AI data centers are hitting a thermal wall, and liquid cooling is becoming the only viable solution to keep next-generation AI chips from overheating. LG Electronics just qualified its 2.6-megawatt Coolant Distribution Unit (CDU) to work with NVIDIA's DSX AI Factory Platform, marking a significant milestone in how the industry manages the extreme heat generated by modern AI accelerators.

What's Driving the Shift to Liquid Cooling?

The problem is straightforward: AI chips are getting more powerful and more power-hungry. With NVIDIA's latest Blackwell architecture and other next-generation accelerators, individual server racks now consume hundreds of kilowatts of power. That's an enormous amount of heat concentrated in a small space. Traditional air-cooling systems, which blow cold air across equipment, simply cannot keep up anymore.

Instead of cooling individual racks one at a time, data centers are shifting toward centralized, high-capacity liquid cooling systems. These systems pump coolant directly through cold plates attached to CPUs and GPUs, absorbing heat far more efficiently than air alone. The coolant then carries that heat away to the facility's main cooling infrastructure, where it can be rejected to the outside environment.

The market is responding rapidly. According to global market research firm TrendForce, liquid cooling penetration for AI chips is expected to rise from 14 percent in 2024 to approximately 60 percent by 2027. That's a dramatic acceleration in just three years.

How Does LG's New Cooling System Work?

LG's 2.6-megawatt CDU is designed to handle the thermal loads of large-scale AI server clusters. To put that in perspective, this single cooling unit can manage the heat output of multiple high-density server racks simultaneously, allowing data center operators to use their physical space more efficiently while maintaining system reliability.

The qualification from NVIDIA is significant because it means LG's system meets strict functional requirements for the DSX AI Factory Platform, a unified framework that covers compute, networking, power, cooling, facilities, and software. This isn't just a marketing certification; it's a technical validation that the system can reliably handle real-world AI infrastructure demands.

LG has now qualified three different CDU models across a range of capacities:

• 600-kilowatt model: Handles smaller, more distributed cooling needs for mid-scale AI deployments
• 1-megawatt model: Bridges the gap between smaller and larger installations, offering flexibility for growing data centers
• 2.6-megawatt model: Classified as a "Large-Scale" solution by NVIDIA, designed for the most demanding AI infrastructure

This tiered approach allows LG to serve data centers of different sizes, from emerging AI facilities to hyperscale operations run by major cloud providers.

Why This Matters Beyond Just Cooling

Liquid cooling is part of a broader ecosystem that LG calls its "Chip-to-Chiller" portfolio. This comprehensive approach addresses thermal management at every level of a data center, from the chip itself all the way to the facility's main cooling systems. The portfolio includes high-efficiency chillers, Computer Room Air Handlers (CRAHs), and direct-to-chip liquid cooling solutions, all designed to work together seamlessly.

"NVIDIA's qualification of our 2.6MW CDU for AI factory infrastructure reinforces our position as a first mover in the next-generation liquid cooling market," said James Lee, president of the LG Eco Solution Company. "We will continue to supply AI data center operators with highly reliable, in-house developed cooling solutions from our comprehensive Chip-to-Chiller portfolio."

James Lee, President of LG Eco Solution Company

The significance of this statement is that LG is positioning itself not just as a cooling vendor, but as a complete thermal management partner. As AI infrastructure becomes more power-intensive and compute-dense, data center operators need solutions that work across multiple layers of their facilities, not just point solutions for individual problems.

Steps to Evaluate Liquid Cooling for Your Data Center

• Assess your power density: Calculate the kilowatts per rack in your facility. If you're approaching or exceeding 50 kilowatts per rack, liquid cooling becomes increasingly necessary for reliability and efficiency
• Check vendor qualifications: Verify that any cooling system you consider meets industry standards like NVIDIA DSX Ready certification, which ensures compatibility with modern AI accelerators and platforms
• Plan for scalability: Choose a cooling solution that can grow with your infrastructure. Modular systems like LG's tiered CDU approach allow you to start smaller and expand as your AI workloads increase

The broader implication is clear: as AI models grow larger and more sophisticated, the infrastructure supporting them must evolve too. Liquid cooling is no longer a luxury or a niche technology; it's becoming a fundamental requirement for competitive AI data center operations.