Collection: Liquid Cooling

Air Has Limits. Liquid Doesn't.

The GPU industry has a thermal problem, and it's getting worse. Each generation of AI accelerators delivers more compute performance — and generates more heat per chip. The H100 SXM5 has a 700W TDP. Next-generation accelerators are projected to exceed 1000W per chip. A single 8-GPU server can generate 8–10kW of thermal load from the GPUs alone, before accounting for CPUs, memory, storage, and networking. At these densities, air cooling is no longer a viable long-term strategy — it's a temporary solution that becomes increasingly expensive and ineffective as power densities continue to rise.

Liquid cooling removes heat from hardware 3,000 times more efficiently than air. It enables rack densities that are physically impossible with air cooling. It reduces cooling energy consumption by 30–60% compared to equivalent air-cooled deployments. And it enables quieter, more compact AI infrastructure environments that can be deployed in spaces where a traditional air-cooled data center would be impractical.

DVUN's Liquid Cooling collection brings these capabilities to AI infrastructure teams who are ready to move beyond the limits of air cooling. We stock direct liquid cooling (DLC) systems, rear-door heat exchangers, and immersion cooling components — covering the full spectrum of liquid cooling approaches from incremental air-cooling augmentation to full liquid immersion.

Three Approaches to Liquid Cooling

• Rear-Door Heat Exchangers (RDHx): The lowest-disruption entry point to liquid cooling. A rear-door heat exchanger replaces the standard rear door of an existing rack enclosure with a liquid-cooled heat exchanger that captures hot exhaust air before it enters the room environment. No server modifications required. Cooling capacity up to 60kW per rack. The right choice for teams that want liquid cooling benefits without modifying existing servers.
• Direct Liquid Cooling (DLC): Cold plates are attached directly to CPUs, GPUs, and other high-heat components, with liquid circulated through the cold plates to remove heat at the source. Dramatically more efficient than air cooling for the highest-power components. Requires server-level integration but delivers the highest cooling efficiency and enables the highest rack densities. Cooling capacity up to 100kW per rack.
• Immersion Cooling: Servers are submerged in a dielectric fluid that absorbs heat directly from all components simultaneously. The most efficient cooling approach available, enabling rack densities of 200kW+ and eliminating fans entirely. Best suited for new deployments where the infrastructure can be designed around immersion from the start.

Liquid Cooling Infrastructure Components

• Rear-door heat exchangers: 20kW to 60kW cooling capacity, compatible with standard 42U enclosures
• DLC cold plate kits: compatible with major GPU and CPU platforms
• Coolant distribution units (CDUs): 50kW to 200kW capacity, with pump redundancy options
• Manifolds and quick-disconnect fittings: leak-resistant connections for reliable liquid distribution
• Immersion cooling tanks: 10kW to 200kW capacity, single-phase and two-phase fluid options
• Monitoring systems: flow rate, temperature, and pressure monitoring for liquid cooling infrastructure

Is Liquid Cooling Right for Your Deployment?

If your rack power density exceeds 20kW, liquid cooling is worth serious consideration. If it exceeds 30kW, it's likely the most cost-effective long-term solution. DVUN's team can help you evaluate the right liquid cooling approach for your specific environment, power density, and budget — and design a transition path that minimizes disruption to existing infrastructure. See our Airflow collection for air cooling optimization options, and our Facility & Cooling overview for the complete physical infrastructure picture. Request a consultation to discuss your liquid cooling requirements.