to remove heat generated by computing equipment and maintain appropriate operating temperatures.
Why do data centers need cooling?
Servers, GPUs and other electronic equipment convert electrical energy into heat. Cooling systems remove that heat to maintain reliable operation.
What are the main types of data center cooling?
Common approaches include air cooling, liquid cooling, direct-to-chip cooling, immersion cooling, rear-door heat exchangers and hybrid cooling systems.
What is air cooling?
Air cooling uses fans and conditioned air to remove heat from servers and transfer it away from the data center.
What is liquid cooling?
Liquid cooling uses a liquid coolant to transfer heat away from high-power computing components and into a cooling or heat-rejection system.
Why is liquid cooling important for AI?
AI GPUs can generate substantial heat, particularly at high utilization and rack density. Liquid cooling can transfer heat efficiently from high-power processors and support dense AI deployments.
What is high-density cooling?
High-density cooling refers to cooling infrastructure designed to remove large amounts of heat from densely packed computing equipment.
What is thermal management?
Thermal management is the process of controlling and removing heat from electronic equipment to maintain safe and reliable operating temperatures.
What is heat rejection?
Heat rejection is the process of transferring heat removed from computing equipment to the external environment.
What is a heat exchanger?
A heat exchanger transfers heat between two fluids without necessarily mixing them.
What is a dry cooler?
A dry cooler rejects heat to the surrounding air without relying on evaporative water consumption.
What is a cooling loop?
A cooling loop is a circulating system through which coolant moves between the heat source and the heat-rejection system.
What is closed-loop cooling?
Closed-loop cooling continuously circulates coolant through a sealed or controlled system rather than consuming the coolant during normal operation.
Liquid Cooling
What is direct-to-chip cooling?
Direct-to-chip cooling is a liquid cooling method that transfers heat directly from processors such as GPUs and CPUs to a liquid-cooled cold plate.
What is direct liquid cooling?
Direct liquid cooling is another term commonly used for cooling high-power processors directly with liquid through cold plates or similar interfaces.
What is a cold plate?
A cold plate is a thermally conductive component attached to a processor or other heat source. Coolant flows through it to remove heat.
What is GPU cold plate cooling?
GPU cold plate cooling uses a liquid-cooled cold plate positioned directly against a GPU to remove heat from the processor.
What is CPU cold plate cooling?
CPU cold plate cooling uses a liquid-cooled cold plate to transfer heat away from a CPU.
What is a CDU?
A CDU, or Coolant Distribution Unit, manages coolant temperature, flow and distribution between the facility cooling system and liquid-cooled IT equipment.
Why is a CDU needed?
A CDU can regulate the liquid cooling loop and, depending on the design, separate the IT coolant loop from the facility cooling loop through a heat exchanger.
What is a coolant distribution system?
A coolant distribution system moves liquid between the cooling source and the servers, racks or other heat-generating equipment.
What is a manifold in liquid cooling?
A manifold distributes coolant to multiple servers, cold plates or cooling components within a rack or system.
What is a quick disconnect?
A quick disconnect is a connector that allows liquid cooling lines to be connected or disconnected efficiently while helping minimize coolant leakage.
What is coolant?
Coolant is the fluid used to absorb and transport heat through a cooling system.
Does liquid cooling use water?
Not necessarily. Liquid cooling can use different coolant fluids and can operate in closed loops. Whether a facility consumes water depends largely on the complete heat-rejection architecture.
Immersion Cooling
What is immersion cooling?
Immersion cooling places servers or selected computing components directly into a dielectric cooling fluid that absorbs heat from the equipment.
What is single-phase immersion cooling?
Single-phase immersion cooling uses a dielectric liquid that remains in liquid form while circulating and removing heat.
What is two-phase immersion cooling?
Two-phase immersion cooling uses a fluid that changes from liquid to vapor when it absorbs heat, then condenses back into liquid as part of the cooling cycle.
What is phase-change cooling?
Phase-change cooling uses a change in the physical state of a cooling fluid, such as liquid to vapor, to absorb and transfer heat.
What is dielectric coolant?
