I’ve spent more than 20 years in enterprise technology, running data centers for major financial institutions, building products at enterprise software companies, working across cloud infrastructure, DevOps, and observability.
I’ve watched a lot of technology waves roll through the enterprise. And in my experience, the ones that catch organizations flat-footed are never the ones nobody saw coming. They’re the ones everyone saw coming and still didn’t prepare for.
Agentic AI is shaping up to be one of those waves.
Chief Product and Technology Officer at Apica.
The conversation in enterprise IT right now is almost entirely focused on the models, which one to choose, how to govern its outputs, whether the ROI is real.
What’s not getting nearly enough attention is what happens to your IT infrastructure when those agents start running. Specifically, what happens to your telemetry data.
Recent Omdia research surveying more than 300 enterprise IT decision-makers across North America and Western Europe should make any CTO uncomfortable.
The explosion is already happening and most organizations are already behind
Start with what’s happening right now, before agentic AI fully kicks in. More than half of the enterprises surveyed reported that their telemetry data volume more than tripled in the past 12 months, with an average increase of 3.7x. AI and ML workloads already account for approximately 43% of that growth. This isn’t a future projection; the majority of respondents are experiencing exponential data growth today.
The forward-looking numbers are where it gets genuinely alarming. Enterprises anticipate a 9.5x average increase in telemetry data from agentic AI workloads within the next two years. That’s the average; nearly half of respondents expect growth somewhere between 6x and 100x. Omdia’s senior market research analyst Keir Walker put it plainly: “This is just the beginning of the growth curve. We are at the beginning of the hockey stick.”
When I say most organizations aren’t ready for this, I’m not editorializing. Roughly two-thirds of organizations describe themselves as only “somewhat prepared” or worse for the data volumes agentic AI will generate. Twenty-two percent haven’t even begun to consider the data implications. Meanwhile, 35% already report widespread agentic AI deployment. The gap between deployment ambition and infrastructure readiness is not a small one.
The cost problem is already worse than most people realize
Here’s the number that stopped me when I first saw it: 69% of organizations actively running agentic AI projects report that their observability costs now exceed their compute and infrastructure costs. The technology meant to support AI has become the single largest line item in the AI budget.
Average annual enterprise observability spend is sitting at $3.17 million, growing at 28% year-over-year. Thirty-five percent of enterprises are seeing budget increases above 52%. Nearly one in five is already spending more than $5 million annually just on observability. Eighty-one percent are actively looking for ways to cut those costs.
And the consequences of not managing this aren’t just financial. Fifty-nine percent of organizations have already terminated or delayed at least one agentic AI deployment because monitoring costs were too high. These aren’t experimental projects being shelved. When observability costs become unsustainable, organizations pause the AI agents responsible for cybersecurity monitoring, compliance, and fraud detection. 27% and 26% of respondents cited exactly those exposure areas as direct consequences of paused or cancelled deployments.
Think about what that means in practice. A company decides it can’t afford to monitor an AI agent handling fraud detection, so it gets shut down. That’s not a technology problem. That’s a risk management failure.
The infrastructure assumption that’s now broken
The observability infrastructure most enterprises rely on today was built for a different era: Predictable data volumes, manageable costs, centralized platforms. Agentic AI breaks all three of those assumptions at the same time.
The fundamental issue is architectural. Most observability platforms are built to ingest everything and then sort it out. That model works when data volumes are relatively stable. It doesn’t work when you’re facing a 9.5x surge in telemetry from autonomous AI systems that are constantly reasoning, acting, and generating signals. You can’t just throw more storage and more backend capacity at this problem. The math doesn’t work, and the budget won’t survive it.
What the organizations pulling ahead have figured out is that you need an intelligent layer between your telemetry sources and your observability platforms, one that routes, filters, enriches, and governs data before it ever hits a costly backend. This isn’t a new concept in infrastructure terms, but the urgency around it is new. Omdia’s research found that enterprises with a telemetry pipeline in place are 50% more likely to be prepared for 6x or greater data growth, and those with mature pipeline deployments are 80% more likely to have avoided the operational cost challenges still constraining organizations in earlier stages.
The pipeline isn’t a workaround. It’s the architectural foundation that makes continued observability investment viable at agentic AI scale. As Omdia principal analyst Torsten Volk put it: “Scalability is really the underlying issue for everything, for security, for cost, for rolling out projects. I don’t see any point of diminishing returns or any slow downward trend in data volume growth.”
The window to act is narrower than most think
Ninety-seven percent of organizations have either implemented or are actively considering a telemetry pipeline solution, and 70% are in active evaluation cycles right now. This market is not moving slowly.
One detail from the Omdia research that I think gets overlooked: Nearly one in four respondents say existing vendor relationships are not a significant factor in their purchase decision. That’s a more open competitive environment than most people assume. The capabilities buyers say they want most are multi-destination routing, data reduction and sampling, and content-based routing; all of which point toward pipeline-first architecture rather than expanding incumbent observability platforms.
The data represents a decision point, not a warning. The organizations that solve the infrastructure problem before it becomes a budget crisis are the ones that will scale agentic AI on their terms. The ones that don’t will find themselves choosing between monitoring their AI deployments and affording them.
That’s not a choice any organization should have to make.
We've reviewed and ranked the best cloud storage.
This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.
The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit