It’s well understood that high-density AI-ready data centers are driving up complexity at the facility level and across the industry – redefining power, cooling, and operational requirements in the process. But an equally critical shift is the increased value and criticality of the assets and equipment that make up modern operations.

The stakes have never been higher and a failure of any kind carries significant weight. The cost of GPUs alone has increased substantially, making infrastructure downtime a much bigger business risk than it once was.

As a result, the ability to manage and respond proactively to potential threats before they take hold across increasingly complex and interconnected facilities has quickly become a defining operational challenge.

Turning up the tension sits another reality: modern facilities are made up of different systems from a range of vendors, all operating under different protocols and assumptions. Liquid cooling is being introduced alongside traditional air cooling, while AI workloads can change rapidly, creating new demands on infrastructure that was never designed for this level of variability.

The margin for error is disappearing, making a holistic, streamlined view of end-to-end operations fundamental to long-term success and scalability.

At the heart of this approach lies modern data center infrastructure management (DCIM) – platforms capable of unifying, normalizing, and presenting actionable data from across each layer of the facility to support critical decision making and underpin risk management.

Today’s platforms must increasingly represent proactive rather than reactive systems, helping operators protect valuable equipment, maximize capacity, and keep critical uptime on track. The first step is having a unified view of what’s happening across the facility – the next is turning this clarity into better decisions.

To explain why legacy DCIM systems are being outpaced in today’s high-stakes data center landscape, Modius’ Craig Compiano, CEO, and John Rhoads, CTO, share what operators should expect from modern platforms – and how next-generation DCIM will be central to finding clarity amid the complexity.

Flying blind

The complexity of modern AI-ready facilities means attempting to identify what needs actioning in and amongst a wide range of disparate data. As a result, it’s all too easy for important details to fall under the radar.

“The big one I’ll pull out is the ability to see how your compute is driving your core building infrastructure,” says Rhoads. “There are very few deployments where you can actually coordinate your compute to your cooling, power, and networking.”

This is a significant blind spot. Operators may be able to see how individual systems are performing, but miss how they interact – or how a change in one area affects the rest of the facility.

A sudden change in compute demand, for example, can create a power or thermal event faster than traditional building management systems or legacy DCIM platforms can respond. Meanwhile, layers of redundancy and protection can mean equipment is repeatedly tripping offline, dialing down compute, or passing workloads elsewhere without oversight of the full impact.

The challenge for today’s operators doesn’t stop at identifying a one-off event that’s taken place across the facility, but in understanding why, what else is affected, and what action should follow.

The power chain itself illustrates the problem. In some facilities, there may be 15 different layers between the transformer and the chip. Tracing these connections is essential when planning maintenance or responding to an upstream failure, yet the documentation required to map them is often incomplete or difficult to bring together.

“When we go into a new project, the level of effort it takes for us to gather the information to be able to encode the software with that power mapping is almost shocking,” says Compiano. “People are virtually flying blind.”

Without a clear operational source of truth, operators may also end up over-provisioning infrastructure simply to maintain reliability. As Rhoads suggests:

“A better understanding of how systems operate could help reduce unnecessary capital expenditure on redundant equipment, while maintaining resilience.”

The data lake

Bringing this understanding together highlights another twist. Many data centers still house a combination of legacy and modern equipment from different vendors, making the task of collecting and integrating telemetry a significant challenge.

“There’s not a data standard that every industry subscribes to,” explains Compiano. “Each piece of power equipment will publish information in a non-standard, unstructured way within that protocol. Then you go into the mechanical systems, and they speak a completely different language again.”

The result is a ‘data lake’ – vast amounts of information with little consistency in how it’s structured or understood.

“You’ve collected all the data, but it’s not time synchronized, aligned, or normalized,” says Compiano. “And so, there’s a data acquisition challenge that exists in every deployment.”

Modius OpenData was designed to solve precisely this problem. By bringing data together from disparate systems and normalizing it during acquisition, the next-gen DCIM platform helps transform fragmented operational technology (OT) data into a consistent, usable structure for the IT layer.

Rather than trudging the lake to collect each and every data point, this process is about collecting the right information in the right way.

