Everybody Wants AI. Who's Paying for It?

Nextera Energy just agreed to buy Dominion Energy for $66.8 billion. And what looks like a utility deal is actually a signal for every AI-oriented executive out there today. AI has become so power hungry that the companies building it are now acquiring the power grid itself, and it's making the entirety of your IT estate more expensive.

So for enterprise AI leaders, that means AI strategy has to go beyond models, pilots, and productivity gains to account for the infrastructure economics underneath it. Things like power, capacity, and cloud consumption are part of it. But in this episode, we'll talk with WWT managing director of automation Mark Wall about the strategic yet often overlooked role of asset utilization and optimization. Because the more expensive intelligence becomes, the less room there is for waste in the IT estate.

Mark's argument is that enterprises already have trap value inside the estate, and he'll offer a playbook for how to capture that value so you can reinvest those savings into the infrastructure AI now demands. From Worldwide Technology, this is the AI Proven Ground Podcast. Let's jump in a looking across the enterprise landscape here, obviously everybody's investing in AI heavily, looking to drive that ROI, but but this stuff is not cheap and it's not necessarily easy to deploy either.

So just to start off, what's standing in the way of all this in terms of enterprise AI execution? Yeah, you know, talking to a lot of clients and we had some advisory sessions as of late. I mean, there's a couple things that constantly come up. You know, the the need to sort of rob Peter to PayPal, right?

Just the cost optimization imperative that's going on. I mean, how how do I fund AI and sort of create new opportunities to sort of fuel that fire? I think the other side of it is tech debt, right? So I want to be able to, you know, have a Ferrari, but like I have a bumpy driveway that has breaks and cracks and everything across the board.

So being able to sort of you know mitigate that is is is a big challenge for a lot of our clients. You know, the other side that we're also seeing is with the explosion of AI and sort of the app development space is the, you know, the the traditional operations and infrastructure side even able to keep pace. So a lot of these challenges we're seeing our clients are sort of you know hindering a little bit to drive the more holistic enterprise AI velocity, but it's it's it's not anything that is you know uncertain that we can we can solve the problem.

Yeah, I mean, businesses have always grappled with that, right? You know, need to run the business versus need to fund the future of the business. How would you advise our clients or organizations out there to to strike a good balance between taking care of what they need to do now, but also keeping an eye on the future to fund innovation, AI, whatever comes next? Yeah, and I think an interesting turn that's coming up with just the sort of you know AI and cyber paradigm is a lot of the back office is now in the front office.

So if you think about it, you know, coming back to basics. So how do I, you know, do things like, you know, optimize, you know, my my purchasing decisions, rationalize some of my inventory and make sure this, I'm not paying for assets that are, you know, no longer necessary or needed. Even simple things like you know, patch management and and sort of you know software hygiene are becoming sort of paramount. And really the ability to sort of look at sort of a life cycle aspect and say, hey, everything from you know purchasing, procurement to deployment to you know, running the day to.

And then something that's particularly coming up quite a bit is is you know decommissioning and just sort of making a full-blown lifecycle capability out of it. Until you can get a lot of that squared away, a lot of the bigger AI challenges are going to be still powerful, but not really where they could be. And a lot of that starts with data and really sort of having a good foundation of data to be able to act upon. If we're talking about more of this life cycle, more of this optimization, AI, meanwhile, is making infrastructure more expensive.

Let's double-click into that life cycle conversation. How does that shift the mindset from what we used to think about as it relates to IT life cycle into what kind of the future of IT lifecycle looks like? Yeah, and like the the joke that you know we we hear all the time is have you tried AI? So what is kind of a you know, almost inception level, you know, type of aspect is I'm I'm I want to do more AI to fuel and power the business, but AI is it has the ability to expose new insights.

So can we take concepts of applying AI and apply it to my infrastructure and apply certain concepts? So the idea of you know being more data driven in how I'm purchasing, being more data driven in how I'm operating, you know, a very common starting point is, you know, really around, you know, mapping, you know, dependency. So if I'm, you know, managing my my life cycle and I need to refresh equipment and sort of help mitigate some of the tech that I have of legacy equipment, I need to prioritize that.

Well, how do I prioritize something that what applications does it tie to? What business units does it tie to? What's the business criticality of some of these assets and the ability to be more data driven in how you think about those things are really going to be key. And so that's a lot of what we're seeing in the industry and well with working with our clients as of late is how do I take a data-driven approach to solving a lot of these lifecycle challenges, starting with insights and rationalization, kind of going into sort of, you know, you know, more unique and automated life cycle planning capabilities of what if I take vendor A and vendor B and do, you know, different sort of what-if scenarios.

