Gartner found 54% of I&O leaders are adopting AI to cut costs. But AI spending itself is heading to $2.5 trillion in 2026. The math needs to include both sides.
I was reading a Gartner survey summary earlier this year and one number pulled me up short. According to Gartner, 54% of infrastructure and operations leaders say they are adopting AI specifically to cut costs. I sat with that for a minute. Because the same Gartner April 2026 forecast (https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026) projects worldwide AI spending at $2.5 trillion in 2026, up from lower levels just a year or two ago. A market that is scaling that fast is not primarily being driven by people who think AI will save them money. Or if it is, the savings story needs to be examined more carefully than it usually is.
Let me say upfront that I find the cost-reduction logic for AI in infrastructure operations genuinely plausible in certain conditions. The category of tasks that AIOps tools handle well is real: log analysis at scale, incident correlation across distributed systems, predictive capacity planning based on historical patterns. Human operators doing this work are slower and more expensive than trained ML systems, once those systems are properly configured. That is not marketing copy. It is an accurate description of what the tooling does when it works. The labor savings are real for organizations that have the right kind of workloads and can implement the tooling without major friction.
The problem is not the technology. The problem is what the "AI saves money" framing systematically leaves out of the accounting.
Running AIOps systems requires compute. The inference infrastructure that drives real-time anomaly detection and incident correlation is not free. Training the underlying models requires GPU cycles. Maintaining the data pipelines that feed those models requires data engineering talent. Monitoring the AI systems that are supposed to reduce the burden of monitoring everything else is a real operational task. The governance overhead for deciding when to trust an AI recommendation and when to override it does not disappear just because a machine made the recommendation. These are costs that belong in the total cost of ownership calculation, and in my reading of how the "AI cuts costs" story gets told inside organizations, they are frequently underestimated or omitted entirely.
This is where the Brynjolfsson productivity paradox, even though it was developed in a very different technological era, stays relevant. Brynjolfsson's observation was that IT investment does not automatically translate to productivity gains, because the organizational changes required to capture value from technology investments are themselves costly, slow, and uncertain. The paradox showed up in the data during the PC era, when productivity statistics did not reflect the massive investment in computing hardware. The resolution, which took years to arrive, was that productivity gains required firms to reorganize work practices, train workers, and change management structures. The technology alone was not enough.
AI in IT operations has the same structure. The tooling exists. The potential for cost reduction exists. But capturing that potential requires organizational changes that are not free: restructuring operations workflows, retraining staff, building governance frameworks for AI recommendations, establishing new accountability structures for when the AI is wrong. Organizations that have purchased AIOps platforms and deployed them without those organizational investments are probably not seeing the cost reduction their vendors projected. And the ones that are seeing it may be measuring only the labor cost reduction, not the total cost of the capability they built.
The workforce dimension of this is something I find myself worrying about in ways that are not entirely comfortable to articulate in a field that is broadly enthusiastic about AI. The tasks that AIOps automates are exactly the tasks that build junior IT operations staff into senior ones. First-level log triage is not glamorous work, but it is how a new operations engineer learns to recognize patterns, develops intuition for what normal looks like in their environment, and builds the knowledge base that eventually makes them good at escalation decisions. If that work is automated away, the learning path changes. The organization ends up with a workforce that is good at overseeing AI systems but has thinner experiential grounding in the underlying operations those systems are managing.
I am not saying this is wrong. I am saying the workforce transition has costs that do not show up in the cost-per-incident calculation. And when the AI system encounters a situation outside its training distribution, which it will, the human expertise available to handle the exception depends on how much of that experiential foundation was built before the automation arrived.
Feldman and Pentland's routine dynamics framework is useful here. Their distinction between the ostensive dimension of an organizational routine (what people say the routine is, the ideal description) and the performative dimension (what actually happens in practice) captures something important about AIOps deployment. The ostensive version of an AI-driven operations workflow is clean: anomaly detected, alert generated, model recommends response, operator approves, issue resolved. The performative version involves the model flagging things that are not actually problems, operators learning to ignore certain alert categories because the false positive rate is too high, the system missing correlations that an experienced human would catch, and a gradual degradation in the feedback loops that would otherwise improve the model over time. The gap between ostensive and performative is where cost estimates go wrong.
None of this means AI in IT operations is a bad investment. For the right organization with the right workload profile, the right implementation approach, and the right governance structures, the math probably works. But "54% of I&O leaders are adopting AI to cut costs" is not evidence that 54% of I&O leaders have done the full accounting. It is evidence that the cost-reduction narrative has broad appeal. The question I would want any of them to answer before booking the savings is: what is your total cost of ownership for the AI capability, including infrastructure, data engineering, governance, and transition costs, and what is your plan for the situations where the system gets it wrong?
As an IS researcher, what I keep coming back to is that the organizational absorptive capacity required to benefit from AI in operations (Cohen and Levinthal 1990) is itself an expensive thing to build. Absorptive capacity is not just technical. It is the organizational ability to recognize valuable external knowledge, assimilate it into existing practice, and apply it effectively. Building that requires investment in people, processes, and structures that take time to develop. The organizations that will see real cost reduction from AIOps are the ones that invested in that capacity before they deployed the tooling, not the ones that expected the tooling to do it for them.
---
claims_checked:
- "54% of I&O leaders adopting AI to cut costs": "https://www.gartner.com/en/newsroom/press-releases/2025-10-29-gartner-survey-54-percent-of-infrastructure-and-operations-leaders-are-adopting-artificial-intelligence-to-cut-costs"
- "AI spending $2.5T in 2026 (Gartner April 2026)": "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
claims_unverified:
- "AIOps use case descriptions (log analysis, incident correlation, capacity planning) are general industry knowledge, not sourced to a single study"
- "Observations about junior operator career path effects are analytical, not sourced statistics"
- "Claims about organizations omitting full TCO are editorial and speculative"
sources_used:
- "https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-forecasts-worldwide-it-spending-to-grow-9-8-percent-in-2026"
- "https://www.gartner.com/en/newsroom/press-releases/2025-10-29-gartner-survey-54-percent-of-infrastructure-and-operations-leaders-are-adopting-artificial-intelligence-to-cut-costs"
word_count: 1050
About the author
Share
More notes
Related notes