NVIDIA AI Power & Cooling

Follow power from the utility to the server. You learn the path through transformers, switchgear, UPS and distribution to the rack PDU, and where each stage can limit capacity.

• From utility feed to the data hall
• Transformers, switchgear and UPS systems
• Power distribution to the rack
• AC and DC distribution
• Phases, voltage and current
• Power supplies inside the server

You build A single-line sketch of the power path for a small AI room, with capacity at each stage.

Power must keep running and stay safe. You compare redundancy schemes, understand what protection devices do when something fails, and see why grounding and bonding matter for dense equipment.

• N, N+1 and 2N designs
• Circuit breakers and protection
• Failover behavior of a server with dual supplies
• Backup generation and batteries
• Grounding and bonding
• Planning maintenance without downtime

You build A redundancy comparison showing what each design survives and what it costs.

Capacity planning keeps projects from stalling. You add up server and network loads, apply derating and headroom rules, and decide how many AI systems a rack, row and room can carry.

• Nameplate against typical draw
• Derating and headroom
• Per-rack and per-row budgets
• Power capping and its effects
• Metering and measuring real draw
• Growth and expansion planning

You build A capacity worksheet that fills a row with AI systems within its power limit.

To remove heat you first need to understand it. You learn how heat moves, how airflow is managed in a room, and why mixing hot and cold air wastes capacity.

• Heat transfer: conduction, convection and radiation
• Airflow through a server and a rack
• Hot-aisle and cold-aisle layouts
• Containment and blanking
• Temperature and humidity ranges
• Measuring thermal conditions

You build A thermal map of an example room showing where air mixes and where it should not.

Air still cools most equipment, up to a point. You learn how air cooling units work, how to size them against a heat load, and how to recognize when density has outgrown air.

• Cooling units and chilled water basics
• Sizing cooling against a heat load
• Fan power and airflow limits
• Rear-door heat exchangers
• Signs that air cooling is failing
• Hybrid designs

You build A sizing check that shows at what rack density an example room can no longer be cooled by air.

Liquid carries far more heat than air. You learn the main approaches, how coolant circulates between rack and facility, and what changes in installation, monitoring and maintenance.

• Direct-to-chip liquid cooling
• Immersion cooling at an overview level
• Coolant distribution units and manifolds
• Facility water and rack loops
• Leak detection and response
• Maintenance and service practices

You build A diagram of a liquid-cooled rack and its loops, with the checks needed at each connection.

Heat removed from the room still has to go somewhere. You learn how heat is rejected outdoors, how climate affects the choice, and how efficiency metrics such as PUE describe a facility.

• Chillers, cooling towers and dry coolers
• Free cooling and climate effects
• Heat reuse at an overview level
• Water use in cooling
• PUE and what it does and does not say
• Reporting efficiency honestly

You build An efficiency comparison of two cooling designs, using PUE and water use.

Power and cooling are watched constantly. You learn which sensors and alarms matter, what safe work around power and coolant looks like, and how to respond when a limit is reached.

• Power and thermal monitoring points
• Alarms and escalation
• Electrical safety around racks
• Safe handling of coolants
• Responding to a cooling failure
• Preventive maintenance routines

You build A monitoring and response plan for a power or cooling failure in an AI room.

Plan one room end to end. You size power, choose a cooling approach, decide how heat leaves the building, and describe how the room will be monitored and kept safe.

• Reading the requirement
• Sizing and routing power
• Choosing the cooling approach
• Planning heat rejection
• Writing the monitoring plan

You build A power and cooling plan for one AI room, with calculations and monitoring.