Artificial Intelligence (AI) may be driving the next generation of data center growth, but in many cases, the bigger challenge often isn’t the AI platform itself.

It’s the infrastructure underneath it.

Cooling systems, power distribution, environmental monitoring, control platforms, and analytics tools all need to exchange reliable operational data. When they don’t, even sophisticated AI tools may struggle to deliver meaningful insight.

Many in the industry would agree that AI is powerful. The harder question is whether the facility underneath can support it.

Even when AI sits at the top of the operational technology stack, interoperability often determines how effectively it performs.

This post covers why many facilities aren’t AI-ready even when they’re brand new, what fragmented data does to AI performance, how the integration conversation is changing for both integrators and OEMs, and what facilities making real progress are doing differently.

Data Centers Are Now Industrial Infrastructure

Something fundamental has shifted in how the industry talks about data centers. Many of the facilities being built to support AI workloads operate more like power plants or manufacturing floors rather than office buildings with servers in them. Power systems, thermal management, and controls layers have become so deeply interdependent that treating them as separate domains may introduce operational challenges.

A cooling decision can influence power usage. A controls decision can affect reliability. The facility has become part of the compute equation.

That shift changes what “integrated” means—and it changes the cost of getting it wrong.

The pressure is significant: AI workloads are exploding, cloud demand is growing, and facilities are scaling faster than ever. In some cases, timelines compress the planning and coordination that integration efforts typically require.

Integrators and operators are often being asked to support:

• Centralized operational intelligence
• Predictive maintenance
• Automated fault detection
• Energy optimization
• Remote operations
• Enterprise-wide visibility

The technology to do all of that exists, of course. The challenge is often the infrastructure underneath them.

Many data centers weren’t designed this way. They evolved with new systems being added during expansions. Different vendors came in at different phases, and legacy infrastructure often stayed in place because replacing it introduced too much cost, downtime, or risk. Temporary integrations became permanent, too.

The result can be facilities where BACnet supports building automation, Modbus connects power infrastructure, SNMP handles UPS systems, proprietary protocols run cooling, and a cloud analytics platform at the top is waiting for normalized, structured data from all of it.

Individually, these systems may perform well. Collectively, without effective integration, consistent data exchange can become difficult.

That integration layer is critical to many AI initiatives.

What Happens When AI Meets Fragmented Data

One common misconception is that AI platforms inherently deliver operational clarity. In reality, their effectiveness depends heavily on the quality of the data they receive.

When the data flowing into a platform is inconsistent, siloed, delayed, or incomplete, the analytics layer may inherit those problems, too.

Here’s what system integrators may encounter:

• Alarms that don’t correlate across systems
• Trend history that disappears intermittently
• Conflicting values for the same condition across platforms
• AI models operating on incomplete representations of the facility

These issues typically reflect underlying data and integration challenges rather than failures of AI itself.

In practice, that means operators may spend more time validating data than acting on it, which is the exact opposite of what AI initiatives are supposed to accomplish.

The fix rarely starts at the analytics layer. It can start at the protocol level, where systems that were never designed to talk to each other need a reliable way to exchange information.

That’s the challenge FieldServer gateways help solve in data centers, industrial facilities, and commercial buildings around the world, long before the dashboards ever light up.

New Build? Don’t Assume You’re AI-Ready.

For years, interoperability was framed as a legacy problem. Old buildings, aging controls, and systems cobbled together over decades by vendors who never had to think about interoperability. The approach was “get the old stuff sorted and move on.”

Today, thanks to the current data center construction boom, this framework is becoming genuinely costly.

New facilities are coming online at a pace the industry has never seen. AI demand, hyperscale expansion, and cloud growth are driving large-scale builds, sometimes under compressed timelines.

The pressure to get online as fast as physically possible is intense enough that the integration planning that should happen during the design phase often gets waylaid.

The assumption is that integration can be sorted out later, after the facility is operational.

It can, but that’s an expensive lesson to learn when live AI workloads are running.

A greenfield data center built right now might have the latest-generation cooling, state-of-the-art electrical infrastructure, and the most capable AI analytics platform money can buy, yet still be operationally fragmented.

This is not necessarily due to poor decision-making. It is often a result of scale and complexity.

A major build typically involves dozens of vendors, contractors, and equipment suppliers, each arriving with their own protocols, data models, and integration assumptions:

• Cooling infrastructure from one OEM ecosystem
• Electrical distribution from another
• Fire and life safety systems under their own certification requirements
• A BMS, a DCIM layer, and enterprise monitoring tools each with their own architecture
• An AI analytics platform selected, in many cases, before the integration questions were fully resolved

In many cases, fragmentation is not intentional, but it can still emerge.