“There’s a reason that all of these different mechanisms exist,” explains Rhoads. “They serve very different functions and meet unique requirements for reliability, latency, feedback, and control.”

The method of data acquisition must also be driven by the application itself. Information may be gathered through direct polling, a backend database, or a broker. Each approach offers different trade-offs around security and disruption.

“That’s something that really differentiates the platform,” adds Rhoads. “The strong backend is highly reliable and has a lot of different options for bringing data in. Our experience lies in identifying the right solution so that we don’t disrupt the base systems.”

The ability of the OpenData DCIM platform to scale is equally important. The same operational view needs to remain usable within a single site and across a growing portfolio of facilities. Thanks to the flexible backend design, the platform is able to provide the right information when it's needed across a fleet of critical sites.

The efficiency equation

Whether it’s invested capital or power secured from the grid, given today’s high-stakes, high-value data center environments, maximizing resources is the name of the game. And it comes down to optimization.

“The pain point is you’ve only got so much power,” says Compiano. “If you don’t leverage that power efficiently, you won’t get the maximum volume of compute, the most tokens created, or whatever else the business is trying to do.”

The focus, therefore, is shifting from simply measuring energy efficiency to understanding the economic output that can be achieved from the available resources.

Power usage effectiveness (PUE) remains a useful measure of facility efficiency, but it can’t be relied upon to tell the whole story for AI workloads. A more meaningful picture comes from connecting compute activity directly to power, cooling, and other infrastructure demands.

Across the data hall, pods, and racks operators need to understand how much excess capacity they have, and where.

“People talk a lot about power, but that’s not always the limiting factor,” adds Rhoads. “It could be space, networking, equipment, or supply chain. So, it’s a question of balancing all of these things to your economic output.”

As Compiano explains, at its heart, the equation is simple – supply must meet demand:

“You have a supply of power and cooling, a set space, and a network. On the other side of the equation sits the demand for each of these elements. If the supply is greater than the demand at every moment in time, you don’t have a problem. That’s it.”

The real complexity lies in presenting this equation in a way that operators can actually use to inform critical decisions.

“It’s got to be very simple. A minus B equals a positive number. I’m safe. I can take that piece of equipment offline and do maintenance, or not,” Compiano continues.

Normalize, predict

Achieving this level of clarity also means looking beyond the immediate moment. Many legacy data centers lack consistent historical information. They may have real-time or historic data for individual systems, but not the joined-up view required to understand how conditions change over time.

Without this foundation, there’s no reliable way to identify trends, forecast demand, or make informed decisions about future capacity.

By bringing datasets together and normalizing them, platforms like OpenData can help operators build a more cohesive picture of the facility’s operational history. This creates the basis for predictive analytics and trend analysis that underpin informed planning.

The benefit is a shift from observed conditions to predictive action. Instead of simply identifying that a piece of equipment is under strain, operators can begin to understand when demand is likely to rise, where risk may emerge, and when maintenance can be carried out with minimal disruption.

“Can I take this piece of equipment offline now to do maintenance because I’ve watched the load pattern over the last week or month?” Asks Compiano. “These types of calculations are dependent on having trend data, storing history, and then having an easy way to access that and do math on it to make strong predictions.”

The real-time operational state and the predictive view aren’t separate functions. They need to inform one another, creating a feedback loop between what’s happening now and what may happen next.

The full picture

For the industry to move forward, next-generation DCIM must go beyond describing what’s happening to help operators understand what to do about it.

Legacy DCIM has primarily been a descriptive tool. It explains where things are, how much resource is being consumed, and how much space is occupied. Next-generation DCIM is a prescriptive advisor.

“The right DCIM platform can be leveraged to fill in your operational gaps,” concludes Compiano. “The journey towards improved management means entering into a prescriptive stage.”

Similarly, Rhoads sees a natural progression for DCIM: “There’s going to be a shift from platforms providing information, to ones outlining what action to take, and eventually to taking that action automatically.”

OpenData is designed to help make this possible by bringing together disparate systems, uncovering the relationships between them, and providing operators with a clearer understanding of how their infrastructure is performing.

With complexity only increasing, the ability to see the whole picture is now the very foundation for managing what comes next.