These are a lot of these common uh use cases we're starting to see across the board. So we've been talking about the business imperative, the cost pressures, the tech debt, the need to fund AI without breaking the business that pays for it. Mark's answer to all of it keeps coming back to the same place: data. Specifically, whether an organization actually knows what it has.

That's where this conversation goes next. A very common starting point, and a lot of clients say, like, you know, the data exists or it exists in silos or silos of excellence was sort of a term I've I've heard before, where it's all great, but I have multiple sources of truth. So the challenge is maybe what you think you have in your CMDB is different from what your vendors are saying, is different from what you have in your financial tracking systems, is different from what you have in your security and monitoring systems.

And the challenge is when you have different sort of versions of the truth, it's very difficult to sort of, you know, make those informed decisions. And a lot of times our our clients and a lot of the folks that we work with are saying, I don't have time to manage that. I'm just gonna kick the can and just pay the renewal or just you know, make it somebody else's problem. So, really the idea is to start this journey, it's to bring some of these sources of truth together and then rationalize it.

We have sort of a concept we're working on with a lot of our clients where we build a call it a data cube. And so I have my asset data, but I bring in different vectors of data. So I say, hey, here's what the OEMs are saying about the data that I have, here's what my CMDB is saying. But then maybe I bring in financial uh records, you know, what is the depreciation scheduling?

What are the maintenance support costs? What are the available change windows and sort of app dependency, you know, aspects of it? And then finally bring in some telemetry and analytics. And by building that data cube and rationalizing all of those different sources together, you can extract what is actually true and have a lot of different sort of information and metadata, if you will, about those assets to say, you know, hey, is this truly in use or not?

What do I need to replace it? Do I need to refresh it? Is this certified with the latest code version or do I need to update it? When is the end-of-life software support coming up?

And what are the available change windows to maybe swap that out? So having that foundation of data to make those decisions and rationalize that to be true really is the first step in this journey. How much more clarity comes from what organizations thought they understood about their landscape, their footprint versus what they get if they can actually pull off that data cube where it's it's ingesting and and spitting out all those different data sources? Yeah.

And you know, the it to the to the sort of tune of a famous painter, a lot of times they find out they've had happy little accidents along the way where maybe they were paying for a lot of assets that didn't exist, or maybe they were plugged in, but nothing was actually running through them. And we've we've um worked with many clients along this journey to save tens of millions of dollars across their enterprises by being more informed. If I can make decisions to refresh six months sooner or be more precise, or, you know, if I have to rob Peter to pay Paul, do it in the right way across, you know, tens of thousands, hundreds of thousands of assets across the board.

You're talking about substantial savings. The other side of this equation is not just efficiency and and optimization, but also risk, sort of the ability to sort of take and prioritize. Maybe I have vulnerabilities, maybe I have compliance and regulatory challenges. I can sort of interrogate that and prioritize and say, well, this has a vulnerability, but it's not, it's not able to be exploited.

So maybe I don't really have to put that on the on the front. And maybe I can focus on the ones that are truly having that risk. So the idea is within this data cube model, within that sort of data rationalization, I can extract different insights that not just help me save cost, optimize the business, but also mitigate a lot of risk and be a lot more agile in how I sort of defend my uh defend my enterprise. So the data cube solves an infrastructure problem, but it turns out it also unlocks the very thing most enterprises are chasing, the ability to actually run AI effectively.

Yeah, and then what's the other side in terms of how it can help enable an AI strategy? Is it, you know, you're not only talking about efficiency and optimization, you've also mentioned risk. What does that do to enable an enterprise that wants to go very fast with their AI ambitions? Yeah, absolutely.

You know, if if if if you have, you know, there's been a lot of conversations with clients, even internally within within uh worldwide around the need to sort of curate the data in the right way. Having just throwing AI at a bunch of random data sources can be valuable, but if there's a lack of confidence in that data, it's gonna, you know, hinder the AI ability to sort of bring out those additional insights to be able to execute. So the challenge you run into is there's still gonna be the need to rationalize and curate that data a bit.

And then it's gonna allow the AI to make, and if you even think about it, think about AI in the in the aspect of just new protein folding and medical and just all these new innovative things that have never come out. You can apply the same concept where I can I can have the AI look at, you know, even my life cycle, my IT operations, and be able to make recommendations, but that is only possible if I really sort of spend a little bit of time curating and rationalizing those different data sources from those vectors of financial telemetry and sort of asset and inventory data.