Still, there’s yet another dimension to greenfield projects that also doesn’t get enough attention: operational novelty.

For example, liquid cooling systems have expanded more recently, power densities require completely different monitoring approaches, and rack-level thermal data that needs to move in real time across systems weren’t part of many original integration specs.

The playbook from the last generation of data centers doesn’t fully apply here.

Trends Reshaping the Integration Conversation

The interoperability challenge isn’t static. Several industry shifts are making it more urgent and more complex at the same time.

Liquid Cooling and Integration Readiness

As high-density AI racks push thermal loads beyond what traditional air cooling can handle, liquid cooling adoption is growing. Some implementations may have limited native integration support, which can create visibility gaps if not addressed during design.

OT and IT Convergence

Enterprise platforms, cybersecurity requirements, and AI tools are increasing the demand for normalized, accessible data. This is placing additional expectations on OT environments to integrate with IT systems, while maintaining operational reliability.

Procurement and Integration Gaps

Equipment is often selected based on performance and reliability criteria first. Integration questions around protocol support, upstream data accessibility, BMS compatibility often come later. That gap is increasingly showing up during commissioning and go-live.

Cybersecurity Considerations

The days of open, flat OT networks are disappearing. Segmentation requirements are changing how data moves between systems, which means interoperability solutions increasingly need to work within security architectures, not around them.

The Question Integrators Are Being Asked Now

The old integration mandate was straightforward:

• Get systems communicating.
• Complete the point mapping.
• Deliver visibility.
• Maintain continuity.

Now the scope has expanded to include:

• AI readiness and cloud connectivity
• OT-to-IT communication and data normalization
• Cybersecurity segmentation
• Long-term operational scalability

The question used to be: “Can these systems connect?”

Now it’s: “Can these systems produce reliable, normalized, consistent data that a higher-level platform can actually use?”

That’s a much bigger challenge.

And it’s one where the integration layer—the part that handles protocol translation, data aggregation, and cross-platform communication—has gone from a technical footnote to a strategic decision.

The interoperability decisions being made today can affect analytics capabilities, AI scalability, and facility flexibility for years to come.

OEMs Are Being Evaluated on Integration, Too

Equipment procurement used to live almost entirely in the world of reliability, performance, efficiency, and cost.Those things still matter but integration readiness is becoming part of the evaluation.

Operators increasingly want to know:

• Which protocols does it support?
• Can the data move upstream easily?
• Will it work inside multi-vendor environments?
• Does it create operational visibility or operational silos?
• Will it integrate cleanly into BMS and DCIM environments?
• Does it align with cybersecurity expectations?

As a result, some OEMs are incorporating integration capabilities into their offerings.

Programs such as the FieldServer gateway OEM program are intended to help embed interoperability earlier in the system lifecycle, reducing the need for retroactive integration under operational pressure.

What the Facilities Making Real Progress Do Differently

The organizations scaling AI and analytics initiatives successfully tend to approach infrastructure the same way. A few things come up consistently.

DO: Treat Interoperability as Foundational Architecture

The facilities that solve interoperability early may reduce delays. The ones that defer it often spend months cleaning up data problems instead of using the insights they paid for.

DON’T: Assume New Equipment Means Equal Integration

A brand-new system can still create an isolated data pocket. Ask about protocol support, upstream visibility, and BMS compatibility before the purchase order, not after the equipment is installed.

DO: Take Legacy Infrastructure Seriously

Older systems are often the most operationally critical equipment in the building. The goal isn’t to replace them on a software-driven timeline. It’s to connect them effectively so they don’t become blind spots.

DON’T: Separate the AI Conversation from the Integration Conversation

In practice, they’re the same conversation. Organizations that treat them separately usually find that out the hard way.

DO: Plan for Growth from the Start

Integration strategies should support expansion without requiring significant redesign.

DON’T: Wait Until Deployment to Ask Integration Questions

By the time the dashboards are live and the gaps are visible, the integration work that should have happened during design is now happening under operational pressure. That can be a harder and more expensive place to solve it.

The Protocol Layer Nobody Talks About Until Something Breaks

Much of the attention in the industry focuses on AI, analytics, and automation. Less attention is given to the underlying data exchange mechanisms that enable these capabilities.

These include:

• Protocol translation
• Data aggregation
• Normalization
• OT-to-IT bridging
• Cross-platform communication
• Legacy integration

FieldServer protocol gateways are designed to support this layer by helping facilitate communication between systems that may not otherwise exchange data effectively.

Before AI platforms can deliver meaningful insights, underlying systems need to communicate reliably. This challenge can apply to both existing facilities and new construction.

Contact us to learn more about interoperability strategies for modern data center environments.