Liquid Cooling
Direct-to-chip & immersion cooling for dense GPU racks
Liquid cooling — primarily direct-to-chip (DLC) cold plates and, secondarily, immersion — has become mandatory for AI racks as GPU/rack power crossed the ~40-50kW ceiling of air cooling toward 120-150kW+ (e.g., GB200 NVL72). It shifts data-center design from CRAH air handling to coolant distribution units (CDUs), manifolds, and facility water loops. Adoption is now the default for frontier training clusters, driving a fast-growing thermal-management supply chain.
Trigger
Racks >40-50kW exceed air-cooling limit
NVL72
~120-150kW/rack, DLC required
Types
Direct-to-chip (dominant) + immersion
How it fits the stack
Liquid Cooling with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.
Liquid Cooling in the AI stack. Liquid Cooling with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.
Graph data (text) — 10 entities, 9 relationships
- Liquid Cooling —manufactures→ Auras / AVC / Cooler cluster (Taiwan thermal)
- Amazon Web Services (AWS) —uses→ Liquid Cooling
- CoreWeave (financing) —uses→ Liquid Cooling
- CoreWeave Data Centers —uses→ Liquid Cooling
- Crusoe Energy —uses→ Liquid Cooling
- Digital Realty —uses→ Liquid Cooling
- Equinix —uses→ Liquid Cooling
- Foxconn (Hon Hai) —depends on→ Liquid Cooling
- Liquid Cooling —competes with→ Immersion Cooling
Depends on ↑ · 1
Feeds ↓ · 21
Context — capital, rivals, policy · · 1