As artificial intelligence (AI) and high-performance computing (HPC) push more compute capacity into the data center, much of the infrastructure conversation has focused on one metric: rack density.
But rising density is only part of the story.
AI is changing the equipment inside the rack, the amount and type of power it requires, how heat is removed, and how supporting infrastructure is integrated. As these requirements evolve together, the physical architecture of the rack must evolve with them.
Much of the AI infrastructure conversation happens at the facility level—utility capacity, power distribution, cooling capacity, and available space. But the rack is where those requirements ultimately converge. It is where power reaches the IT load, heat must be removed, connectivity is managed, and equipment must remain accessible and serviceable.
Increasingly, the rack is becoming more than a structure for mounting IT equipment. It is becoming an integrated infrastructure platform where compute, power, cooling, cabling, and serviceability must work together.
AI Is Putting New Demands on Rack Architecture
Traditional data center racks were designed around relatively predictable equipment dimensions, power requirements, and cooling strategies. AI infrastructure challenges many of those assumptions.
GPU-based systems and other high-performance computing platforms can concentrate substantially more compute—and therefore more power and heat—within a single rack. At the same time, equipment may be larger and heavier, power distribution more complex, and liquid cooling increasingly integrated alongside air cooling.
These changes create a more interconnected set of physical infrastructure requirements. Available facility capacity does not automatically translate into deployable rack capacity. Power, cooling, space, cabling, and structural requirements must all come together at the point where the IT equipment is installed.
At higher densities, cabinet design decisions also become increasingly interdependent. Cable pathways can affect airflow and service access. Cooling infrastructure can affect usable rack space. Power architecture can influence equipment placement and cabling. Even rack dimensions and structural requirements may need to change to accommodate larger, heavier compute platforms.
Rather than addressing these elements independently, organizations increasingly need to consider how they work together at the rack level.
Power Is Becoming a Rack-Level Design Decision
As compute density increases, so does the importance of how power is delivered to and distributed within the rack.
In conventional data center architectures, power distribution and rack selection could often be treated as closely related but distinct infrastructure decisions. High-density AI environments are making that separation more difficult.
Open Rack environments, for example, may use 48V DC power architectures with busbars and power shelves integrated directly into the rack. This changes the physical relationship between power distribution, IT equipment, cabling, and available rack space.
The larger point is not that one power architecture is appropriate for every AI deployment. Rather, the power architecture for high-density AI racks increasingly influences the rack architecture required to support it.
As power requirements rise, selecting the rack before understanding how power will be delivered, distributed, and managed can limit flexibility later in the design process.
Read also: How to Choose a High-Density PDU for AI Racks
Cooling Can No Longer Be Separated from Rack Design
The same principle applies to cooling.
Higher-density AI systems are accelerating the adoption of direct-to-chip liquid cooling, rear door heat exchangers, and hybrid approaches that combine liquid and air cooling.Even in predominantly liquid-cooled environments, some components may continue to reject heat to the surrounding air.
That means rack architecture must account for more than simply moving sufficient airflow through IT equipment.
Designers may need to consider how cooling connections enter the rack, where supporting components are located, how residual air-cooled loads are managed, and whether technicians can safely access and service equipment once the complete system is installed.
The question is therefore shifting from “How do we cool this rack?” to “How does the rack architecture support the cooling strategy?”
Considering cooling early in rack design can help avoid conflicts among thermal infrastructure, power components, cabling, and service access as density increases.
Physical Form Factors Are Evolving With the IT
AI is also contributing to changes in the physical form factor of rack-mounted infrastructure.
AI compute platforms can be larger, deeper, and heavier than conventional IT equipment, making usable rack depth, equipment clearances, and structural capacity increasingly important design considerations. Supporting components for power distribution, cooling, and cabling can add further weight and consume additional space within the rack. As a result, physical rack specifications need to account for the complete installed system—not simply the dimensions of the servers themselves.
The traditional 19-inch EIA rack remains an important architecture for enterprise, hyperscale, HPC, and other data center applications. At the same time, some AI and high-density computing platforms are driving the need to support alternative rack architectures.
The Open Rack Specification introduced a 21-inch equipment space and OpenU (OU) mounting standard, along with an architecture designed to integrate IT equipment with supporting power infrastructure. For organizations deploying IT platforms built around these standards, the rack must align with those requirements from the beginning.
This does not mean Open Rack architectures will replace traditional EIA racks. Different workloads, equipment platforms, and data center environments will continue to require different approaches.
Instead, it points to a broader shift: rack architecture is becoming more closely aligned with the architecture of the compute it supports.
For infrastructure teams, that makes understanding the IT roadmap increasingly important when planning the physical environment around it.
Standardization and Customization Have to Coexist
As AI infrastructure expands, organizations face another challenge: deploying increasingly complex systems repeatedly at scale.
Standardized rack architectures can help establish a common foundation for equipment integration and deployment. Open Rack specifications, for example, provide common dimensional and interface requirements that manufacturers and technology providers can design around.
But a standardized architecture does not make every deployment identical. Facilities still differ. Compute configurations differ. Power availability, cooling strategies, cable pathways, deployment schedules, and operational requirements can vary from one site—or even one deployment phase—to another.
That makes standardization and customization complementary rather than opposing goals.
A common architecture can create repeatability, while engineering flexibility allows the physical infrastructure to be adapted around the specific requirements of the deployment. As rack density and infrastructure complexity increase, striking that balance becomes increasingly important—particularly for organizations evaluating Open Rack architectures.
For a deeper look at the physical infrastructure considerations involved in Open Rack deployments, read our white paper on Building for OCP: Physical Infrastructure Considerations for Open Rack Deployments.
The Rack Is Becoming Part of the AI Infrastructure Strategy
For years, the rack could sometimes be treated as a relatively straightforward physical layer decision made after major IT and facility requirements had already been established.
AI infrastructure is making that approach harder to sustain.
Power, cooling, compute, cabling, and physical form factor are becoming increasingly interdependent. Decisions in one area can create constraints—or opportunities—in another. The rack is where these requirements ultimately converge, making it an increasingly important part of the overall AI infrastructure strategy.
The right architecture is not defined by density alone. It is defined by how effectively the physical infrastructure brings compute, power, cooling, connectivity, and serviceability together around the requirements of the deployment.
As those requirements continue to evolve, organizations will need rack platforms capable of evolving with them.
Build for Open Rack V3 With CPI
Chatsworth Products (CPI) brings its infrastructure engineering and manufacturing expertise to Open Rack deployments with CPI’s OCP Open Rack V3 (ORV3).
Designed to meet Open Rack V3 specifications, CPI’s ORV3 is available in standard and customer-specific 44OU and 48OU configurations, as well as high-power rack (HPR) configurations. The platform accommodates OCP-standard 48V DC busbars, power shelves, and IT equipment, with supporting infrastructure options to help customers integrate the rack around their specific deployment requirements.
Backed by CPI engineering expertise and U.S.-based manufacturing, CPI’s ORV3 provides a customizable platform for organizations deploying AI, HPC, and other high-density computing environments.