Here is a problem that is sitting quietly inside a lot of finance and IT conversations right now, and I think it is going to get significantly louder over the next twelve months. And I want to be clear upfront, this is not just a large enterprise problem. If your organisation has cloud spend and is deploying AI agents, this conversation is relevant to you, whether you have 200 people or 20,000.
Most organisations signed their cloud agreements before they had any meaningful picture of what running AI would actually cost. Whether that is an Azure Consumption Commitment, an AWS pricing deal or a standard Microsoft 365 subscription, those agreements were sized against known workloads: compute, storage and SaaS licensing. Predictable, relatively stable, manageable. The way most organisations govern technology spend was built for that world. It is not built for this one.
Three things are happening simultaneously, and almost nobody is treating them as a single problem.
The first is agent proliferation. Organisations are not deploying one AI tool anymore. They are deploying many, copilots, autonomous agents, data processing pipelines and inference workloads running across whatever foundation model the business decided to try last quarter. This is as true for a 300-person professional services firm running Microsoft Copilot and a couple of automation agents as it is for a large enterprise with a dedicated data science team. Each deployment has a consumption model that is fundamentally different from a traditional SaaS licence. Token usage, inference compute and training cycles. The costs are real-time, variable and in many cases sitting in departmental budgets with no central visibility. According to the FinOps Foundation’s 2025 State of FinOps report, the number of organisations now actively managing AI spend doubled in a single year, from 31% to 63%. That doubling is not a sign that people have got on top of it. It is a sign that the problem arrived faster than anyone expected.
The second is the committed spend collision. Average monthly AI spend jumped 36% year on year between 2024 and 2025. That spend is landing inside cloud agreements that were structured with completely different assumptions. The people who negotiated the cloud deal are often not the same people deploying the agents, and the governance model connecting the two frequently does not exist at any size of organisation. A mid-market business that signed a three-year Azure agreement eighteen months ago almost certainly did not model what Copilot, Azure OpenAI and a handful of autonomous agents would add to their monthly consumption. Finance is looking at one number. IT is looking at another. Nobody is looking at both at the same time.
The third is the visibility gap. Only 43% of organisations track cloud costs at the unit level, according to Gartner. Most cannot tell you with confidence what a specific AI workload is costing per business outcome, let alone how it maps against committed spend. And yet CloudZero research found that 91% of organisations feel confident in their ability to evaluate AI ROI. That gap between confidence and actual visibility is where the surprises live, and they are not always pleasant ones.
The tooling to address this is maturing. IBM’s Apptio and Kubecost, alongside a new generation of platforms built specifically for AI and data cloud cost management, are bringing the kind of real-time cost intelligence to AI workloads that traditional FinOps brought to compute and storage. But the tooling is only part of the answer. The Flexera State of the Cloud report found that 84% of organisations still struggle to manage cloud spend, and AI is adding a layer of complexity that most existing approaches were not designed to handle.
So what should any organisation, of any size, actually be doing right now?
Three practical things. First, map your AI workloads against your cloud spend position before your next renewal, not after. If your Azure or AWS agreement was sized twelve or eighteen months ago and your AI consumption has grown since, you are either heading for an overage or you have committed spend sitting underutilised that could be working harder. Either way that conversation is better had now.
Second, establish clear ownership of the connection between what the business is deploying and what the organisation is financially committed to. In larger organisations this means expanding the FinOps function explicitly to cover AI. In smaller ones it might just mean one person owning both conversations. The size of the organisation does not change the need, it just changes the scale of the solution.
Third, if you are running data platforms like Snowflake or Databricks, treat cost governance as an architectural decision rather than an afterthought. Their consumption models mean costs can escalate faster than almost any other technology investment. Building visibility and controls in from the start is significantly easier than retrofitting them later.
The honest reality is that most organisations, large and small, are behind where they need to be on this. The AI investment decisions were made quickly, which was probably right. The financial architecture around those decisions has not kept pace, which is a problem that compounds over time.
That is the convergence worth paying attention to. And it is not waiting for anyone to catch up.
As organisations continue to scale their Al initiatives, bridging the gap between deployment speed and cost governance will be critical to achieving sustainable, long-term value.
Want to learn more? Connect with one of our specialists:
Fabienne Porquet, Marketplace Sales Specialist – SCC UK Sales Software
Author: Andy Dunbar, Managing Director, Software & Security (UK)
We also recommend reading The Real Challenge Behind ServiceNow’s New Licensing Model