US Liquid Cooling editorial · Reviewed · 4 min read
Many liquid-cooled servers still need air cooling. A direct-to-chip system may collect heat from selected processors while memory, power components, storage, network hardware or other equipment continues to reject heat into the room.
Some newer platforms are designed for much more complete liquid heat capture. The label “liquid cooled” alone does not tell you which arrangement a proposed rack uses.
Ask for the liquid-capture fraction
The liquid-capture fraction is the portion of heat transferred into the liquid circuit under the stated operating conditions. The remaining load needs another cooling path, usually involving air.
Lenovo's Neptune direct-water-cooling standards explicitly discuss residual heat and the air-cooling equipment needed to manage it. This makes the capture fraction an input to facility planning, not just a marketing comparison between cooling technologies.
Obtain it for the complete proposed configuration and relevant workloads. A CPU cold plate's presence is not a measurement of the rack's overall liquid fraction.
A small percentage can still be a substantial load
For an illustrative 120 kW rack with 80% of its heat entering liquid, the remaining air-side load is 24 kW. Across ten such racks, that is 240 kW before adding other room loads.
The calculation is air-side heat = total relevant heat × (1 − liquid fraction). The percentages here are assumptions for explanation, not a typical value or a product rating.
This is why an installation can achieve a high liquid-capture percentage and still require meaningful air management. The air-side system needs the actual remaining load, its location and the equipment's inlet requirements.
“Fanless compute tray” is a precise boundary
NVIDIA's October 2026 OCP announcement describes fanless liquid-cooled compute trays within its updated MGX architecture. That is a development worth following, but a statement about a compute tray should not automatically be expanded to every item in an entire data hall.
A facility may contain different rack generations, storage, networking and infrastructure with different requirements. Identify which components are included in a “100% liquid cooled” claim and how supporting equipment is handled.
For a mixed deployment, keep a separate schedule for each equipment group rather than applying the newest platform's cooling description to the whole room.
Can a rear-door heat exchanger handle the remainder?
Potentially, if the selected rear-door system, airflow and liquid conditions match the residual duty and rack arrangement. Other projects may use in-row cooling, room air handlers or a combination.
Vertiv's direct-to-chip cooling explanation describes liquid cooling as an infrastructure integration problem, including the continuing role of air-side systems in hybrid arrangements.
Do not double-count the heat. If a rear door transfers residual server exhaust heat into a water circuit, that load belongs in the relevant heat-rejection calculation. Its presence does not make the residual heat vanish from the facility energy balance.
Plan normal operation and loss of liquid cooling separately
A server's normal air-cooling requirement does not establish what happens after loss of liquid flow. Some equipment may reduce power, stop workloads or require shutdown. Existing room airflow is not automatically an emergency substitute for cold plates.
Ask the OEM for the supported fault response and required coordination with facility controls. Then determine whether the remaining air-side equipment can manage the transient conditions described, or whether the required response is controlled IT shutdown.
This should be a documented sequence, not an assumption that “the fans will take over.”
What belongs in the equipment schedule?
For each rack configuration, request:
- Total design electrical load and heat boundary.
- Heat captured by each liquid circuit.
- Residual air-side duty and permitted inlet conditions.
- Required airflow and any exhaust or containment constraints.
- Normal and fault-mode operating sequence.
Check the schedule again when the hardware changes. Replacing a few trays or adding network equipment can change the air/liquid balance without changing the rack's name in an asset register.
Our rear-door heat exchanger overview explains one possible air-side solution. For an existing building, use the liquid-cooling retrofit guide to turn the heat split into an infrastructure review.
Sources & further reading
- Lenovo Neptune Direct Water-Cooling Standards > Lenovo Press ↗
- Amid Ecosystem Adoption, NVIDIA Opens Vera Rubin Architecture to Help Scale AI Factories Worldwide | NVIDIA Blog ↗
- Understanding direct-to-chip cooling in HPC infrastructure: A deep dive into liquid cooling ↗
Explore the company directory → · All liquid-cooling guides →