Once that's there, it really allows AI to sort of grow and be successful. The other side of that, too, is around observability and telemetry. So for AI to sort of help with the term, you know, agentic operations and AI ops sort of terminology from before, having that accuracy of what my, you know, asset and inventory is, along with the current state and sort of the health and well-being of my assets and my infrastructure, allow these sort of self-healing concepts. So I can sort of apply better insights to my environment with that robust data foundation and then allow action to be self-healing, to create efficiencies to drive that velocity.

But it all starts and stops with that, you know, sort of initial data foundation to be uh accurate and curated. Well, Mark, this strikes me as something that's not maybe an easy subject to broach with senior leaders or boards. It's not exactly a flashy object, IT optimization refresh, things like that. So, how do we advise IT leaders to make the case for this?

I mean, what you're saying makes perfect sense, but like we said, it's not necessarily, you know, the flash in the pan that a lot of AI stuff is out there right now. Yeah, no, it it really is. And I think, you know, there's a term that that we've seen around almost called buying outages. So with a lot of the tech debt that I mentioned before, a lot of the sort of, you know, cyber risk and a lot of the robbing Peter to Pay Paul narratives going on, you need to sort of have those conversations with leadership with the board to really say, you know, if we don't address this, where do you want the outage to be?

Do you want it to be in the wealth management system? Do you want it to be in the business generation? Do you want it to be in the expense applications? Really, the only way for us to prioritize that is to sort of put together this sort of you know data-driven insights and model.

And then you have the ability to say, well, you mentioned I I need to, you know, forego for budget reasons, replacing this, you know, these types of infrastructures. Well, let me show you that the downstream impact of what that's gonna result in if we don't patch or replace that immediately. So to be able to have those data-driven insights and be able to map back to the business, you know, application stack, the the critical business functions, those, those pieces of IT infrastructure are supporting, it's gonna give you the ability to say, where do you want the outage to be?

Right. Where do you want the the risk and how do I manage that risk in the best way possible? So for really the the the concept now and and the idea of the conversations with leadership is all about managing risk. How do I rob Peter to pay Paul?

But how do I make the best decisions and quantify that so I can clearly articulate it to leadership in the board? And and how should a leader articulate all this? Well, I think it's a it's definitely a balance between the technology and the business, but I think making the mapping and and how does that technology sort of tie into those business aspects? So the ability to have, as I mentioned before, those siloed sources of truth.

If I can bring those together and map that this application, this infrastructure stack is supporting this business function and some of the revenue and business criticality tied to that, make that a much easier conversation. So the ability to have those data insights are going to be key, but we got to balance to say, okay, here's what this technology stack is powering from a business standpoint, but here's the inherent tech debt and sort of, you know, technology risks that are associated with that.

Maybe it's based on vulnerabilities, maybe it's based on end of life, maybe it's based on previous performance of incidents and outages to that technology component. So it's really the idea of how do I map those business functions and the and the and the and the criticality of that to the underlying infrastructure. And then really taking that infrastructure and creating sort of a risk profile against that that includes financial, cyber, and operational components and telling that story together.

So that way you solve not only the business side, but also the technology side in that conversation. Let's live in a world right now where this IT leader has convinced senior leaders or the board to make some of this refresh or to optimize its IT in whatever um area. You can't just say, oh, well, now we miss this perceived outage that was somewhere down the line. So how do you read out on ROI so that you keep that momentum, that flywheel going?

What should you be looking at to make sure or to prove that all this is working? Yeah, absolutely. And I think a lot of it is, you know, people look at it from just the number of incidents, the labor that's involved in responding to that, the the sort of cost of the software and and sort of, you know, all you know, capabilities that surround that element. I think what's also very important is, as I mentioned earlier, tying that back into what is that business function that it's supporting.

So if I have a wealth management application that is generating, you know, $100 million a year in revenue, what's the impact of that from an outage perspective, inclusive of just the operational and personnel and all the sort of IT costs? I think that is what is key. And then you have to balance something else with sort of the AI and a lot of the security conversations going on now and everything moving at the speed of AI, you have to think to yourself, what's worse? Being breached because I didn't patch something fast enough or having a potential outage of, you know, an untested bug or feature, and being able to sort of balance that sort of seesaw of, you know, operational confidence with sort of immediate cyber response is another critical aspect into how we need to think about sort of you know ROI business criticality, um, but also impact to the uh to the organization as well.

There's one more dimension to this that we haven't talked about yet. The target is moving. AI is pushing out to the edge, data centers are exploding in size and cost, cyber threats are multiplying. Asset management, what we used to think as tracking routers and switches, is becoming something fundamentally different.

