Advanced GPU Liquid Cooling for AI Servers

AI servers pack increasingly powerful GPUs into limited server and rack space, creating much higher and more concentrated heat loads than traditional computing systems. As AI workloads grow rapidly in scale and intensity, GPU liquid cooling is shifting from a specialized solution to a core requirement for high-density AI infrastructure, helping maintain stable performance while supporting continued growth in computing density.

Cooling Challenges for AI Servers and GPUs

The rapid growth of AI computing is pushing conventional air and single-phase liquid cooling systems to their practical limits. Common cooling challenges include:

• Rapid Growth in Chip Power

Next-generation AI accelerators, including GPUs and TPUs, are moving into the kilowatt-class thermal design power (TDP) range, with selected high-end processors exceeding 1.5 kW. NVIDIA’s Rubin platform, for example, has a TDP of 2.3kW per GPU, and the upcoming Rubin Ultra is expected to exceed 3kW. At the same time, advanced server CPUs from NVIDIA and AMD are approaching the 600 W range. This rapid increase in chip-level heat means thermal management can no longer be treated as a secondary design consideration; it is now a key determinant of sustained server performance.

• Ultra-High Density Computing

As more GPUs and accelerators are packed into each server and rack, more heat is generated within the same limited space, creating concentrated heat loads that are increasingly difficult to manage.

• Surging LLM Inference and Memory Demands

LLM workloads span three major stages: training, fine-tuning, and inference. Today’s high-compute AI servers focus heavily on training, but as more AI models are deployed in real-world applications, servers must continuously process a growing volume of inference requests. During these workloads, GPUs perform the AI calculations, and nearby high-bandwidth memory (HBM) supplies the required data at high speed. Keeping both components under sustained load concentrates more heat within the limited server space, placing greater demands on the cooling system.

• Limitations of Conventional Cooling

Air cooling requires increasingly powerful fans to move enough air through compact servers, increasing energy consumption and noise. While conventional single-phase liquid cooling can remove higher heat loads than air, it heavily relies on chemical coolants (such as PG-25) and requires higher flow rates and fluid velocities to carry heat away via temperature change alone. This drives up pump energy consumption and makes pressure drop and leak risks significantly harder to manage. Furthermore, chemical additives pose environmental concerns and long-term maintenance risks, making single-phase cooling increasingly difficult to scale sustainably. 

To address these challenges, CreAlien developed W2PC.

CreAlien’s W2PC (DI Water Two-Phase Cooling)

CreAlien’s W2PC is a proprietary GPU liquid cooling solution developed for AI servers and other high-density computing environments. Key technologies of the W2PC system include:

• DI Water Two-Phase Cold Plates

Installed directly on high-power GPUs and other AI processors, the cold plates allow DI water to absorb massive chip heat and undergo a phase change from liquid to vapor, carrying concentrated heat away far more efficiently than single-phase sensible heat transfer.

• In-Rack and In-Row Coolant Distribution Units (CDUs)

CreAlien offers in-rack CDUs for localized rack-level cooling and in-row CDUs for higher-capacity, multi-rack deployments, allowing W2PC to support AI infrastructure at different scales.

• Chiller-less, Free-Cooling Ready Design

The warm-water design can eliminate the need for mechanical chillers. When outdoor conditions are suitable, the facility can use outdoor air to release heat (free cooling), significantly simplifying cooling infrastructure and reducing CAPEX. 

• PFAS-Free, Sustainable  Cooling

By utilizing pure, non-toxic DI water as its cooling medium instead of chemical blends, W2PC eliminates PFAS-related environmental hazards, supporting a truly sustainable AI computing ecosystem. 

Together, these innovations  help AI infrastructure achieve higher computing densities with greater energy efficiency and long-term reliability.

How W2PC Empowers  Next-Generation AI Computing

W2PC supports the cooling requirements of modern AI servers and GPUs in several ways, including:

• Preventing Thermal Throttling and Maximizing AI Performance

Stable thermal management helps AI processors maintain consistent performance during demanding training and inference workloads. By efficiently removing heat from high-power GPUs and CPUs, W2PC helps prevent thermal throttling and supports AI servers in maintaining peak clock speeds during continuous, 24/7 operation.

• Optimizing AI Infrastructure  Efficiency

W2PC’s chiller-less warm-water design reduces the energy and infrastructure required to remove heat from AI servers. This helps data centers achieve a lower power usage effectiveness (PUE), reduce total cost of ownership (TCO), and support more sustainable AI operations.

• Simplifying Deployment

W2PC features a modular, plug-and-play architecture that reduces deployment complexity and infrastructure requirements. The system supports both new AI infrastructure projects and upgrades to existing facilities, helping organizations adopt advanced liquid cooling with less disruption.

• Providing a Scalable AI Cooling Ecosystem

AI deployment can range from single-rack enterprise systems to hyperscale data centers, with each scale requiring different cooling capacities and infrastructure. CreAlien offers three W2PC solutions to support organizations as their AI infrastructure expands.

• VC-Lite: Designed for SMEs, single-rack AI Deployments, and lightweight LLM applications.
• VC-Pro: Supports enterprise AI, smart manufacturing, private AI clouds, and existing data center retrofits.
• VC-Mega: Built for hyperscale data centers and cloud service providers (CSPs) requiring megawatt-level cooling capacity.

W2PC Validation and Deployment Applications

W2PC has undergone proof-of-concept (POC) testing and is being introduced across different AI computing environments, from research facilities to enterprise and industry-specific applications.

• Tier 1 Server System Manufacturer – Proof-of-Concept Validation

Completed proof-of-concept (POC) validation with a Tier 1 server system manufacturer, demonstrating the feasibility of W2PC for next-generation AI server cooling.

• National Cheng Kung University – Green High Efficiency Computing Demonstration Site

VC-Lite is planned for deployment at National Cheng Kung University's green high-efficiency computing demonstration site, supporting energy-efficient AI computing.

• Enterprise AI Applications – Scalable AI Infrastructure

VC-Lite and VC-Pro are designed for seamless deployment in enterprise AI server environments, supporting both new installations and existing data center upgrades.

• Industry-Specific AI and Thermal Engineering Collaboration

CreAlien’s industry experience includes collaboration with a European thermal engineering firm on advanced thermal management. W2PC also supports specialized AI applications in smart manufacturing and medical imaging, including work with a Taiwan-based medical imaging company

Whether you are deploying enterprise AI servers or planning a hyperscale AI factory, CreAlien delivers a liquid cooling solution designed around your computing density, cooling capacity, and facility requirements. Contact us to discuss the right GPU liquid cooling solution for your next AI server deployment.