Enterprises, like telcos, face a world dominated by AI and other hype, while their own technical reality is still pretty pedestrian. There are some points of congruence between hype areas and real planning focus, but even in those areas the influences driving enterprise network and IT planning are more complicated than responding to a new tech wave. That’s what makes it worthwhile to have a look at them.
I got 288 enterprise comments on tech infrastructure planning and spending so far this year. All these included discussions of their “driving priorities”, as one put it, and none of the popular hype targets, most notably AI or the cloud, were cited by more than a quarter of the group as primary, much less as the top, considerations in network and IT planning and deployment. However, over half said that both the cloud and AI were factors to weigh in the future, and a third said there would be some impact on this year’s budgets.
You can probably guess enterprises’ top priority; cost management, cited first by 217 of the 288. Second, cited by 46, was improved QoE. Optimizing the cloud was cited as top driver by 17, and AI by 8. I can’t give you reliable data on what percentage of data center switching budgeting was driven by AI, but surely nothing like the article I cited earlier would suggest. Data center switching changes were budgeted by 199 of the 288, though, and it was the top network equipment spending proposed. All 25 enterprises who listed AI or the cloud as the driving priority expected to spend more on data center switching, showing that where there was significant planning emphasis on either area, it drove significant changes in the data center and its network.
Outside the data center, things are mixed. Of the 288, 54 said they expected to spend on edge-LAN upgrades, most in their main site. Another 52 expected to spend on SD-WAN, with slightly more than half that group already SD-WAN users. Only 17 said they were budgeting for edge routers to attach to VPNs other than SD-WAN. None indicated they were budgeting for more WAN traffic due to cloud usage or AI.
If these numbers surprise you, let me repeat a point I’ve often made. Surveys are largely useless because people don’t tell the truth on them. Many will shade their responses because it looks better to say that you’re doing cloud or AI planning when asked what’s driving your 2025 budgets, even if the questions aren’t structured to favor that answer (which, regularly, they are). But let’s continue, keeping in mind that the data I provide is an analysis of spontaneous comments, which I feel are a better measure of reality.
Looking explicitly at network spending, only 8 enterprises mentioned WiFi 6 or 7 in their comments, and in all these cases the improved technology wasn’t a driver of the upgrade, but just a beneficiary of an upgrade driven most often by a larger number of users or higher rate of bandwidth consumption per user. Enterprises agree the more modern standards are best if you are upgrading, though.
In the SD-WAN case, of 288 enterprises, 221 said they were under pressure to lower their WAN service costs, but only 148 planned to attempt that, including the 52 who had budgeted for new or increased use of SD-WAN. Another 57 said that they had assessed SD-WAN but would not likely adopt it unless it was offered by a credible (preferably a current) operator rather than something they had to create themselves. Of the 52 with specific SD-WAN plans or deployments, 5 were based on an operator-provided service. IMHO, this means there is a significant interest in SD-WAN services, and that there’s a barrier to roll-your-own SD-WAN models.
In AI networking, 6 of the 8 said that they were deploying a “small AI cluster”. Only four offered any comment on cluster size, but all of these planned less than 100 GPUs. Interestingly, all of these were associated with already-underway deployments. Of 122 who said they expected to deploy AI down the line, the majority believed they’d use “agents” that were described in a way most consistent with small-model AI or even basic ML. Only 12 expected to deploy large-scale (LLM) generative AI, and this group also (interestingly) mentioned less-than-100 GPU limits. Of the 122 who expected to deploy AI, the primary reason (held by 98) was data sovereignty, with another 25 saying they didn’t want the usage-price-AI-service hanging over them. The impression I get from AI plans is that enterprises are still trying to come to terms with AI adoption costs and benefits.
Within the 288 enterprises, 190 said their network and IT spending was going to be focused on sustaining current applications and missions. Only 11 said they had any new applications in mind; the rest made no comment on the topic. Of the 11, two said that the new stuff would account for more than fifteen percent of their total IT spending, five said it would account for less than ten percent, and the others didn’t characterize it. This shows that “project spending” continues to run far below historical levels (over the last 40 years, it’s averaged 37% of total IT spending). Of the 288, 38 said that “productivity projects” would be the only source of new money in IT, and the remainder didn’t even mention the topic. Only two enterprises suggested what new projects might be, and both mentioned IoT.
What’s surprising here is that many more (122) enterprises indicated they expected AI adoption, yet none suggested AI would be a source of productivity projects. Does this mean enterprises don’t see things like the AI copilot applications as AI adoption, or as a valid target for increased spending, or as a way to improve productivity? I can’t tell from the comments, but I have my own idea.
Which is? Which is that enterprises haven’t really figured out AI. It gets good ink, as we say in the media business, so they believe that all the cool, upstanding, people are looking at it, which of course means they are. But this attitude, back in the 1980s, resulted in a third of enterprises telling me they used gigabit Ethernet when there were no commercial products available. The generative AI model doesn’t require planning, insight, or really anything but interest, and so it’s been the dominant AI approach. However, will enterprises pay for that? They tell me that they won’t, and if that’s true then everyone who believes in AI will need to help enterprises work out its value proposition and set a path to realizing it.