This episode is supported by Riverbed. The Riverbed platform provides open full stack observability, enabling customers to optimize their digital experiences by using AI to prevent, identify, and resolve IT issues. It is very, very much um forcing a need to respond quicker and have everything be dynamic and be updated. I was just on with a client earlier this morning that on the edge of their environment, they need the ability to patch, to unplug, to isolate, but they need the ability to almost instantaneously map this asset at this time ties into this business function, this business unit.

Here's the VP that's in charge of that organization, here's the business owner, here's the downstream applications that are tied from a dependency standpoint in real time. You know, the the idea is if it hurts, rub some AI on it, right? But the ability to leverage AI to then do that dynamic correlation and rationalization in real time and have that, all that additional metadata. So when we think about asset management, it's not just binary in a sense of like, here's the asset, here's some information, and we update it every six months, every year, every couple years.

It's the ability to sort of interrogate an asset and and at any point in time dynamically have a sense on what is it doing, what is it connected to, what is the health and well-being of it, and what is the sort of you know, connective tissue to the rest of the organization. If I take it out, if I upgrade it, if I replace it, what is the sus, you know, the the potential impact of that? So asset management is is is changing dramatically with the ability to have immediate insights, but also if I make a change in the environment to update and rationalize that almost immediately as well.

Yeah. I mean, if the core idea here is that AI is making infrastructure more integral and more expensive to the enterprise, do you get the sense that that leaders understand exactly uh exactly how expensive that infrastructure is with AI, or is that just a continuous shifting price point? Well, I think you know, a lot of clients are saying, you know, with the leveraging the frontier models and and it and it wildly depends on the organization and their maturity and their adoption and their needs, but it's it's coming to the point of with these frontier models and the cost of that, it almost might be worth building out, you know, uh an on-prem AI infrastructure or sort of an edge, you know, component.

So the ability to sort of take and and and and sort of quantify the not only the token spend, but also the cost of operations and the cost of the infrastructure that support that. So the ability to sort of say, hey, I'm not only mapping that infrastructure to you know the the business criticality and the business function, but now that I have the ability to sort of look at you know these assets and these and these from a cost, from a financial, from a risk perspective, it gives me the idea to say, hey, I need to optimize and fund more AI, but I still need the infrastructure to support it.

Can I leverage a more intelligent life cycle approach to squeeze every ounce? Efficiency from the the sort of iceberg, if you will, of infrastructure that's supporting it? How do I rationalize my inventory? How do I look at different solutions in real time?

How do I look at even things like lead time? And if I if I want a best of breed solution, but you know, maybe there's you know 18-month lead time, but there's something that's maybe more expensive that I can get sooner, what's the cost-benefit impact of what the that decision would be? So the idea of lowering my operational cost, OpEx, CapEx, through a lot more efficiency in how I'm I'm managing and driving my life cycle will ultimately allow those dollars and cents that are saved to be funded back into AI.

And that ultimately creates that flywheel effect. The more AI I leverage, the more I can optimize my environment, the more I can sort of put back into the AI bucket for the rest of the organization. Yeah, maybe to put a little bit of that into real-world instances, Nextera Energy is buying Dominion Energy for $66.8 billion, or is agreed to buy them for $66.

8 billion, which to me says that AI is becoming a power and infrastructure and optimization game. I mean, those are those are massive companies. But what is that signal to to the rest of the market in terms of how we need to think about lifecycle? I mean, maybe it's going to be repeating a lot of what you've already said, but what is the signal with that deal that others, even outside of the utility industry, should be looking at?

Well, I think it's shifting that, you know, the the ability to have power and the ability to fuel and and drive the the AI machine, the machinery, the GPUs, the data center infrastructure is paramount. It's going beyond a sort of engineering game. Now it's a physics game, like to the point of how do how do I just generate that much power and cooling to fuel this? We see a lot of clients, and this is sort of the shift as well.

Is I think there's even talks of a of you know a megawatt rack at some point in the future, right? Because of how dense this is. So a lot of the challenges we're seeing with clients and the market signals we're seeing is power is is becoming paramount. Some of our clients' data centers only have so much power in space, which ultimately drives a very core need to create as much efficiency as possible.

So, how do I go and build best of breed infrastructure that is consuming and the most, you know, the least amount of power per output of whatever I need, not even counting GPUs, but even storage, compute, network, those sort of core underlying infrastructure components, be able to sort of rationalize that using sort of what we talked about before by doing automated optimization of your environment, right? Doing that on a real-time basis of your assets as part of your life cycle, freeing up power and cooling space so I can put in more high performance compute.

