Cooling technology Technology

Direct-to-chip liquid cooling

A metal plate sits directly on each processor, and liquid pumped through it carries the heat out of the rack.

DLCcold platesCDUAI racks

Direct answer

What this record says

Direct-to-chip liquid cooling is a scaling cooling architecture using water, usually with glycol, in the loop that reaches the servers. It is typically applied at Commonly selected for 50–150+ kW AI racks. Last reviewed Aug. 9, 2026 against 1 source.

1 sourceReviewed Aug. 9, 2026
Maturity
Scaling
Coolant
Water, usually with glycol, in the loop that reaches the servers
Typical density
Commonly selected for 50–150+ kW AI racks
Suppliers tracked
6 suppliers
MaturityScaling
Cooling mediumWater, usually with glycol, in the loop that reaches the servers
Typical applicationCommonly selected for 50–150+ kW AI racks

Architecture

How heat moves

Direct-to-chip cooling places cold plates on the highest-heat components, usually graphics processing units and central processing units. Water or a water-glycol mix stays liquid as it moves from a rack manifold through each plate. It returns warm to a coolant distribution unit (CDU), where a heat exchanger passes the load to facility water without mixing the loops.

The architecture fits new high-density rows and phased retrofits where supported servers are available. It captures only the components touched by plates, so memory, drives, power supplies, and network devices can leave a material air load. Facility teams need primary water at the required temperatures and flow, space and power for CDUs, compatible piping, water treatment, leak detection, controls, and enough air cooling for the remainder.

Published plate or CDU capacity is tied to inlet temperature, flow, fluid mix, pressure loss, and approach temperature; compare products at the same conditions. Ask which exact server configurations are approved, what percentage of rack heat enters the liquid, and what happens after a pump, power, control, or facility-water failure. Also decide who owns fluid chemistry, quick-disconnect inspection, commissioning, alarms, and warranty boundaries across the server, manifold, CDU, and plant.

01Chip
02cold plate
03rack liquid loop
04coolant distribution unit
05building water

Why teams choose it

  • Removes heat at the hottest point, before it reaches the air
  • Preserves rack-based service when servers are designed for the loop
  • Can deliver warmer return water than room-air systems, which can help heat reuse

What to validate

  • Connections, water quality, and leak detection all demand disciplined operations
  • Memory, power supplies, and networking still shed heat into the air
  • The coolant distribution units have to be sized and made redundant to match the compute rollout

Connected records

Companies in this category

View all companies in this category

Connected records

Required equipment

View all required equipment

Connected records

Tracked deployments

View all tracked deployments

Evidence ledger

Sources and evidence

The links below show where the factual claims came from. Supplier specifications remain supplier-reported unless the record names independent operating evidence.

  1. 01
Record information
Record ID
DCC / TECH / DIRECT-TO-CH
Record reviewed
Aug. 9, 2026
Record first published
Aug. 9, 2026
External sources
1
Search visibility
Offered to search engines

Latest coverage

Intelligence mentioning Direct-to-chip liquid cooling

View all intelligence mentioning direct-to-chip liquid cooling

Decision support

Guides and comparisons mentioning Direct-to-chip liquid cooling

Direct answers

Frequently asked questions

Direct answers drawn from the record, its comparison fields, and the evidence linked below.

What is Direct-to-chip liquid cooling?

Direct-to-chip cooling places cold plates on the highest-heat components, usually graphics processing units and central processing units. Water or a water-glycol mix stays liquid as it moves from a rack manifold through each plate. It returns warm to a coolant distribution unit (CDU), where a heat exchanger passes the load to facility water without mixing the loops. The architecture fits new high-density rows and phased retrofits where supported servers are available. It captures only the components touched by plates, so memory, drives, power supplies, and network devices can leave a material air load. Facility teams need primary water at the required temperatures and flow, space and power for CDUs, compatible piping, water treatment, leak detection, controls, and enough air cooling for the remainder. Published plate or CDU capacity is tied to inlet temperature, flow, fluid mix, pressure loss, and approach temperature; compare products at the same conditions. Ask which exact server configurations are approved, what percentage of rack heat enters the liquid, and what happens after a pump, power, control, or facility-water failure. Also decide who owns fluid chemistry, quick-disconnect inspection, commissioning, alarms, and warranty boundaries across the server, manifold, CDU, and plant.

Where is Direct-to-chip liquid cooling typically used?

Commonly selected for 50–150+ kW AI racks. We class the approach itself as scaling.

What are the main advantages of Direct-to-chip liquid cooling?

Removes heat at the hottest point, before it reaches the air. Preserves rack-based service when servers are designed for the loop. Can deliver warmer return water than room-air systems, which can help heat reuse.