AI Infrastructure Readiness Drives 2027 IT Budget Priorities

Hey guys, thanks for the intro. We're here with Jack Lodge, who's Executive Vice President for Customer Success at NWN, and we're having a little chat about, well, where are people going to start allocating their budgets for 2027? Because, well, I know that everybody's at the beach or at some sort of lake house, but the reality is we need to start planning how we're going to spend money in 2027 if we expect to be ready.

So, Jack, how you doing? Welcome to the show. I'm doing great, Mike, and thanks for having me.

Where are people starting to kind of focus on for 2027? The problem as always is that there's more things we can spend money than we have on. It's an age-old problem.

But this year, is there anything different than you're seeing happening? Yeah. I think, obviously, AI is a dominant trend across the industry, and everybody's talking about AI, AI adoption.

Agentic workflows are now a big focus. But what's interesting across our customer base, and we really provide infrastructure solutions for our customers, think sort of endpoint across a network to a data center, whether that's enterprise data center, cloud solution, application and security across the top of that. And certainly, AI and agentic workflows in that environment are a big area of focus.

But one of the things we're working with our customers is that the infrastructure in most customers has a lot of technical debt that isn't prepared to support AI investment. And so while there's a lot of talk about the large language models and the agentic capabilities, the reality is that a lot of CIOs are working with some foundational infrastructure upgrades they need at the desktop, at the network, to really be able to support some of those agentic models. So we're seeing a lot of focus on that refresh cycle.

To your point, a lot of these agentic workflows involve AI agents that will consume a massive amount of data, and maybe everything from my existing laptop to the fundamental wireless network just isn't really designed to consume that amount of data and networking bandwidth. And the other side of that thing is that it's spiky. It's unpredictable, so I don't know when these AI agents are going to kick off those types of workloads.

So do I have to be careful about maybe limiting the amount of investment I have in agentic AI because I don't have the infrastructure, or what are people thinking through here? Yeah. It's really an and, not an or.

But I think I read a study recently that 40% of agentic workflows, sort of projects that are currently in process are going to be scrapped because they failed, and they didn't fail because of the agentic capability or the large language model. They failed because the infrastructure wasn't capable of supporting it, or there wasn't proper governance around sort of the identity and management of those agents. And you mentioned some of the infrastructure, really two primary drivers.

I think Gartner said that by the end of 2026, every new PC that's deployed will be an AI PC, meaning it has NPUs. And NPUs allow you to do local processing of AI activities, and that's important for a couple of reasons. Number one, it limits that amount of network traffic that needs to go back and forth and sort of de-stresses the network.

And number two, NPUs are much more efficient at processing AI data than GPUs, which is really the big driver in the hyperscalers and the neo clouds, and where all the trillions of dollars of data center investment is going. So I think that the refresh of the endpoint across customers is a huge piece in getting those AI-enabled laptops, PCs in place. And then you touched on it from a network perspective, most enterprise networks are architected for a human-to-platform type of a communication.

So it's a lot of north-south traffic, it's very spiky, and it's generally asymmetrical. So it's basically a download model where user goes out to a cloud and application and is pulling down data, and that happens sort of very sporadically. In an AI environment, it's a lot of agent-to-agent communication, which is much more continuous, and it's much more east-west, and it requires a lot more latency sensitivity because these requests continue to build on themselves.

And so if you don't have a network that's specifically architected to support that type of traffic, your AI application's not going to work. The other thing I hear people struggling with is the data management side of this equation, and that relates to the security conversation because as this comes together, people are starting to realize, well, yeah, no matter how many guardrails I put together for the AI agent, the AI agent is kind of programmed to accomplish its mission by any means necessary. And so they will explore any kind of weakness in our system to get at the data that they're trying to get at, and the end result is they're kind of exposing all these issues we have with lack of controls and things we never did many, many years ago that we probably should have.

So are people going to have to revisit their entire data security strategy? 100%. I actually watched one of your prior podcasts on zero trust.

