Rapidly expanding AI-driven data centers are becoming critical infrastructure, much like refineries and industrial plants. Emerson’s Aaron Crews joins me to explain that increasing computational demand is driving higher power and cooling requirements, making reliability, automation, and advanced control strategies essential.

Aaron highlights how industrial-grade automation systems like the DeltaV™ automation platform, traditionally used in sectors such as chemicals and energy, are increasingly relevant for data center operations because they focus on 24/7 reliability, redundancy, and optimizing complex processes.

We explore how advanced control techniques like model predictive control (MPC), data fabric architectures, and AI-assisted operations can optimize cooling systems, improve uptime, and manage highly dynamic AI workloads more efficiently.

We also discuss the challenges of executing projects at high speed and scale and emphasize modular approaches like electronic marshaling to reduce integration complexity and the cost of change.

Why it Matters

AI data centers are becoming foundational infrastructure for global digital systems, but their scale, energy demands, and complexity are pushing traditional IT facility designs to their limits. Applying proven industrial automation and control strategies can significantly improve reliability, energy efficiency, and operational resilience while reducing risk in mission-critical environments.

Key Takeaways

• Data centers are shifting from IT facilities to industrial-scale critical infrastructure requiring 24/7 high-reliability operations
• Advanced control methods like model predictive control can optimize cooling systems and reduce the need for excess capacity
• Unified data models and “data fabric” approaches enable better visibility, integration, and AI-driven decision-making across complex systems
• Modular automation design (e.g., electronic marshalling) helps accelerate project delivery and reduce costly late-stage engineering changes

Transcript

Jim: Hi everyone, I’m Jim Cahill with another Emerson Automation Experts podcast. Today I’m joined by Aaron Cruz, who leads our product strategy for PLC and SCADA platforms. We’ll be discussing the evolving landscape of control and data centers especially as they grow increasingly large and more sophisticated in this era of ever increasing AI. Welcome, Aaron.

Aaron: Thanks, Jim. Thanks for having me. I appreciate it.

Jim: Let’s kick things off by let me ask you, you know, data centers are starting to look more like industrial facilities and office buildings. What’s driving that change?

Aaron: Obviously there’s a lot of activity around data centers and AI and everything right now. So the whole AI revolution driving this crazy build out of data centers. But as those capabilities kind of continue to grow and as the use cases for AI continue to grow, then the scale, like you mentioned, And then the reliability that’s required to keep those data centers in operation to keep all those demands working, 24 by 7, that drive continues to increase.

And so where, before we had a situation where you had basically a data center, it was stood up, obviously you wanted to be running, as reliably as possible, but now that’s not just serving the cloud or serving somebody’s app or something like that, but it’s serving critical infrastructure for the country and for the world. And so the role of what a data center was in terms of just being able to give us software and scalability and some of those things to critical infrastructure basically changes the risk profile and the scale obviously associated with those kind of projects.

Jim: Yeah, and those servers get ever and ever more power hungry than it just seems like what they consume in the way. And what’s required is just no longer your basic office building anymore. Now, that shift, why does it create opportunities for automation systems like DeltaV?

Aaron: Systems like DeltaV basically designed for those critical infrastructure applications. So for years, we’ve been building chemical plants and refineries and critical manufacturing of all types. And those control systems were evolved to basically give you 24 by 7 operations, right, for decades of life. And so you have a situation where you have an industry that’s transitioning into more of that critical infrastructure role, and you’ve got a set of automation solutions that are basically built for that purpose. And so that’s where we’re starting to see those worlds converge.

Jim: Yeah, that makes sense. If you have systems designed to handle refineries, LNG plants, pharmaceutical plants, as the complexity grows in data centers, they’re entering that realm of this critical infrastructure like those others. I guess traditionally data centers have relied heavily on infrastructure redundancy to maintain their reliability. Is there a different way they should be thinking about reliability?

Aaron: I think so. I think that if you’re starting to build out a factory, or in this case an AI factory, and you want to make sure that it’s always available, that equipment can fail. It’s natural to build out process redundancy and redundant equipment. But for other industries like the ones that DeltaV serves, of course, yes, you have some process redundancy, but more importantly, you don’t want it to fail at all in the first place. So if you have systems that are designed for that 24-by-7 operation that are designed for really, really low failure rates, then you’re going to be even better off, if that makes sense.

So just to give you an easy example, if you’ve got a lineup of 21 chillers for your cooling system and each one’s on an independent PLC, you’ve got 21 chances for a PLC to fail. Now, when one fails, you’re only losing one chiller and you have a lot of process redundancy to make up for it. But if you could take modern control strategies, put it on top of that, then you have all 21, you know what I mean, all the time. Because if one piece of your automation system fails, you have redundancy there to support it. So that’s like a simple example where you can get orders of magnitude more reliability out of a system by having the right system architecture.

