Enterprise data centers will need to evolve toward supercomputing
AI workloads are driving a new era of efficiency across infrastructure, power, cooling, and operations
The Data Center Perfect Storm
The forces reshaping modern data center infrastructure.
Enterprise data centers are entering a period of change that many organizations are not fully prepared for. The rapid adoption of artificial intelligence is driving a dramatic increase in computing demand while pushing power consumption, heat density, and infrastructure costs to levels that traditional enterprise environments were never designed to handle. These forces are converging to create what we call The Data Center Perfect Storm.
Data centers have always adapted to rising demand for computing. What makes this moment different is the nature of the workloads driving that demand. Artificial intelligence and machine learning require levels of computational intensity, parallelism, and system performance that have far more in common with high-performance computing than with the transactional and database workloads that enterprise infrastructure was originally built to support.
In short, many data centers are beginning to encounter the kinds of constraints that high-performance computing environments have been dealing with for decades.
The challenges created by the Data Center Perfect Storm are not entirely new. High Performance Computing environments have been operating under similar constraints for decades—extreme computational density, demanding power requirements, aggressive cooling challenges, and relentless pressure to maximize efficiency.
What HPC organizations learned long ago is that solving these problems requires more than simply deploying faster processors or larger systems. Achieving large gains in capability and performance requires corresponding gains in efficiency across the entire computing environment—from infrastructure and cooling systems to system architecture, workload management, and utilization practices.
Enterprise data centers are now beginning to encounter similar pressures as modern processors and accelerators consume far more power and generate far more heat than previous generations. Rack power densities that once seemed extreme are becoming increasingly common, and the infrastructure required to support them is changing rapidly. We explore the specific technology trends and numbers behind these shifts in the sections that follow.
The Explosion of Compute Demand
The demand for computing has been rising steadily for decades, driven by economic growth, expanding digital services, and the increasing instrumentation of nearly every aspect of modern life. Businesses, governments, and research organizations now generate enormous amounts of data through sensors, applications, transactions, and connected devices. That data must be stored, analyzed, and increasingly used to drive automated decision-making.
In many ways, this growth is simply the continuation of long-term trends. As organizations become more dependent on digital technologies, the amount of computing required to support them naturally increases.
Artificial intelligence, however, is accelerating this growth dramatically. The rush to develop and deploy AI systems has become a strategic priority across nearly every sector of the economy. Organizations are experimenting with machine learning models, deploying generative AI tools, and exploring large language models in an effort to gain competitive advantage or improve operational efficiency.
What makes these workloads different is the computational intensity required to make them useful. Training AI models involves processing vast amounts of data through complex, multi-layered algorithms in order to discover patterns, relationships, or predictive insights. Even after models are trained, the process of generating predictions or responses—known as inference—can require substantial computing resources, particularly when large models must respond quickly to user requests.
While AI workloads can technically run on almost any computer, training and operating modern models within practical timeframes requires highly parallel systems equipped with accelerators such as GPUs, large memory pools, high-speed interconnects, and fast storage. These architectural characteristics closely resemble the systems used for decades in scientific computing and supercomputing environments.
In other words, the workloads now emerging across enterprise data centers increasingly resemble the kinds of computational problems long associated with high-performance computing.
We explore the infrastructure technologies and system architectures designed to support these workloads in the Systems & Integrators (LINK), System Components (LINK), and Performance Optimization (LINK) sections of this site.
The Power Problem
As computing demand increases, so does the amount of electricity required to support it. For many years, improvements in processor efficiency helped keep power consumption in check, even as computing capability increased. But those trends are changing as modern processors and accelerators push the limits of what current semiconductor technologies can deliver.
Today’s CPUs, GPUs, and other accelerators consume significantly more power than earlier generations of computing hardware. High-performance GPUs used for AI training and inference now operate at power levels that would have seemed extraordinary only a few years ago, and future generations are expected to push those limits even further. At the same time, other components—such as high-speed networking, large memory configurations, and fast storage—also contribute to rising power consumption within modern servers.
