News & Insights

Data center operators often see liquid cooling as just an equipment choice, but it’s actually a comprehensive facilities design issue. What begins as a simple reaction to rack densities over 200 kilowatts can escalate into changes in power systems, space layout adjustments, and infrastructure interdependencies that many don’t expect until constraints become apparent during construction or operation.

The trend is clear: power levels per rack have increased from over 100 kilowatts to over 200 kilowatts. At these densities, liquid cooling becomes essential. However, it’s less obvious how this equipment-level choice impacts entire facilities. Data centers rely on a balance of power, space, and cooling—like a three-legged stool. Altering one aspect without considering the others isn’t feasible. Operators who view liquid cooling solely as a mechanical system often face infrastructure challenges later—during construction, after project completion, or even when trying to fulfill customer demands in operational facilities.

The technology shift enabling change

Recent developments have accelerated the adoption of liquid cooling and altered its infrastructure needs. At CES 2026, NVIDIA unveiled the Rubin platform, which is designed to operate with warm-water supply loops around 45°C (113°F). This change is about more than just chip temperature limits. It significantly impacts how facilities are constructed.

Newer chips operating at higher temperatures reduce the need for mechanical cooling and chillers. Facilities can use direct air-to-water heat rejection instead of compressor-heavy chilled water systems. This change significantly reduces power consumption and improves power usage effectiveness, allowing more power to go to IT equipment while placing less stress on mechanical systems.

The OEM ecosystem is aligning around this shift. Manufacturers are expanding liquid-cooled rack solutions optimized for higher-temperature platforms. This industry support reduces perceived risk for operators, insurers, and lenders about whether these cooling methods will prove reliable and properly supported over facility lifespans.

For owners and operators, the practical and financial implications are significant. Silicon roadmaps now align with liquid-cooling architectures that support warmer supply loops and potentially eliminate chillers entirely. This opens the door to shifting from capital-intensive mechanical cooling plants to simpler, more scalable thermal designs. The operational energy footprint decreases, and the complexity of commissioning and maintaining large-scale cooling plants is reduced, especially in high-density AI deployments.

The hybrid reality

The first infrastructure reality operators must accept is that facilities will never be 100 percent liquid-cooled or 100 percent air-cooled. The actual split typically runs 80/20 or 70/30. Network equipment still requires air cooling. High-density compute demands liquid cooling. Each has different infrastructure requirements.

This hybrid reality influences early design choices that shape long-term flexibility. It’s essential to consider both cooling types from the beginning because infrastructure needs differ. Liquid-cooling loops follow different distribution routes than air-handling systems. Physical space allocations vary. Power requirements change. These are core decisions that either restrict or support everything that comes afterward.

The challenge intensifies in existing facilities where space becomes the biggest constraint. The question becomes how to fit all the required infrastructure into existing space. Many facilities simply can’t accommodate hybrid infrastructure without major modifications that cost nearly as much as a new building.

Multi-tier thermal architectures

High-temperature liquid cooling creates opportunities to rethink facility thermal design from the ground up. Rather than forcing all equipment into a single temperature regime, multi-tier architectures layer cooling loops to match diverse thermal characteristics of modern compute loads.

A standard design features a primary high-temperature liquid loop built to supply temperatures around 105°F, effectively capturing heat from liquid-cooled racks. Secondary and tertiary loops function at sequentially lower temperatures to accommodate equipment with stricter thermal requirements. This layered approach allows for mixed rack populations surpassing 200 kilowatts per rack, all while maintaining optimal thermal performance and energy efficiency.

Advanced controls dynamically blend supply temperatures and route flows based on real-time workload intensity and rack density. High-heat sections operate from warm loops when possible, while more sensitive equipment receives cooler fluid as needed. The result is a cooling system responding proportionally to actual workload instead of running worst-case assumptions continuously.

By operating with warmer supply loops, facilities reduce or eliminate mechanical chilling, expand hours of economizer operation, and materially lower parasitic electrical load per megawatt of IT. For owners, this means reduced operating expense volatility tied to compressor energy, improved power usage effectiveness at scale, and more stable performance under extreme ambient conditions.

However, there is good reason for technical caution, since there is a critical point where sufficient cooling capacity is essential to handle short break scenarios. High-density racks generate significant heat, transferring it into small volumes of liquid even at elevated levels temperatures.

