Beyond the limits of air: Why liquid cooling is becoming a strategic imperative for AI

In the rapidly evolving information technology landscape, the data center of the future has arrived ahead of schedule. As AI workloads move to become core business drivers, the underlying infrastructure is hitting a thermal wall. 

For CIOs, the transition from traditional air cooling to liquid cooling is no longer a niche technical choice for supercomputers—it is a critical financial and operational investment in long-term enterprise viability.

Working together, HPE and NVIDIA are redefining what’s possible, but realizing the full potential of these advancements requires a fundamental shift in how we cool the infrastructure that powers them.

The physics of performance: Why air is no longer enough

For decades, air cooling has been the workhorse of the data center. However, the math is changing. Standard CPUs and early-generation GPUs could be kept within operational limits by moving massive volumes of air through the rack. But with the advent of next-generation GPUs like the NVIDIA Blackwell series, power densities are skyrocketing.

• The power density gap: A fully populated modern GPU rack can draw upwards of 132 kW, with next-generation architectures projected to reach 240 kW per rack—and eventually 1 MW per rack.
• Thermal limits: As chip density rises, the thermal tolerance of these processors is actually decreasing. Air simply does not have the thermal conductivity required to pull heat away fast enough to prevent thermal throttling.
• The cooling-to-compute ratio: Traditional facilities can spend up to 40% of their total energy budget just on cooling. Without efficient architecture, organizations are forced to sacrifice compute capacity to stay within power and thermal limits.

Liquid cooling as a strategic financial asset

When evaluating liquid cooling, the conversation must move from “how much does it cost?” to “what does it enable?” Investing in liquid-cooled infrastructure today provides a multi-year competitive advantage.

1. Superior ROI and TCO

While the upfront capital expenditure (CapEx) for liquid-cooled systems can be higher than for air-cooled systems, the long-term total cost of ownership (TCO) is significantly lower.

• Energy efficiency: Liquid cooling can reduce energy consumption by up to 40%. For a 1 MW data center, this translates to hundreds of thousands of dollars in annual OpEx savings.
• Payback period: Typical breakeven points for liquid cooling deployments are now reaching the 18–24 month mark, with a 3-year TCO advantage of 15–25% over air-cooled alternatives.

2. Maximizing real estate and density

In many markets, the greatest constraint is not hardware, but physical floor space and power availability. Liquid cooling allows for significantly higher rack density, enabling CIOs to:

• Do more with less: Condense three or four air-cooled racks into a single liquid-cooled rack.
• Accelerate deployment: HPE modular solutions, such as the HPE AI Mod POD, allow for rapid scalability without the 3–5 year wait time for new building construction.
• Improve the energy profile of high-density AI deployments, especially when paired with facility designs that support heat reuse or lower mechanical cooling demand.

3. Sustainability as a business imperative

ESG targets are no longer optional feel-good metrics. They are core to corporate governance and regulatory compliance.

• PUE excellence: Liquid cooling can drive power usage effectiveness (PUE) down to 1.15 or lower, compared to the industry average of 1.6 for air-cooled facilities.
• Carbon footprint: HPE has demonstrated that liquid cooling can reduce a data center’s carbon footprint by as much as 87%.
• Heat recovery: Advanced systems allow for the reuse of heated water for local heating or other industrial processes, turning waste heat into a community asset.
• 100% fanless architecture: HPE introduced the industry’s first 100% fanless direct liquid cooling system, which eliminates the noise and mechanical failure points of traditional fans while maximizing energy efficiency.
• Flexible deployment models: Whether through retrofitting existing facilities or leveraging HPE Financial Services for reduced payment plans, CIOs have multiple paths to adoption.

Summary checklist for the future-ready CIO

Strategic driver	Practical impact
Performance resilience	Prevent thermal throttling to maintain consistent AI training and inference service-level objectives.
Financial optimization	Lower annual OpEx by up to 40% through energy-efficient cooling.
Sustainability	Meet aggressive ESG goals with PUEs nearing 1.10.
Scalability	Support rack densities of 70 kW to 100 kW+ required for next-gen NVIDIA Blackwell GPUs.

The choice to move to liquid cooling isn’t just about managing heat—it’s about removing the bottlenecks that can stall your AI strategy. By investing in liquid cooling today with partners like HPE and NVIDIA, you aren’t just buying hardware; you are buying the agility, efficiency, and performance required to lead in the era of AI. Explore HPE’s rack-scale AI solutions and flexible financing options to upgrade your facility.

**************

As AI becomes increasingly central to economic competitiveness, scientific advancement, and national priorities, organizations require infrastructure that balances performance with security and sovereign control. Together, HPE and NVIDIA co-engineer rack-scale AI systems that integrate AI computing, high-performance networking, and supercomputing expertise to support large-scale AI workloads. This provides enterprises, governments, and research institutions with a trusted foundation for sovereign AI initiatives while maintaining control over critical data, models, and operations.