Cloud vs. On-Premises: Infrastructure for the Evolving Data Center

Every few years, it seems, the pendulum swings between the public cloud and on-premises data centers. 

Recently, the cloud has largely dominated, with organizations rapidly deploying cloud-hosted artificial intelligence applications. However, we’re already starting to see the limitations of this approach. As business and IT leaders increasingly seek to bring AI models to their data (rather than pushing their data out to AI models), there is a growing need for infrastructure at the edge of the enterprise network. 

This rapid change creates a significant challenge for organizations: How can they build data center facilities meant to last for decades when the technology landscape changes every few years? 

The answer lies in modularity. By outfitting facilities with modular (and sometimes custom) infrastructure, organizations can future proof their data centers and prepare for continued changes in the IT landscape over the coming decades. 

At Eaton, we have long specialized in the “gray space” within the data center: infrastructure on the front end that supports IT equipment, including uninterruptible power supply systems, switches, backup generators and cooling systems. 

Recently, we have made inroads into the “white space” area where the IT hardware lives. One prominent example is our recently announced collaboration with NVIDIA. We’re working with the chipmaker to design best practices, reference architectures and innovative power management solutions tailored to support high-density graphics processing unit deployments — leading the transition to high-voltage direct-current power infrastructure in AI data centers. 

In both the gray space and white space arenas, we are working to offer customers modular and custom solutions that will meet their needs both today and tomorrow. 

For example, we recently worked with a customer that needed 200 rack power distribution units. The catch? Our standard product wasn’t a fit for their needs. Instead, they requested a custom build. Typically, that sort of request would take 18 months or more to fulfill, due to the bottlenecks associated with product development. But we were able to start shipping units in just four months, allowing the customer to meet its very specific needs, which also increased its opportunity  to become a primary global supplier. 

In the past, we’ve mostly focused on product lines. Today, we’re taking a more holistic look at our customers’ data center needs and how we can help meet those needs. For instance, a customer might tell us they have a certain number of NVIDIA NVL72 racks that they need to power. Rather than simply pointing them to our product catalog, we will work with them to determine how much power they need to support that infrastructure and then design a solution that fits their build. 

It's important for organizations to think about the way their energy needs are likely to evolve in the future. We know that power demand is growing exponentially. Rack power density has increased from 20 to 120 kilowatts, and we’ll see 600kW to 1 megawatt before long. Leaders might look at future needs and determine that their organization will eventually require more power than they can access from utilities. In these cases, organizations might consider onsite power generation with Eaton’s recent collaboration with Siemens Energy; or, they might build smaller infrastructure distributed across various locations that will better meet their overall computing growth requirements.

Regardless of the investments organizations make today, leaders must recognize the ongoing need for updates to address future requirements. The key is to design data centers that can evolve as new technologies come along. 

Data center investments shouldn’t just solve today’s problems. They should create a roadmap that helps organizations solve the problems of the future.