As artificial intelligence (AI) transforms industries—from personalized healthcare to self-driving cars—it’s placing new demands on data center infrastructure. AI workloads require massive amounts of power, compute, and cooling, especially as organizations deploy high-density GPU clusters.

But legacy data centers are hitting their limits, and fast. Meeting these demands requires a fundamental shift in how data centers manage power and heat.

Cooling plays a bigger role in data center operations than many realize. According to the International Energy Agency (IEA), cooling accounts for around 40% of a data center’s total energy use—almost equal to the energy used by the IT equipment itself. That percentage can climb even higher in facilities running high-density workloads.

But it’s not just about efficiency—it's about uptime. Overheating risks hardware failure, degraded performance, or outages, which can cost thousands—or even millions—per hour in lost productivity or customer impact

Traditional data centers rely on air cooling: circulating chilled air through server rooms using fans and HVAC systems. While effective for standard enterprise workloads, this approach begins to break down when used with dense, high-power GPU deployments.

As densities rise, air becomes a less efficient medium for heat transfer. More power is required to keep air cool and moving, which increases operating costs. Air-cooled systems typically top out at 15–20 kW per rack, while advanced row-based cooling with containment can stretch to 30 kW—but even that’s reaching the limit.

And in hotter climates, like Texas or Arizona, the cost and complexity of keeping air-cooled facilities within safe temperature thresholds grows even more dramatically. Simply put, blowing more cold air into a room isn’t a scalable or sustainable solution for AI infrastructure

To meet the demands of AI workloads, the industry is moving toward liquid cooling. Unlike air, liquids can absorb and transfer heat much more efficiently, making them better suited for dense, high-power environments. Liquid cooling supports rack densities of 20 kilowatts and up.

There are a few key approaches:

Each method helps reduce energy use, improve performance, and extend equipment life—all while enabling the kind of rack densities that AI requires. Many companies are leaning into hybrid cooling strategies, which combine one or more of these techniques, potentially with air cooling, for maximum data center cooling efficiency

While IDC reports that 22% of data centers already use some form of liquid cooling, most legacy facilities weren’t built to support it. Traditional data centers were designed around air cooling, with lower rack densities, limited chilled water infrastructure, and HVAC systems that cap out around 20–30 kW per rack. As AI workloads push densities higher, operators are hitting physical and thermal limits. A 2024 Uptime Institute report found nearly half of operators cite cooling constraints as a key challenge when deploying high-density compute, especially for GPU-intensive environments.

Retrofitting to support liquid cooling isn’t always practical—or affordable. Upgrading a 1 MW data center to accommodate liquid-cooled racks can cost millions, especially if it involves structural modifications, new plumbing, and power upgrades. In water-stressed regions, evaporative cooling isn’t a sustainable option, further limiting retrofit feasibility. While some sites can be upgraded with hybrid solutions, many operators are opting to deploy AI infrastructure in purpose-built environments that are liquid cooling-ready from day one.

At Evocative, we’re designing infrastructure with AI workloads in mind—not retrofitting for them after the fact. Our modern data centers are engineered to support high-density, liquid-cooled environments, giving customers the flexibility to scale without compromising performance or uptime. Whether you're training large models or supporting inference at scale, we provide the power, space, and thermal design needed to handle next-gen workloads.

Our facilities leverage closed-loop liquid cooling systems that eliminate evaporation, significantly reducing water usage. In water-stressed regions, we deploy air and refrigerant-based systems instead of evaporative cooling to minimize environmental impact. And with our modular architecture, we can achieve up to 30% reductions in Power Usage Effectiveness (PUE)—translating into energy savings and lower operating costs for our clients. It’s all part of our strategy to future-proof infrastructure for the AI era.

Liquid cooling is no longer experimental—it’s becoming the new standard for high-performance computing. If your organization is leaning into AI, now is the time to evaluate whether your cooling strategy is ready for what’s next. Contact Us to discuss your data center cooling needs.