Jim: Yeah, and it would seem even the control strategy could adjust to that loss of a chiller in there to be able to handle it, more than independent islands controlling each one could be able to handle that.

Aaron: Yeah, exactly.

Jim: So what are the biggest opportunities for advanced control in the data center cooling systems?

Aaron: Cooling, I think, is one of the places where there’s just an obvious opportunity. We talked about how we’re taking some of the lessons from other industries and bringing them into the data center space in that previous example from a reliability perspective. But there’s no reason why you couldn’t do the same thing from a process optimization perspective as well.

So, if you have a chemical reactor, you’re trying to prevent a runaway, thermal runaway. We’re also trying to prevent thermal runaway in these data centers as well. So in the past, what you would do is just do simple, like simple coordination between all those chillers to say, okay, this is the power demand, this is the heat that’s gonna be generated, therefore we need to run these many at whatever rates. And that was sort of okay.

But in the AI world, those demands are a lot more dynamic. And those simple control strategies can’t adjust fast enough. And so what ends up happening is you end up basically running full tilt. You end up running excess capacity to handle that increase in variability on the front end. And this is like process control 101. This is like what we’re built for. And so the control strategies that we would use to prevent that kind of runaway thermal reaction in a chemical plant are almost ideal for that kind of use case.

So the main one that we would use, obviously, is something like model predictive control. So PredictPro and DeltaV or Aspen DMC3™, where you would be looking at those demands, understand the model of the facility, what the delays are in that model, and you can feed forward and actually predict what the cooling load’s going to be and use that to optimize how you run your chillers.

Jim: You know, and it seems like with just the sheer kilowatts that are added to these racks to handle these massive processors that are running in there, you’re moving up from more traditional valves you’d find in an office building to industrial-strength valves and the sensors that go with it. And I think all that ties in with the same equipment that you might find in a refinery or LNG plant is now being specified for that, which improves the whole control process.

Aaron: And then you start worrying about measurement quality. high reliability and uptime. Equipment that’s gonna run for a long time and the diagnostics that you need, if there’s any issues with any of that equipment. So it’s more than just data.

So people are coming around to the idea of let’s bring all the data together, but a single process value is just one aspect of the data. So there’s a lot of parts, moving parts to that. But if you think about as that comes together, what’s the opportunity? for operators, it’s huge, because could you, could you imagine running a refinery where you just handled all of that process variability with excess capacity?

We’ve squeezed so much capacity out of our critical infrastructure over the years through these kind of optimizations. That’s why we’re able to run with optimal energy usage, run as lean as we can, and to basically unlock that latent capacity in our infrastructure.

Jim: Yeah, you’re seeing that right now in refineries around the world, just full tilt going and making things. Everyone is talking about AI and data centers. Where does automation and advanced control fit into that conversation?

Aaron: Yeah, I mean, Obviously it’s funny. It’s kind of a meta thing that you have like, you have an AI factory and then you’re saying, okay, well let’s also use AI to control it or let’s use AI to help with running it or let’s use AI to even help with building it. And so there’s a lot of different places where AI can help. We’ve been using AI-based tools to help our project teams go faster. So that’s one aspect of it.

Then certainly when we get in and implement the control system, things like DMC, model predictive control, that’s artificial intelligence that’s baked into the system to help manage the dynamic control loads of the system. So we’ve had a lot of data center engineers and operators come to us and say, do you have some AI that can help me manage these dynamic loads and switch my controls strategies around and things like that. And the answer is like, well, yeah, absolutely.

We’ve been doing AI and controls for a while. And so we can definitely apply that. And then now what’s becoming kind of even more interesting is now when we can move into operations and we say, oh, okay, We’ve got a lot of different systems. We have all that data together. How do we make sense of it. So how do we do the analytics and stuff like that, but also how do we make the right operational decisions? So as an operator, how do I have the guidance from all the data that’s available to know what to do, right, in a given situation?

And so one thing we didn’t talk about is cybersecurity and connectivity and things like that. That’s where we’re solving these same problems in those other industries. So we bring in like air gapped AI that can run kind of inside the Purdue model, local to the site, and still can advise the operators on what the right moves are and things like that. I think we’re still, I don’t know anybody that’s like, you know, closed loop AI, you know, Claude code is gonna run my plant or whatever at this, you know, stage, but But human in the loop, like AI advising is absolutely like here today and baked into our control systems.