The result is that power density within data centers is increasing rapidly. Rack configurations that once operated comfortably within traditional enterprise power envelopes are being replaced by systems that require far more electrical capacity to operate effectively. In some environments, rack power levels that would have once been considered extreme are becoming increasingly common.
These trends have significant implications for the infrastructure that supports modern computing environments. Power distribution, facility design, and electrical capacity are becoming critical considerations in the deployment of AI and other high-performance workloads. The detailed technology trends and power requirements behind these changes are discussed in greater depth in the sections that follow.
One thing computers are exceptionally good at is converting electricity into heat. A lot of heat.
Why Air Cooling Has Reached Its Limits
For decades, most data centers relied almost entirely on air cooling to remove heat generated by computing systems. Servers pulled cool air from the data center environment, passed it across internal components, and exhausted the resulting warm air back into the room. Large air handling systems then removed that heat and maintained stable temperatures throughout the facility.
This approach worked well when server power densities were relatively modest. But as modern processors, accelerators, and high-performance networking components consume more electricity, the amount of heat they generate has increased dramatically. As a result, the thermal load produced by today’s high-performance systems is pushing the limits of what traditional air cooling can manage efficiently.
The physics behind this shift is straightforward. Liquid is far more effective than air at capturing and transporting heat. While air cooling requires conditioning the entire volume of a data center in order to keep equipment within safe operating temperatures, liquid cooling focuses directly on the components that actually generate the heat. By removing heat at or near the source, liquid-based systems can dramatically reduce the thermal load placed on the surrounding environment.
For this reason, many high-performance computing systems have relied on liquid cooling technologies for years. As enterprise data centers begin deploying increasingly dense computing infrastructure for AI and other demanding workloads, these technologies are rapidly becoming more relevant outside traditional supercomputing environments.
We explore and explain the different liquid cooling technologies, how they work, and the vendors developing them in the Cooling (LINK) section of this site.
The Cost Reality
Alongside rising power consumption and increasing heat density, organizations are also facing a significant increase in the cost of modern computing infrastructure. The systems required to support AI training and large-scale inference workloads are substantially more expensive than the enterprise servers that most data centers have traditionally deployed.
A major driver of these costs is the growing reliance on accelerators such as GPUs and other specialized processors. These devices deliver enormous computational capability, but they also represent a large share of the cost of modern AI systems. In many cases, the accelerators themselves cost more than the servers that host them.
At the same time, the scale of computing required for modern AI workloads can be dramatically larger than traditional enterprise applications. Training advanced models or supporting large inference deployments requires clusters of systems connected by high-speed networking, large memory configurations, and high-performance storage systems capable of moving vast amounts of data.
Some organizations assume that public cloud providers will allow them to sidestep these challenges by shifting the infrastructure burden elsewhere. In some situations this can be an effective short-term strategy, particularly for experimentation, development work, or handling occasional spikes in demand. But for sustained, high-utilization workloads, cloud-based computing can become significantly more expensive than operating infrastructure directly.
As a result, organizations are facing increasing pressure to ensure that the infrastructure they deploy is used as efficiently as possible. When systems are expensive, power consumption is rising, and demand for computing continues to grow, inefficiency quickly becomes costly.
This is one reason why high-performance computing environments have historically placed such a strong emphasis on utilization, workload scheduling, and system-level efficiency. Many of those same principles are becoming increasingly relevant as enterprise data centers adapt to the demands of AI-driven computing.
We explore these topics further in the Data Center Assessment (LINK), Utilization & Workload Management (LINK), and Performance Optimization (LINK) sections of this site.
What Supercomputing Already Learned
Many of the challenges now emerging in enterprise data centers are not new. High-performance computing environments have been operating under similar constraints for decades. Supercomputing centers routinely manage extremely dense computing systems, massive power requirements, complex cooling challenges, and constant pressure to maximize performance within limited budgets.