Full hydronic modeling of mechanical systems helps minimize thermal storage requirements while predicting critical temperatures during failure scenarios with much higher accuracy than traditional approaches.

Power system implications

Liquid cooling fundamentally changes facility power requirements in ways that catch operators unprepared. The most significant shift is that backup power must now support cooling systems, not simply IT loads. Weak points are unacceptable in cooling systems supporting high-density compute. Since GPUs are expensive, and a single rack failure can cost millions of dollars, reliable cooling during power events becomes non-negotiable.

This drives changes throughout electrical infrastructure. Power distribution must be flexible enough to accommodate dynamic loads. Medium voltage is becoming standard for large facilities to handle the scale efficiently. Overhead busway allows flexible power delivery as rack configurations evolve. Single points of voltage transformation increase efficiency and reduce losses, which matters enormously when every percentage point of efficiency translates to available IT capacity.

Design for adaptability

When designing for flexibility and future-proofing, the goal is expandability and modularity. Future additions of capacity or redundancy should happen without impacting existing operations. This means deliberately leaving space for infrastructure that isn’t installed on day one but might be needed as density evolves or customer requirements change.

The constraint goes beyond square footage to operational accessibility. Maintenance teams need realistic access to equipment over facility lifespans. Installation-ready, constructible, clash-free designs account for operational reality from the beginning.

This level of integrated thinking across disciplines separates facilities that adapt gracefully from those that hit constraints requiring expensive retrofits. The impact shows in how different trades and disciplines interconnect:

• Mechanical decisions affect electrical capacity
• Structural loads change with equipment density
• Controls integration spans all systems.

Approaching any element in isolation guarantees problems downstream.

When complexity may be worthwhile

Multi-tier liquid cooling systems are more complex to engineer than single-loop or traditional chilled-water plants. They require integrated modeling, careful hydraulic design, and advanced control strategies. This complexity gets absorbed during engineering and commissioning, not in daily operations. Operators often experience fewer emergency cooling events, more predictable performance envelopes, and reduced manual intervention because systems are designed to adapt automatically.

The question becomes whether that engineering investment delivers value for specific facilities and business models. Not all IT equipment is suitable for the elevated coolant temperatures typical of high-efficiency liquid cooling systems. Legacy servers, certain storage arrays, or non-optimized components have strict thermal specifications that limit compatibility, risking performance throttling, reduced reliability, or warranty issues. Careful evaluation of each technology’s specifications is essential before deployment.

Facilities with suitable hardware running sustained rack densities above 100 kW and electricity rates at or above $0.10/kWh generally reach break-even on more complex liquid cooling system investments within 18 to 24 months. Through the lifecycle, this architecture can deliver between 10 and 30 percent lower cost of operation through dramatically reduced cooling energy, increased compute per square foot, and improved power usage effectiveness.

For existing facilities, space constraints often make the decision. Getting hybrid infrastructure into buildings designed for air cooling alone requires creative solutions that may or may not be economically viable. For new construction, the calculation weighs upfront engineering complexity against decades of operational efficiency and flexibility to accommodate density growth without plant redesigns.

How can Salas O’Brien help?

Salas O’Brien brings over 30 years of technical expertise to the mission critical market. Our approach starts with listening and understanding what different customers’ business models require, which informs design decisions throughout the project. Whether you’re maximizing compute capacity per site, working within physical constraints, or addressing environmental concerns, we translate your objectives into building designs that work.

Our capabilities span full facility. We conduct full hydronic modeling to optimize thermal storage and predict system performance during failure scenarios. We coordinate integrated design across mechanical, electrical, and structural disciplines because we understand the three-legged stool—changing one element requires adjusting the others. We model maintenance access during design to deliver installations that remain serviceable throughout their operational lives.

We go the extra mile as part of standard design process, not as an added service. You receive installation-ready, constructible, clash-free designs that account for both technical requirements and operational reality. As your facility evolves and density requirements change, the flexibility we build in during design phase protects your long-term investment.

To discuss how liquid cooling decisions will impact your facility infrastructure and how we can help you design for both immediate needs and long-term adaptability, contact one of our data center experts below, or email us at [email protected]

For media inquiries on this article, reach out to [email protected].