The rapid growth of AI workloads has fundamentally reshaped the data center market. The latest GPU servers demand significantly higher heat densities, rendering traditional air cooling inadequate. The solution is Direct-to-Chip (DTC) liquid cooling.
However, fast technological adoption is always accompanied by risks. A rupture in the liquid cooling pipeline can lead to system shutdowns and potential damage to valuable GPU racks. Losses can amount to millions of dollars. Such incidents raise a crucial question: how can DTC be implemented while maintaining high levels of reliability and safety?
The Problem: Leaks and Human Error
Leak Risk Is Never Zero. Even with high-quality installation, micro-fissures, joint depressurization, or material defects are still possible.
Piping Located Above Racks creates an additional threat: a breach could cause coolant to drip onto expensive electronics.
Shortage of Qualified Specialists in regions experiencing rapid data center growth increases the probability of errors during installation and maintenance.
The Result: Any damage to the cooling loop can lead to downtime or catastrophic financial losses.
The Solution: Transition to Negative Pressure
We propose modernizing the liquid cooling principle of operation.
Instead of the classic scheme where fluid is supplied under pressure, the pump is moved to the “return side” of the loop. The system now operates in a negative pressure mode—the fluid is “sucked” by the pump rather than being pushed through the pipelines.
We fully understand: the risk of leaks in liquid cooling systems is never zero. Therefore, instead of battling the consequences, we have altered the fundamental operating principle—we switched the loop to a negative pressure mode. In this configuration, even in the event of depressurization, liquid will not come into contact with the equipment. It’s a simple yet robust engineering solution that drastically reduces the threat of failures and protects the data center from millions in losses.
How It Works:
If a crack or loose connection appears, the system does not eject liquid outward.
Instead, air is drawn into the loop.
Installed bubble sensors (ultrasonic or optical) instantly detect the ingress of air.
A signal from the sensor automatically shuts down the racks and the pump, preventing overheating and further escalation of the incident.
Advantages of the Approach
Server Protection: Eliminates the risk of conductive liquid reaching the electronics.
Resource Savings: With minor damage, coolant is not lost; repairs are limited to localized pipe segment replacement.
Early Diagnosis: Air in the loop is a natural indicator of a leak, allowing the problem to be detected before serious consequences occur.
Automation: The detection and emergency shutdown process completely eliminates the human factor.
Conclusion
The transition to a negative pressure liquid cooling system is a significant step forward in data center infrastructure development.
This modernization not only meets the demands of GPU workloads but also enhances the reliability of data centers, mitigating the risks of costly incidents.
DTC IS TECHNOLOGY THAT COMBINES ENERGY EFFICIENCY AND SAFETY.