In these environments, simply adding more hardware has never been a sustainable solution. Supercomputing organizations learned long ago that achieving meaningful gains in capability requires improving efficiency across the entire computing environment. Infrastructure design, cooling systems, system architecture, interconnect performance, workload scheduling, and overall system utilization all play critical roles in determining how much useful computing work can be delivered.
As AI workloads spread across enterprise IT, many organizations are beginning to encounter the same physical and economic constraints that supercomputing centers have been managing for years. Power availability, cooling capacity, and infrastructure costs are becoming key factors that influence how computing systems are designed and deployed.
The technologies and operational practices developed within the HPC community provide valuable lessons for organizations navigating these challenges. High-speed interconnects, accelerator-based architectures, advanced cooling technologies, workload management systems, and a relentless focus on efficiency have long been essential components of successful supercomputing environments.
As enterprise data centers evolve to support AI-driven workloads, many of these same technologies and practices are becoming increasingly relevant outside the traditional HPC world.
The Infrastructure Efficiency Framework
Navigating the Data Center Perfect Storm requires more than simply deploying faster processors or larger systems. As computing demand grows and infrastructure constraints become more pronounced, organizations must focus on improving efficiency across the entire computing environment.
In high-performance computing environments, this approach has long been essential. Every watt of power, every unit of cooling capacity, and every cycle of compute time must be used as effectively as possible. Achieving that level of efficiency requires attention to many different aspects of the computing infrastructure, from the physical design of the data center to the way workloads are scheduled and executed.
As enterprise data centers adapt to the demands of AI-driven computing, these same considerations are becoming increasingly important. Organizations must understand how their existing infrastructure is being used, how efficiently their systems operate, and where improvements can be made to support new workloads.
At Olds Research we examine these issues across several key areas of infrastructure efficiency:
Data Center Assessment
Understanding how existing infrastructure is being used and identifying opportunities to improve efficiency and free up capacity.
Cooling
Technologies and approaches for managing dramatically increasing heat loads generated by modern computing systems.
Systems & Integrators
Architectures and integration approaches designed to support high-performance AI and data-intensive workloads.
System Components
Critical technologies—including accelerators, interconnects, memory, and storage—that determine system capability and efficiency.
Performance Optimization
Techniques and tools that help organizations maximize the performance delivered by their computing infrastructure.
Utilization & Workload Management
Scheduling, orchestration, and operational practices that ensure expensive computing resources are used effectively.
Infrastructure & Power Management
Facility-level technologies and strategies that support modern compute densities while managing energy consumption and operational costs.
Each of these areas plays an important role in helping organizations adapt their data centers to the demands of modern computing. Throughout this site we explore the technologies, operational practices, and vendor ecosystems that are shaping this transformation.
Looking Ahead
The forces shaping modern data centers are not temporary. Artificial intelligence, data-intensive applications, and increasingly complex digital services are continuing to drive demand for computing at unprecedented levels. At the same time, the physical realities of power consumption, heat generation, and infrastructure cost are becoming impossible to ignore.
For many organizations, this represents a significant shift in how computing infrastructure must be designed and operated. Systems will become denser, power requirements will grow, and the efficiency of the entire computing environment will become a critical factor in determining how much useful work a data center can deliver.
In many ways, the future of enterprise data centers will resemble environments that the high-performance computing community has been operating for years. The technologies, architectural approaches, and operational practices developed in those environments provide valuable guidance for organizations navigating this transition.
The goal of Olds Research is to help explain these changes, explore the technologies that are shaping them, and highlight the companies and ideas driving the evolution of modern data center infrastructure.
The Data Center Perfect Storm is already underway. The organizations that adapt most effectively will be those that understand the changes happening now and begin preparing their infrastructure for the next generation of computing.