Jim: Yeah, I think that role as an advisor, but human in the loop, that smartest operator out there whispering in the ear saying, hey, we need to check this or check that, but being done that, yeah, that makes complete sense. Maybe one day, way off in the future, but not now. Some operators are heavily invested in PLC and SCADA environments. Does moving towards a platform approach like what DeltaV offers require a complete replacement strategy?

Aaron: If you think about the way that these data centers are getting built today, I mean, as Emerson, I would love for everybody to just use all of our stuff. And like you said, valves and the transmitters and all the industrial stuff to make it happen. But the way that the data centers are getting built today, the technologies are evolving. There’s a lot of different vendors that are involved. There’s a lot of different systems, you know, from the electrical stuff, power generation side.

The cooling, which we’ve been talking about, and bringing all of that stuff together, it’s probably unrealistic that you’re going to have just a pure unified. system. For the critical loads and the things that, the places where you can, having a DCS, or more centralized approach makes a lot of sense. But still, you’re gonna have subsystems and things that you need to integrate.

So I think from our perspective is we’re gonna have to meet, from our perspective, we’re gonna have to meet operators where they’re at in terms of that. automation system maturity and the mix of subsystems that they’re going to have. So it’s important that we’re flexible to basically go from a PLC-based architecture to a DCS, PLC and SCADA to a DCS architecture or something sort of in between even where maybe you have DCS for that high redundancy, critical load stuff. but still have visibility to everything else that’s out there.

So one of the important technologies that Emerson has as well is I would call it like a data fabric at that L2 in the Purdue model where all those independent systems can tie in and we can have consistent visibility across the operation.

Jim: And I think that’s particularly important as AI is coming into it instead of having these islands of data around that aren’t connected, that they’re brought in together to put that layer on top of that data fabric. Yeah, that makes a lot of sense.

Aaron: Yeah, I think it makes sense from both, just getting the data together, right, and historizing it and alarming and having visibility, but also getting it out to the other consumers who need it. So that’s another piece of the trend, right, is not just having local data, but being able to bridge together multiple sites across the campus, get data egress to different vendors that are going to need to support that.

And that’s another big advantage around with using this kind of industrial automation infrastructure is the security models that are built around that are designed for that. So it’s easy to not be a target from a cyber perspective and just do your own thing and weave it all together however you want to do it.

But we move into critical infrastructure space and the I would just say the regulatory environment is different, the expectations are different, and obviously these facilities can have bigger targets on their back from a cyber perspective as well.

Jim: Yeah, that’s so true. Now we’ve spent a bit of time here talking operational side of things, but this data center expansion is happening at an extraordinary speed. So how does automation impact the project execution side of the equation?

Aaron: I think that when we look at just the pure speed that these projects are having to move at, it seems extremely challenging. And it’s tough like on the supply chain side. Just getting all of the equipment that you need to get, what vendor is going to actually be able to deliver on time and that sort of thing.

The knowledge piece is becoming as scarce as anything. And so there’s, it’s tough. What we’ve seen is basically the speed is the speed. And so if we don’t have enough definition, we’re just going to go forward without good definition. And that’s like, that seems like, not maybe the best idea, but what’s the alternative. And so that’s what we’ve seen.

People are, they don’t have full definition. Maybe we know 40% of the scope. Maybe we know 60% of the scope, and we’re just going to go. And that’s fine until you get later, you know, into the project. And now the decisions that you would have made if you had that definition earlier, they become extremely costly. And the cost of change shoots through the roof and then the projects get delayed. So you think you’re going fast and you end up going slow. You think it’s expensive and it’s even more expensive. And so those things are really not great. To put it mildly, I guess.

To get back to your question though, the thing that, the other place where we’ve been investing is DeltaV, right, is in how can we use the technology to help with that problem. And so we started this quite a few years ago, right, when we started thinking about building technologies for projects. So capital projects, you know, that lack definition, that need to move quicker, where we want to limit the cost of change is a solved problem as far as we’re concerned.

And so we have technology like electronic marshaling with charms, where you basically can have your cabinets, like one standard design of your cabinets, go and deliver those, install them wherever with a lack of definition. And then once that definition arrives late in the project, you can just plug the IO into any available slot. You can assign that IO to any available controller. So you have cloud of IO, cloud of controller.

The engineering happens all separately, so there’s no dependencies between the software and the hardware. And all of that comes together at site. And it’s just sort of like magic because it’s designed into the system to do that. And that goes back to that first problem that I was talking about around the expertise challenge. So if the humans in the loop, they got to stay in the loop as we talked about, that they have to deliver these projects and do the systems integration, if we can make their lives a lot easier, And then it helps everybody.

