Artificial intelligence is transforming how data centers are designed and built. Over the past few years, the infrastructure required to support AI workloads has shifted dramatically. Power density is rising, cooling demands are increasing, and traditional facility designs are being pushed beyond their limits.
For companies across the infrastructure ecosystem, including Data Center Floor Tiles (DCFT), these changes are creating new challenges but also major opportunities. As AI systems grow larger and more power-intensive, data center operators are increasingly turning to hybrid cooling strategies that combine air and liquid cooling.
Understanding why this shift is happening can help organizations prepare their facilities for the next generation of AI infrastructure.
AI Is Driving a Massive Increase in Rack Power Density
One of the biggest changes in modern data centers is the rapid increase in rack power density.
Just a few years ago, the average data center rack consumed around 20 kW of power. Today, AI training clusters are pushing racks toward 200 kW, and industry forecasts suggest densities could reach 400–800 kW per rack within the next two years.
This growth has happened incredibly quickly. In fact, the ramp from 20 kW to 200 kW racks has occurred in less than two years.
At the same time, data center investment has exploded. U.S. data center capital expenditures have grown from roughly $15 billion four years ago to an estimated $500 billion today, largely driven by AI infrastructure demand.
These workloads require far more compute power, which means facilities must handle dramatically higher heat loads than traditional enterprise environments.
Why Cooling Strategies Are Evolving
For decades, most data centers relied almost entirely on air cooling to regulate temperatures across the data hall.
AI hardware is changing that equation.
Modern GPU servers generate intense heat at the chip level. Liquid cooling has therefore become an effective method for removing heat directly from high-performance components. However, air cooling still plays a critical role in maintaining environmental stability throughout the room.
Because of this, many modern facilities are adopting hybrid cooling architectures that combine both approaches.
In a hybrid system:
- Liquid cooling removes heat directly from GPUs and CPUs
- Air cooling manages the surrounding environment and residual heat
- Airflow management maintains stable operating conditions across the data hall
This combination allows facilities to capture heat more efficiently while maintaining flexibility as hardware densities evolve.
Physical Cooling Limits Are Becoming Clear
Cooling capacity is not just about technology. Physical space also becomes a constraint at large scales.
Rooftop cooling equipment, for example, typically requires 12–15 square feet of roof space per megawatt of cooling capacity. When facilities begin operating at tens of megawatts of compute capacity, rooftop space alone cannot always support the required cooling infrastructure.
Consider a high-density data hall producing 50 MW of heat within a 30,000-square-foot room. The roof area required for cooling equipment can quickly exceed what the building can support.
Hybrid cooling strategies help solve this problem by capturing heat directly at the source while distributing cooling loads across multiple systems.
AI Clusters Create Sudden Heat Surges
Another challenge is how AI workloads behave.
Large GPU clusters often run tightly synchronized training jobs. When these workloads begin, compute demand can increase almost instantly. Some facilities see power load changes of 80–100 MW within seconds.
Mechanical cooling plants cannot ramp that quickly.
These sudden load increases can create localized hot spots across rows of racks. Computational fluid dynamics models often show these events as concentrated heat plumes forming in the data hall.
Hybrid cooling systems help manage these thermal spikes. Liquid cooling removes heat directly from the processors while air systems continue to stabilize the surrounding environment.
Data Center Commissioning Is Changing
The rapid pace of AI hardware deployment is also changing how data centers are commissioned.
Traditionally, infrastructure systems were fully tested before servers were installed. Today, IT deployments are moving so quickly that infrastructure and hardware often come online simultaneously.
Many AI facilities now commission their systems in phases, powering up equipment in stages and testing performance using simulated workloads.
Hybrid cooling architectures support this process by providing operational flexibility while rack densities evolve during deployment.
Hardware Supply Chains Are Under Pressure
The rapid expansion of AI computing has also strained global supply chains.
Across the industry:
- Major hyperscale providers have reserved server supply through 2027
- Memory now represents a large portion of server build costs
- Flash shortages have even triggered renewed demand for traditional hard drives
Meanwhile, advanced GPU platforms developed by companies like NVIDIA are pushing the limits of interconnect performance and compute density.
Software ecosystems remain critical to adoption as well. Widely used frameworks such as TensorFlow, PyTorch, and scikit-learn continue to shape which hardware platforms gain traction in AI environments.
AI Data Centers Are Becoming Compute Factories
Perhaps the biggest change is how these facilities are viewed.
Traditional data centers were often treated as buildings designed to house servers. AI facilities are increasingly operating like industrial production environments.
Instead of producing goods, these facilities produce compute capacity and AI inference workloads. Every system inside the building must support that output as efficiently as possible.
That includes power delivery, airflow management, cooling infrastructure, and structural components within the data hall.
Infrastructure Must Evolve With AI
As rack densities continue to rise and AI workloads become more demanding, data center infrastructure must evolve alongside them.
Hybrid cooling systems, flexible facility designs, and scalable infrastructure will be essential for supporting the next generation of high-performance computing environments.
For infrastructure providers like DCFT, the future of AI data centers will depend on building environments that support efficient airflow, structural reliability, and adaptable cooling strategies.
Facilities that can balance both air and liquid cooling technologies will be better positioned to support the rapidly growing demand for AI compute.
Contact DCFT to learn more about cooling your AI data center