A dielectric coolant is an electrically non-conductive fluid that can come into direct contact with electronic components.
Is immersion cooling better than direct-to-chip cooling?
Neither is universally better. Direct-to-chip cooling targets selected high-power components, while immersion cooling surrounds the computing equipment with cooling fluid. The best option depends on density, workload, infrastructure and operational requirements.
Data Center Power & Efficiency
What is PUE?
PUE stands for Power Usage Effectiveness. It is a data center efficiency metric comparing total facility energy consumption with the energy consumed by IT equipment.
What is a good PUE?
A lower PUE generally indicates that a greater proportion of the facility's energy is being used by IT equipment rather than supporting infrastructure. The appropriate target depends on facility design, climate and workload.
What is rack power?
Rack power is the amount of electrical power consumed by equipment within a data center rack, typically measured in kilowatts.
What is power density?
Power density describes how much electrical power is consumed within a particular physical area or volume.
What is IT load?
IT load is the electrical power consumed by computing and networking equipment, excluding supporting facility systems.
What is total facility load?
Total facility load includes IT power plus the power consumed by cooling, power conversion, lighting and other supporting systems.
What is UPS?
UPS stands for Uninterruptible Power Supply. It provides backup power and helps protect IT equipment from power interruptions and disturbances.
What is redundancy?
Redundancy means having additional infrastructure or capacity available so that a component failure does not necessarily interrupt operations.
What is N+1 redundancy?
N+1 means a system has one more unit of capacity than is required for normal operation, allowing one component to fail or be taken offline while maintaining the required capacity.
Water & Sustainability
What is waterless data center cooling?
Waterless data center cooling refers to cooling architectures designed to eliminate or minimize ongoing consumption of fresh water for cooling and heat rejection.
What is a zero-water data center?
A zero-water data center generally refers to a facility designed to eliminate ongoing water consumption for cooling, although the exact definition and system boundaries can vary.
What is water-efficient cooling?
Water-efficient cooling is designed to reduce water consumption while still providing the required thermal performance.
Why does AI increase data center water concerns?
AI can increase compute density and therefore cooling requirements. In water-constrained locations, the cooling architecture can become an important factor in the overall water footprint of a data center.
Can liquid cooling be waterless?
Yes. A closed-loop liquid cooling system can circulate coolant without continuously consuming fresh water. The facility's final heat-rejection system determines overall water consumption.
What is evaporative cooling?
Evaporative cooling uses the evaporation of water to remove heat. It can be effective but consumes water during operation.
What is dry cooling?
Dry cooling rejects heat to ambient air without relying on evaporation as the primary heat-rejection mechanism.
Is waterless cooling more energy efficient?
Not necessarily. Waterless cooling can reduce water consumption, but the energy requirements of the complete cooling and heat-rejection system need to be evaluated separately.
Modular Data Centers
What is a modular data center?
A modular data center is a prefabricated or standardized data center environment that can integrate computing, power, cooling and supporting infrastructure into deployable modules.
What is a containerized data center?
A containerized data center uses a standardized enclosure to house computing and supporting infrastructure.
What is a prefabricated data center?
A prefabricated data center is manufactured or assembled partly in a controlled factory environment before being transported to the deployment site.
What is a modular AI data center?
A modular AI data center is modular infrastructure specifically designed around the high-density compute, power and cooling requirements of AI workloads.
Why use a modular data center?
Modular infrastructure can provide advantages in deployment flexibility, standardization, scalability and potentially speed of installation.
Are modular data centers scalable?
Many modular systems are designed so that additional capacity can be added in stages as compute requirements increase.
Can modular data centers use liquid cooling?
Yes. Modular data centers can integrate liquid cooling, direct-to-chip cooling, immersion cooling or hybrid cooling depending on the design.
Are modular data centers suitable for GPUs?
Yes. Modular infrastructure can be specifically engineered around high-density GPU servers, including the required power and cooling systems.
AI Data Center Infrastructure
What is a GPU data center?
A GPU data center is a facility designed to deploy and operate large numbers of GPUs, typically for AI, HPC or other high-performance workloads.
What is an AI data center rack?