Almost every client we're working with that has sort of these data center modernization approaches are looking for a more intuitive way to optimize the sort of, I'll call it back office infrastructure, which is still the backbone of what we're doing, to be able to free up power cooling space and capital to fuel the larger sort of high performance GPUs and sort of the AI platforms that are going to power the business. So a lot of the market signal is saying, hey, we need to come together to build and deploy as much energy and and and as safely as in as possible.

But what that downstream is telling our clients is you need to do the same thing within the four walls of your organization. Yeah. So if if the margin for error for managing these IT assets is is shrinking, let's think about a client or an organization out there that you see as doing it well. I think a lot of the good ones are, you know, sort of always being curious and always being open-minded to, you know, we'll use the word micro experimentation towards a North Star of sort of having everything be data driven and sort of continuous optimization.

So, really, those, the, you know, a lot of the clients that we work with, particularly one, is really embracing this concept of, hey, I need to continuously interrogate and rationalize what I'm getting from the vendors, what I'm getting from my CMDB. How do I constantly sort of rationalize and shrew that core asset insights and then map that against things like change control windows, resource availability, vulnerabilities. And that way I have a more accurate state of that environment.

This client has, you know, even in the order of over 150,000 network assets across the globe. So the ability in real time to prioritize and say, hey, I want to replace this. Well, what is my velocity for change, even? And the idea is they've saved tens of millions of dollars by not just ordering a bunch of equipment, having it sit in a warehouse, and they're not depreciating, they're not writing off losses on those assets.

They're able to more optimize and purchase in in sort of real time and plan accordingly, then the ability to have not just more accurate insights to plan, prioritize, and optimize, but leveraging, you know, automation to be able to go from they went from 400 days to 40 days from procurement, place and order to production. So it took over a year for that value to be derived from when they purchase something. Now it's a little bit over a month, leveraging a lot of the new kind of automation capabilities.

And then finally, as we kind of talked about, there's sort of a third bucket, insights, actions, and then a lot of this sort of resiliency. So now that I have this, these insights, and now that I have a lot of the automation in place, they're able to sort of drive self-healing infrastructure. So if there's an incident, if there's an outage, or even you know, there's upcoming, you know, potential signals from power supplies and various things like that, they can auto-generate tickets, they can maybe fail over links.

There are more data-driven, more AI-powered infrastructure that's allowed them to be more resilient, save costs, and ultimately just mitigate a lot of the risks with just the velocity of what's going on in the market with AI. Yeah, lots of use of the word automated or AI-driven or AI powered. But another thing that you said at the top of this of that answer was curiosity. So, how much room for human creativity or how much need for human creativity is needed in some of these discussions for lifecycle optimization?

Does that help? Or are we looking just to be kind of black and white data-driven? Yeah, I think there's it's always shades of gray at this point. So the idea, especially with AI, is the superpower, if you will, of AI is it's non-deterministic in a lot of sense.

So the idea to make recommendations, it's the same thing with humans. You know, be curious. So a core key capability is to have, we're seeing a lot of clients do this. We have this internally, is how do I build sort of AI playgrounds and proving grounds and safe spaces to experiment, to test, to test and do something wrong and something happens, but it's in a controlled way to have those happy little accidents.

So it's very much needed and encouraged. And it should be a percentage of you know your team's time to drive experimentation. And really the concept is you need to remove barrier to entry. So a lot of what we're doing on our teams and within our clients and internally, you know, here at Worldwide is you know, remove that barrier to entry.

So if you have a couple free hours, you can safely experiment, you can safely try things out. You know, hey, maybe I want to apply sort of a new, you know, kind of AI workflow that automatically optimizes, you know, some of the storage, you know, things that are that are going on. You should be able to give your teams the ability to sort of have that freedom. So having that curiosity is paramount.

Encouraging that sort of exploration is important, but you really need to make sure you have a safe way and an easy way that removes the barrier to entry to experiment for a lot of this to be successful as well. Okay, thanks to Mark for joining. The lesson here is that AI strategy can't be separated from infrastructure strategy anymore. Every enterprise wants the upside of AI, faster decisions, smarter operations, new efficiencies, new business models.

But Mark's point is that the future does not get funded on invention alone. It gets funded by clarity, knowing what you own, what you use, what is aging, and what's vulnerable. This episode of the AI Proven Ground Podcast was co produced by Nas Baker and Kara Kuhn. Our audio and video engineer is John Knoblock.

My name is Brian Phelps. Thanks for listening. See you next time.