What could boost enterprise spending on network equipment? That’s a question that network vendors obsess over, but one that is (or should be) just as important to any vendor who wants to sell to enterprises. There’s also a corollary question, which is “Have tech vendors gotten focused on consumers, and if so what will that mean for the future?” Let’s try to deal with both questions.
For the first question, almost every enterprise (of almost 500 I’ve gotten comments from) says the same thing; “New business cases”. Enterprises have always broken down IT/network spending into two groups, sustaining that they already have and supporting things that are new. The former group, enterprises say, is all about doing more for less, period. The latter is what spawns real changes in spending, because it brings a new set of benefits that can justify higher costs.
The specific current poster child for new business cases is AI, of course, but there’s a problem here that introduces our second question. Enterprises do really believe that AI could result in a significant new spending spree, both on network equipment/services and on the other IT areas. However, they do not believe that the classic copilot or chatbot models of public AI services will do that. They think AI agents will, but they’re not yet fully confident. Why? Because almost nothing useful to them seems to be talked about in the agent space, because it’s too complex and too specialized.
Everything in tech publicity is driven by clicks, so there’s a strong incentive for a publication to push stories that generate them. That encourages populism and simplicity. Populism means that a lot of people find it interesting, and simplicity means that most will understand them. Simplicity also makes it easy to write the stories, which is important if you want to push out a constant supply of new material. AI is interesting, but what drives populism with AI is making it interesting to a lot of people. That’s not the case for any sort of AI that’s hosted rather than delivered as a service, and not the case for any sort of AI that has to be created by the user, wherever it’s run. AI agents that cull data and offer insights are business tools, not personal tools, and so it’s hard to get them really covered well. That means that companies have to muddle through their own learning curve just to figure out whether AI is helpful, and what kind of it is best, and how much it will cost and save. Then they have to work through it, largely on their own in many cases. They’re doing that, but it can’t happen fast enough to be a 2025 driver of growth for vendors.
Even in 2026, it’s not clear just how much AI will push up network or IT spending. The reason is that enterprises are still working through just what the business cases will look like. The value of AI in making better business decisions isn’t exactly intuitive, and running experiments to determine it requires actually deploying AI agents in realistic scenarios, difficult given a lack of useful guidance in that areas.
There have been some successes with AI agents, and the enterprises who’ve succeeded with the concept seem to have followed a common path that, regrettably, hasn’t percolated through the markets overall. To make agents work, you need to organize a business in the IT domain, provide a context to which agent pieces can attach. In some cases, this has actually involved creating a digital twin, and in others some form of graph or state/event relationship. This implies that the best AI agent applications are dealing with real-world, real-time, behaviors.
This may be the key to the whole “make enterprises spend more” story in a nutshell. AI is good at complexity, something humans can handle if they have the time. In real-world situations they may not have it, and not only that, real-world situations often include variables that are…well…variable. Human analysis of the current state of a real-time system not only takes time, it relies on the factors that determine state being stable through the period of analysis and use, which they probably won’t. The advantage of contextualizing AI agents to handle the real world, then, is that if it’s done right it can actually accommodate the real world’s disorder.
This real-world stuff also introduces a whole category of business case, empowering the 40% of workers who are out in the real world rather than sitting in an office. This group’s productivity, measured by their unit value of labor, is actually higher than that of the office group. The problem with this group’s empowerment is that you need real-world context to assess what they need and how to deliver it, which means sensors of various types and a way of organizing their information. Empowering office work has generated all the business benefits that justify current IT, and if pulling in the 40% only matches that, imagine what it could do to IT spending.
There’s plenty of money on the table to enrich network vendors, and IT providers too. The problem is that realizing it means doing a lot of work on the business side, and few vendors are willing to do that. IBM has really been the only one who shows up in my chats as an evangelist for the “business AI” model, and even IBM has been slipping a bit into the “personal AI” story to get its share of the ink that the tech and financial world dispense. That’s only impacted their PR persona, though; enterprises still say their sales teams know AI in business terms.
That demonstrates two things, I think. One is that it’s possible to position your AI to, as an old political book once said, “Stand tall in Georgetown”, meaning with the media, and still keep yourself grounded in reality. The other is that it’s hard for any AI provider to work to develop the business side unless they have a lot of product in play that would be promoted by such a move. All market education tasks are risky, but absent a clear path from a smart prospect to a committed buyer, they’re suicidal.
You might think I missed some driver here. Why wouldn’t enterprises look to reducing their own opex as a driver for network spending? Well, I did get detailed comments on this from 42 enterprises, and that’s going to be the topic of my next blog.