And sort of identity management of individuals has been a challenge for the industry to date. Right now, there are 144 agents for every human operating in the enterprise. Every one of those agents has an identity.

Every one of those agents not only interacts with humans, but they interact with each other. So identifying who those agents are- Putting role based guardrails around each of them and being able to limit their capability. I hear a great anecdote.

I was at a Cisco executive conference, and one of the sales leaders stepped up and talked about this great experience he had with Claude in his personal life. So it was his anniversary, and he asked Claude to go book a reservation at this very fancy restaurant so he could impress his wife. Well, Claude went online and the online reservation system was down.

So Claude went and built itself a voice bot, called the restaurant, and booked the reservation. Right? So to your point, they're going to use everything available to them to accomplish the task.

They lack judgment. He thought that was a great story. " Right?

We need to be able to identify and put guardrails and not allow these agents to just go rogue and go willy-nilly and get access, but really understand. And a big part of this is just the ability to log and audit every one of those transactions, understand who is the agent, what are they doing, and how do we ensure that it's sanctioned activity and that they're not going rogue. The other side of that too is that the AI agents are going to communicate with each other, and some of these AI agents will inherit our permissions that we give them as the end user that created them.

But others are going to be either autonomous or more semi-autonomous, and they'll be just independent entities that are sitting there that have been deployed to manage some task somewhere. And it's not clear to me that people have an understanding of how to apply and manage identity to an AI agent, because it's neither fish nor fowl. Right?

It's not human, but it's not a traditional non-human identity either. Yeah. No, that 100% is the challenge in the security space as we go forward.

And I think as we deal with CIOs, and this is a budgeting conversation and where are you going to invest budget next year. Number one on the priority list is in that security space, in the identity management. I think everybody wants to take advantage of the massive productivity gains that are available by building out these agentic agents, but nobody wants to risk the enterprise in the process.

And so I think priority one in terms of investment is investing in the right security infrastructure so that you can know what agents are being deployed in your environment, what are they given scope to do, how do you audit and ensure that they don't, so that you don't have agents building agents and deploying additional agents, right? Yeah. So that is 100% the crux of the security concern and, again, an area of required investment as a number one priority as we think about 2027 budgets.

Now most CFOs that I know, despite the rise of AI, are not exactly going to be thrilled about signing off on double-digit increases on IT budgets and probably will not. " Which means that I got to go in and kind of make the whole IT architecture more efficient so I have money left over to fund the AI. So is that part of what we're trying to do here?

100%, right? " I do think the promise of some of these agentic capabilities to be able to retire some of your legacy run rate spend, and be able to invest is a virtuous cycle. And yeah, at times it's a bit of a chicken and the egg.

But for us, sort of the three priorities that we see are, as I said, the identity management and security. Number two, it's refreshing your core infrastructure, PC and network in particular, to be able to take advantage of this. Three, it's really treating your GPU utilization as a sourcing strategy, not a capital strategy, and figuring out the right mix of neo clouds and hyperscalers to be able to go and understand sort of the economics of the token environment that's now driving.

And in doing all of that, drive down your run rate expense so that you're actually self-funding some of this investment. And most organizations have a natural refresh cycle coming up on PCs and network. It's sort of in the standard cycle.

And so you get a bit of a two for one. You're not just refreshing for the sake of upgrading. You're actually refreshing to enable and transform into an AI-enabled infrastructure.

So as we kind of walk through that a little bit, how much are people going to continue to buy and kind of manage themselves, and how much will they eventually ship more to consuming or relying more on services, AKA managed services that you guys provide? Because historically, I think maybe it was kind of 20% of spending went to services, and a lot of it was just internal IT budget for people mainly. Mm-hmm.

But is that going to flip, and are we going to finally get to a point now where more people are going to look at IT and security more as a service that I consume rather than a thing that I hire a bunch of people to staff? " So, managing the infrastructure becomes something that you can have a service provider like NWN provide and sort of operate that environment, and then putting their human resources, their skilled creative resources on the how do I build agentic workflows over the top of that to be able to really enable my business to take advantage of all these productivity gains that are available? Do you think also that we might have to revisit the hierarchy of IT?

