The cloud was just starting to feel manageable. After years of growth and experimentation, organizations had finally brought discipline to cloud cost management. FinOps practices were maturing, cloud efficiency was measurable, and teams were aligning around budgets, accountability, and optimization strategies.
Then AI made its grand entrance.
In the newly released FinOps in the AI Era: A Critical Recalibration report from CloudZero, based on 475 responses from senior leaders across cloud-mature, AI-active organizations, one thing is clear: Artificial Intelligence has thrown the existing FinOps playbook into disarray.
Even as companies matured operationally, efficiency metrics declined. Visibility disappeared. Forecasts missed wildly. And cost management, long the domain of IT and infrastructure teams, became a challenge shared across product, engineering, and finance.
This is a look at what’s changing, why it matters, and how organizations can recalibrate for the new AI-driven reality.
The Paradox: Maturity Is Up, Efficiency Is Down
Let’s start with the headline finding: FinOps is more widely adopted than ever. Formal cloud cost programs nearly doubled from 39% to 72% year-over-year. Budget assignment hit 87%. Chargebacks climbed to 64%. Four out of five organizations now have a formal FinOps function.
And yet, efficiency declined across the board.
The Cloud Efficiency Rate (CER), the percentage of cloud spend allocated to production workloads, dropped sharply. The median fell from 80% to 65%. The top quartile slipped from 92% to 85%. Even the best performers got worse.
Why? AI is the most obvious factor. Two out of five organizations now spend $10M or more annually on AI, and nearly a third spend more than $20M.
But AI doesn’t just add cost. It changes what cost looks like. Projection accuracy is way off, with one in five organizations miscalculating their AI spend forecasts by 50% or more.
AI Costs Are Fundamentally Different
Cloud spend has historically been driven by provisioned capacity, and infrastructure decisions made by engineering teams. These are stable, predictable, and easy to track. In contrast, AI spend is:
- Consumption-based: Think tokens, API calls, and user interactions.
- Volatile: Costs fluctuate based on model selection, prompt design, and end-user behavior.
- Fragmented: AI workloads run across hybrid cloud, private cloud, hosted APIs, GPU providers, and more.
- Opaque: Billing data is scattered across systems with different formats, intervals, and levels of granularity.
In short, AI costs are harder to see, harder to predict, and harder to control. And yet, most organizations are still trying to manage them with traditional cloud tools and frameworks.
Visibility Is the First Domino
The most common challenge cited in the survey? Lack of visibility.
60% of respondents ranked it among their top three issues, with 25% calling it their number one. That’s higher than forecasting accuracy, token pricing, or even allocation itself.
Even AI-native companies are feeling the pain. Despite being the segment most invested in custom dashboards and cost observability, forecasting is deeply unreliable. More than a third (36%) of them miss their predictions by upwards of 50%.
This signals a systemic issue: organizations are budgeting for AI, but they don’t understand what’s driving those budgets. They’re watching the meter run without knowing which product, customer, or behavior is responsible.
Without visibility, there is no accountability. And without accountability, optimization is impossible.
Old Levers Aren’t Enough
The traditional FinOps toolkit includes commitment-based discounts, right-sizing instances, negotiating enterprise agreements, and using third-party resellers. These still matter of course, but they don’t solve for the biggest inefficiencies introduced by AI.
Where is the real leverage today? The survey data points to three underused, high-impact tactics:
- Code Optimization: Still underutilized at 29%, despite proven examples of dramatic cost savings.
- Customer-Level Allocation: Only 43% of organizations track AI costs by customer.
- Transaction-Level Visibility: Just 22% can trace AI costs down to specific usage events.
Engineering efficiency is particularly potent. Where commercial levers are incremental, code and architecture changes can be multiplicative. That’s especially the case in AI-heavy environments where model selection, prompt length, and token usage significantly influence spend.
Pricing Without Insight = Risk
One of the most striking disconnects in the data is around AI pricing.
Nearly nine out of 10 (84%) of organizations say they factor AI costs into pricing, primarily through cost-plus models. But only a fraction actually track costs at the level needed to make those pricing models accurate.
That means many companies are pricing AI based on assumptions or averages, not actual usage. Without customer-level data, they can’t tell whether a feature is profitable or loss-making, and whether that loss is intentional (a strategic investment) or unintentional (an exposure).
As AI becomes more deeply embedded in products, pricing on guesswork becomes a serious margin risk.
Recalibrating FinOps for the AI Era
What separates the organizations adapting well from those falling behind? The report outlines five key areas:
- AI-Specific Visibility: Monitoring tokens, models, and behavior-specific consumption patterns.
- Granular Allocation: Connecting spend to customers, transactions, and features.
- Engineering Investment: Prioritizing code efficiency over just financial levers.
- Pricing Alignment: Tying AI costs to pricing models with real data, not estimates.
- Real-Time Guardrails: Identifying overages immediately, not after the invoice.
In short, tooling and technology are only part of the solution when one looks to optimize. There’s a cultural and thinking process that needs to be applied. Organizations need to shift from managing infrastructure to understanding how cost flows through behavior, features, and value.
The Path Forward
Cloud took more than a decade to stabilize. AI entered quickly, is moving faster than cloud ever did, and introduces chaos into the equation. We see this in how AI spend is rapidly approaching cloud spend levels just three years after it became mainstream.
What’s key is this: the organizations that invest in visibility and allocation now will have the data advantage when the market matures.
FinOps isn’t broken. It never was. It’s just being stress-tested right now. It needs to evolve like any practice when times are changing. And that evolution requires tighter links between engineering, product, and finance so that everyone can see what’s happening, respond in real time, and ensure AI spend drives business value.
Speed matters when the wave hits, but being primed and ready for that wave is even more important.
Access CloudZero’s new report, FinOps in the AI Era: A Critical Recalibration, now.
Related Categories