And we have less dependency on that critical piece of the supply chain as well. And so, as Emerson, we’ve been super focused on ease of use, human-centered design, making everything plug-and-play, and taking the effort out of that integration task that shows up right at the end of the project, where all the pressure’s on to get everything commissioned and going.

Jim: you had mentioned the electronic marshaling. It’s interesting because that actually came out of the offshore oil and gas business, which is in a lot of ways trying to get to first oil as fast as they can, getting started, not having a complete definition of everything, late changes, the distance to get offshore to the platform to do it. And a lot of that came from our customers to say, hey, solve this problem. And it just so happens it’s not just an oil and gas thing, but with this tremendous rush to do data centers, it’s a great technology for that too.

Aaron: Yeah, exactly. And it just, it’s like we’re repeating ourselves a little bit, but it’s like we see our customers come to us with these challenges and that’s the good part about being a company that’s touched so many different industries. that are having to solve those exact same challenges. And we just have to introduce that technology to this world so that they understand how powerful it can be.

Jim: Yep, that’s excellent. The industry often discusses concepts like unified namespaces and data contextualization. How do you view that challenge?

Aaron: I think context really matters. the asset frameworks, how the data is structured, that matters a lot because when you’re trying to put optimization on top of it, you need to know the state of the process. And the state of the process has everything to do with, to understand that, you have to understand how the data is structured and how the asset is set up underneath.

And then if you bring all the data together, then that’s where this unified namespace idea comes in. It’s like, oh, in this PLC we call the tags this, and in this PLC we call the tags that, and we need to have a unified namespace. It’s a little bit of a foreign concept to us because we’re used to a unified name space. So the DeltaV world, it’s built on a unified name space that’s baked into the system. And the contextual data, both from like an S88, S95 perspective and from a physical control system infrastructure perspective, is already baked into our system.

And then when we do data egress with like the DeltaV Edge and bringing data up into higher levels, higher levels of the Purdue model and giving other people visibility to the data and all of that stuff, it’s already in context. And so that goes back to what we were talking about earlier, which is basically, I would just call it like a level 2 data fabric, which provides that unified namespace. Ideally, You do that at the asset level, and then that way your maintenance and everything else becomes a lot more straightforward. But even in a situation where you have a lot of different vendors doing a lot of things their own way, we can bring that data together into a common namespace, give it the proper context, pull it into one unified view of the facility, and also pass it up into the enterprise.

Jim: I think from our conversation, data centers have grown from something that’s part of the office building into really critical infrastructure, much like refineries and offshore production platforms, pharmaceuticals, all that. So what would you say Emerson is able to bring to this critical infrastructure of data centers?

Aaron: Emerson, I mean our technology, our automation systems and technology are built for critical infrastructure. The systems, the way they’re designed, the redundancy that’s built into that is designed for 24 by 7 operations. The way the system’s designed from a cybersecurity perspective is designed for those critical infrastructure applications, the capability inside of those systems to drive the optimization and unlock the capacity of that infrastructure.

The way the system’s designed to help projects move faster to deliver critical infrastructure projects on time and on budget. And The support network that’s built around the world to keep that control infrastructure alive, well, and serving are all things that we’ve been using in those other industries. So it makes perfect sense for us to take that into this new type of infrastructure in AI data centers that’s being built out today.

Jim: That does make complete sense. It’s joined the ranks of all those other big critical industries that make our world really what it is today. Okay, I’m going to wrap it up with one final question for you. So what is the one thing that the data center operators may need to unlearn as they rethink their automation strategy?

Aaron: I think they just have to be open to a change in approach. There’s an inclination that we’ve done it once and it worked, so therefore we’re going to print many and repeat that. But if you have issues that exist and you’re not willing to evolve your approach to eliminate those issues, then they’re just going to compound. And so, what we’ve been doing is partnering with those operators to help them understand how we can evolve their architectures, evolve their project delivery approach, and evolve their operations and optimization strategies so that they can take advantage of those capabilities and not have to relearn the lessons of the past.

Jim: I think rethinking what you do because the fact that you are critical infrastructure now and it makes a whole lot of sense to be able to do that. If you’d like to learn a little bit more, visit the DeltaV Data Centers section on the Emerson website and learn about some of the solutions and things you can do to make those projects run smoother and keep them operating to the reliability standards that you need.

Aaron, thank you so much for joining us. joining us today and sharing your expertise with our viewers and listeners.

Aaron: Well, thank you so much, Jim. I really appreciate the time.