An AI data center rack is a rack designed to accommodate high-performance AI servers and the higher power and cooling requirements associated with them.
What is high-density AI infrastructure?
High-density AI infrastructure is designed to support large amounts of AI compute within a relatively small physical footprint.
What is AI server cooling?
AI server cooling is the thermal management infrastructure used to remove heat from AI servers and their high-performance processors.
What is GPU thermal management?
GPU thermal management is the process of controlling and removing heat generated by GPUs during operation.
What is rack-level cooling?
Rack-level cooling manages heat generated by equipment within an individual rack, potentially using air, liquid or hybrid technologies.
What is facility-level cooling?
Facility-level cooling is the broader infrastructure responsible for rejecting heat from the data center and maintaining appropriate environmental conditions.
Data Center Infrastructure
What is a data center?
A data center is a facility containing computing, networking, power, cooling and other infrastructure used to operate digital systems and services.
What is a hyperscale data center?
A hyperscale data center is a very large computing facility designed to support massive amounts of compute and storage capacity, typically operated by major cloud or technology companies.
What is a colocation data center?
A colocation data center provides physical infrastructure where multiple organizations can house their servers and IT equipment.
What is edge computing?
Edge computing places computing resources closer to where data is generated or consumed, reducing the need to send all workloads to centralized facilities.
What is data center infrastructure?
Data center infrastructure includes the physical systems required to operate IT equipment, including power, cooling, racks, networking, security and facility systems.
What is mission-critical infrastructure?
Mission-critical infrastructure is designed to support systems where downtime can have significant operational or financial consequences.
AI Hardware & GPU Generations
What is NVIDIA Blackwell?
Blackwell is a generation of NVIDIA GPUs and accelerated computing platforms designed for demanding AI and HPC workloads.
What is NVIDIA Hopper?
Hopper is a generation of NVIDIA accelerated computing architecture used extensively in AI and HPC applications.
What is an H100?
The NVIDIA H100 is a data center GPU designed for AI and high-performance computing workloads.
What is a B200?
The NVIDIA B200 is a Blackwell-generation data center GPU designed for high-performance AI computing.
What is GB200?
The NVIDIA GB200 is a Blackwell-based accelerated computing platform combining NVIDIA Grace CPU technology with Blackwell GPUs.
Why do new GPU generations affect cooling?
As GPU performance and power requirements increase, the amount of heat generated by each processor can also increase. This can drive higher rack densities and greater demand for advanced cooling.
Infrastructure Planning
How do you choose a data center cooling system?
Consider the processor platform, rack density, IT load, climate, power availability, water constraints, existing infrastructure, deployment timeline and future scalability.
When should liquid cooling be considered?
Liquid cooling should be evaluated when processor power and rack density create thermal requirements that may be difficult or inefficient to manage using conventional air cooling alone.
Can an existing data center be upgraded for AI?
Potentially. The feasibility depends on available power, cooling capacity, rack configuration, structural capacity, networking and the ability to integrate the required infrastructure.
Can a traditional data center support GPUs?
Yes, depending on the facility's power, cooling, rack density and networking capabilities. High-density GPU deployments may require significant infrastructure upgrades.
How do you calculate data center cooling requirements?
Cooling requirements are primarily driven by IT power and the amount of heat generated by the computing equipment, along with environmental and infrastructure factors.
How do you calculate GPU rack density?
GPU rack density can be evaluated using the total electrical power and compute equipment installed in the rack, typically expressed in kW per rack.
What should you consider when building an AI data center?
Key considerations include GPU requirements, power availability, cooling architecture, rack density, networking, water availability, site conditions, deployment timeline, scalability and total cost of ownership.
Need Help With an AI Infrastructure Project?
Understanding the terminology is the first step. Choosing the right infrastructure is the next.
If you're planning an AI data center, GPU cluster, HPC deployment or high-density compute environment, we can help you evaluate the cooling, power, modular infrastructure and deployment requirements.
Where there is a strong fit, we can facilitate introductions to experienced technology providers specializing in AI infrastructure, liquid cooling, direct-to-chip cooling, immersion cooling and modular data centers.
Discuss Your Project →