And by that, I mean today we have all these silos, and there's somebody who's the database specialist, storage specialist, and so on and so forth. But as I look at the way AI kind of operates, maybe the time has come to kind of rethink how those silos are functioning and maybe collapse them or integrate them. Or is there a movement afoot to kind of rethink how we manage IT from the ground up?

There absolutely is, Mike. I think NWN goes to market and organizes around seven core offer areas, and those offer areas roughly align to those technology silos. You've got endpoint computing, you've got network infrastructure, there's data center infrastructure, server and storage, cloud-based services.

We focus a lot on communication applications, so voice, video, contact center that companies use, and then certainly security as a technology that spans those. The layer of services that companies are looking to us to apply are... They're vertically focused on those technology stacks, but they're horizontally focused as well on the user experience of the user sitting at a laptop, going across infrastructure to get access to data to be able to go deliver some outcome.

And so where AI becomes so powerful, and as the market moves broadly from a telemetry-based managed services model, where you're pulling discrete data off of a thing and taking action based on it, to much more of an observability and insight type of a model, where you're now pulling data and insights off of the entire managed infrastructure, but applying a layer of AI inference to understand not only what's happening vertically, but how that impacts things horizontally and how you actually operate that environment to be able to support the agentic workflows and the business outcomes that are driving over the top of it. So we absolutely see a desire to break down the silos and really operate more horizontally across those infrastructures to be able to deliver the business outcome. As the EVP for customer success, what do you see customers doing these days that just kind of makes you shake your head a little bit and go, "Folks, I wish we could be a little bit smarter than that"?

Some of it is continuing to invest in a siloed fashion, right? And different organizations were organized in different ways and continue to make investment decisions based on what I would call legacy thinking, about how you run a legacy IT infrastructure. And I think a lot of it has to do with where you engage at a customer's organization and who has a more strategic purview to the strategies of the customers and the outcomes they're trying to deliver versus maybe a department level leader who's still focused on protecting a way of doing things that has sort of been the way they always do it.

And so, for us, it's really about articulating with the customer what are the business outcomes you're looking to address, and then how do you design technology solutions to deliver those business outcomes versus the historical technology for technology's sake and sort of investing in these different vertical silos. To your thought about that, do you think, are you seeing people become more proactive about all this and forward-thinking, or is this going to be another instance where we all go to the school of hard knocks and learn all this stuff the hard way? Mike, I think it, like anything, it varies across our customer base.

NWN serves 6,000 customers across North America, and they really run the gamut on that sort of either maturity curve or risk-reward continuum, however you sort of choose to look at it. We've got a number of customers who are early adopters and who are absolutely changing their thinking, who are investing in sort of AI-forward initiatives, and who recognize that AI is not a thing unto itself. It's dependent upon infrastructure, and you have to build and rethink the way that you operate.

So bolting on an AI chatbot to an existing process can get you maybe an incremental gain, but it's not going to transform your business. Redesigning your business around agentic capability is really where the gains are, and that's a big leap for a lot of customers, right? Change is hard.

Change management is hard, especially in established businesses. So many of our early adopter customers are out on the bleeding edge of that, and we're working with them to help them redesign. As the maturity curve grows, we'll have that sort of major middle part where more and more customers, it becomes more mainstream, and there'll always be the laggards, right?

The folks that are less willing to take the risk or less willing to make that change. But as I said, we serve 6,000 customers. We do a lot in highly regulated industries, so state and local government, healthcare, federal government, probably 40% of our business in that area, 60% in the commercial space.

So customers are taking different views across those different segments. All right. Well, folks, you heard it here.

Look, AI is going to change everything. There's nothing out there that AI will not impact, but perhaps the worst thing you can do, to Jack's point, is if you're engaged in legacy thinking, it isn't going to work out so well. Jack, thanks for being on the show.

I appreciate it, Mike. Thank you. All right.

And back to you guys in the studio.