Much of the industry's discussion has centered on whether liquid cooling will replace air cooling. While that debate captures headlines, it might not always capture the reality facing many current projects.
For the foreseeable future, most data centers will not transition entirely from one cooling technology to another. Instead, they will evolve toward hybrid cooling architectures that strategically combine air and liquid cooling solutions based on your climate’s demands. This can include specific workload requirements, rack densities, and long-term operational goals.
The objective is no longer selecting a single cooling technology. Designing flexible infrastructure that can accommodate and future growth without requiring wholesale replacement of existing systems has become essential.
This changes the conversation for owners, engineers, operators, and technicians. As industry experts continue to explore and test emerging technologies in real environments, discussions will center around liquid cooling while also considering how to successfully design for, integrate, and maintain both liquid and air solutions. Many will benefit from evaluating how to bridge the gap between legacy data center air cooling systems and high watt density GPUs.
AI Is Changing the Cooling Conversation
While kilowatts per rack is a key component, power density tells only part of the story. AI workloads also introduce different thermal behaviors than traditional enterprise computing. Large language model inference, machine learning training and other GPU-intensive applications generate heat differently depending on how computing resources are utilized. Some workloads create relatively stable thermal loads, while others experience rapid fluctuations that require cooling systems capable of responding quickly to changing operating conditions.
This means thermal management is becoming increasingly dynamic. Designing for peak capacity alone is no longer sufficient. Cooling systems must also provide operational flexibility, responsive controls, and the ability to support a broader range of workload profiles over the life of the facility.
At the same time, hardware development cycles continue to compress. Processor manufacturers are introducing new generations of AI accelerators at an unprecedented pace, making it difficult to predict future cooling requirements with confidence. Infrastructure designed around today's hardware may be expected to support substantially different thermal loads only a few years later. This uncertainty presents one of today's greatest design challenges.
For engineers, that creates uncertainty about design conditions and adds to design considerations. For owners, it challenges investment decisions and partnership dynamics. For project managers the industrialization of data centers changes timelines and stakeholder engagement. And for operators and technicians, it introduces new equipment, controls, fluids, maintenance requirements, and operating procedures, including experts, into environments where criticality, safety, and reliability remain paramount.
While we can’t predict exactly what future hardware will require, and how the markets will respond, we can create enough flexibility to adapt as requirements become clearer. That shift, from optimizing for today's equipment to planning for tomorrow's flexibility, is becoming a defining characteristic of AI-ready infrastructure.
Air + Liquid Cooling Together
Air cooling has reliably supported data centers for decades and will continue to play a role. Modern airside systems can support substantial loads, manage residual heat, serve mixed-density environments, and maintain appropriate conditions throughout the data hall.
But as heat becomes increasingly concentrated within high-density computing equipment, removing that heat can become more challenging.
Liquid provides another heat-transfer mechanism. Because of its physical properties, liquid can transport significantly more heat in a smaller volume than air. Direct-to-chip approaches can also capture heat closer to where it is generated rather than relying entirely on heat transfer to the surrounding air.
That doesn't make air obsolete. It creates an opportunity to use each technology where it makes sense.
A high-density AI deployment might use liquid to remove a significant portion of the heat generated by processors while still relying on air cooling for residual heat and other components. Elsewhere in the same facility, lower-density racks may remain entirely air cooled.
That is where hybrid cooling becomes less about choosing technologies and more about developing a strategy.
Build on What You Already Have
For existing data centers in particular, hybrid cooling can provide a path forward without starting over.
Many organizations introducing AI capacity are working within facilities designed long before today's high-density workloads were anticipated. Existing chilled water systems, air cooling equipment, controls, piping, and other infrastructure represent significant investments.
Instead of assuming these systems need to be replaced, project teams can evaluate their unique environment with what can be retained, what may need to be expanded, and where new cooling technologies can be strategically integrated.
This also changes how teams should think about future readiness.
Being "AI-ready" means leaving room for tomorrow: mechanical space for future equipment, appropriately planned piping pathways, connection points, control capabilities, and infrastructure that can accommodate changing flow or capacity requirements.
In many cases, the goal is to make tomorrow's upgrade an integration project rather than a complete redesign.
Plan for the People Who Will Operate It
Beyond design, a successful hybrid strategy also needs to account for how the facility will be operated and maintained.
Introducing liquids closer to IT equipment creates new considerations for monitoring, leak detection, fluid quality, filtration, pressure, flow, controls, and maintenance. Air and liquid systems also need to operate as parts of the same thermal ecosystem rather than as independent pieces of equipment.
For operators, visibility becomes increasingly important. Monitoring temperature, flow, pressure, and other operating conditions can help teams understand system performance and available capacity as workloads change.
Serviceability should also be considered during design. Where will technicians access equipment? Can components be isolated and maintained without disrupting critical loads? Are there clear procedures for servicing both air and liquid systems? Do facility teams have the training needed to work confidently around new technologies?
Those questions are easier and less expensive to address during planning than after equipment is installed.
Start Small, Learn, and Scale
Not every organization or facility needs to make the transition at the same pace.
For some, a dedicated AI pod or pilot deployment may be the best starting point. Smaller deployments allow teams to validate cooling concepts, understand controls, establish maintenance procedures, train personnel, and identify potential challenges before expanding the architecture across a larger environment.
This is also why collaboration matters. Owners, engineers, operators, service teams, IT stakeholders, and cooling specialists each see different parts of the challenge. Bringing those perspectives together early can help ensure the final design works not only on paper, but throughout its operational life.
The Future Is Flexible
AI is changing data center cooling, but the industry's evolution is more nuanced than a wholesale transition from air to liquid.
Air cooling will remain essential across many applications. Liquid cooling will play a growing role as higher-density computing creates new thermal demands. And increasingly, the two will need to coexist within the same facility.
The key takeaway is to work with a trusted partner whose mission is to provide the right cooling technologies for your climate where they make sense while remaining reliable, maintainable, and adaptable to whatever comes next.