A—0101 / 05
AI & GPU infrastructure
Context
Training and inference clusters concentrate large thermal loads into a small number of racks.
Engineering challenge
Moving heat away from the chip reliably as equipment generations change.
AI infrastructure / Thermal engineering
Advanced liquid cooling and thermal infrastructure engineered for the demands of high-density AI computing.
The thermal challenge
Fig. 01 — Direct-to-chip cold plates
Heat captured at the source
As computing density increases, thermal management becomes an increasingly important part of infrastructure design.
THERMGRID focuses on the cooling systems, engineered components and supporting infrastructure that help operators manage demanding thermal requirements — from the chip to the point where heat leaves the building.
Cooling approach vs. rack densityConceptual
Air cooling
Rear-door HX
Direct-to-chip
Immersion
Lower densityHigher density →
Indicative only. The right approach depends on the facility, the equipment and the system design.
Engineered systems
How the system works
A simplified direct-to-chip architecture. Actual cooling architectures vary by facility, equipment and system design.
Fig. 06 — Liquid cooling loop, conceptual
Conceptual illustration · not to scale · not a design specificationView our systems
Liquid cooling technology
A coolant distribution unit sits between the racks and the building. How it is configured, connected and maintained shapes how the whole cooling system performs.
Where THERMGRID fits
05 environments
From requirement to system
Stage 01
Establish operational requirements, computing density, thermal loads and facility constraints.
Stage 02
Develop an appropriate cooling architecture and define the system requirements.
Stage 03
Coordinate system components and integration requirements within the project scope.
Stage 04
Provide the relevant documentation, technical coordination and ongoing support where offered.
Featured project
Illustrative case study
An example of how a THERMGRID case study is presented: a new high-density hall planned around direct-to-chip liquid cooling from the outset.
Client
Demo client — not a real organisation
Facility type
High-density AI computing hall
Scope
Secondary loop, CDUs, heat rejection interface
Status
Illustrative template content
01 Context
Project context
A new hall intended for GPU clusters, with racks too dense for air cooling alone.
02 Challenge
Engineering challenge
Delivering liquid cooling to every row while keeping maintenance access clear.
03 Requirements
System requirements
Redundant distribution, separated loops and a phased build-out path.
04 Architecture
Proposed architecture
Row-level CDUs on a closed secondary loop, linked to facility water and dry coolers.
05 Installation
Installation considerations
Overhead headers coordinated with cable routes, with isolation points per row.
06 Outcome
Project outcome
A coordinated cooling system ready for phased commissioning. Values below are demo data.
Illustrative
served by the secondary loop
Illustrative
distribution redundancy
Illustrative
planned build-out
Fig. 07 — In-row CDU, side panel removed
Product imagery
Technical capabilities
Every system is defined around the facility. This sheet shows how technical information is presented — the values are examples only.
Sheet TG-CDU-R
Example values
System category
Coolant distribution unit, in-row
Intended application
Direct-to-chip liquid cooling for high-density racks
Available configuration
Single or redundant pumps, front service access (example)
Integration considerations
Facility water interface, controls and monitoring, rack manifold connection
Documentation
General arrangement, schematics, O&M information (on request)
Project-specific requirements
Capacity, redundancy and connection details defined per project
Why it matters
Fig. 08 — Instrumentation on the secondary loop
01
Thermal design
The cooling architecture sets what the computing room can support over its life.
02
System integration
Racks, distribution and plant work as one system only when they are designed together.
03
Operational requirements
Redundancy, controls and monitoring follow how the facility will actually be run.
04
Maintainability
Access, isolation and documentation decide how safely a live system can be serviced.
Insights
Systems
What Is a Coolant Distribution Unit?
The component between the racks and the building, explained.
Design
Designing Thermal Infrastructure for High-Density Computing
The thermal design decisions for high-density computing that are hard to change later.
Integration
Key Considerations for Data Center Cooling Integration
Interfaces, controls and commissioning: where data center cooling integration succeeds or fails.