Your AI spend is out of control. You just don’t know it yet.
Most executives have never seen a complete inventory of the AI tools their organization actually runs. The ones who have describe the same reaction: surprise, then concern, then a grudging acknowledgment that it makes sense.
Last week, I wrote about Shadow AI — the difference between the AI tools companies approve and the ones employees actually use. And this gap between what is approved and what’s in use isn’t just a security problem. It is also a financial one. Frankly, it’s growing faster than most budgets are designed to keep up with, so leave catching it aside.
Understanding and managing AI Spending is crucial for organizations to maximize their technology investments. By prioritizing effective AI Spending, businesses can ensure they are getting the most value from their AI tools.
So this week, let’s talk money.
The bill you didn’t budget for
Here’s a number worth sitting with: organizations spent an average of $1.2M on AI-native applications in 2025 — a 108% year-over-year increase. That figure captures sanctioned spend. The unsanctioned spending is harder to count, but the pattern is consistent across organizations of all sizes.
That multiplier isn’t an outlier. 78% of IT leaders reported unexpected SaaS charges due to consumption-based AI pricing models — up from 65% the year before. The pattern is predictable: a vendor offers generous pilot credits, a team adopts the tool, usage scales to production, and the invoice that arrives bears no resemblance to the budget line that approved the pilot.
Vendors lure customers with generous pilot credits, yet scaling to production routinely reveals 500–1,000% cost underestimation. That’s not a rounding error. That’s a fundamentally broken budgeting assumption baked into how most enterprises still think about software costs.
Why consumption pricing breaks enterprise budgeting
Traditional enterprise software was predictable. You bought seats. You knew what you’d pay. You negotiated at renewal.
AI doesn’t work that way. Most AI pricing today is consumption-based — tokens, API calls, inference runs, or hybrid models that layer usage fees on top of base subscriptions. Consumption models were designed to align cost with value. CIOs and CFOs discovered they were also deeply unpredictable, which is the thing enterprises hate most about any cost line.
“When you add AI and consumption-based pricing, we’re talking about more budget volatility and pressure on in-year spend, which kills innovation.”
– Jez Back, Cloud Economist & Global Offer Leader, Capgemini.
The problem compounds when you factor in how AI spending actually flows through an organization. In most enterprises, AI costs are split across IT and business units. Teams swipe corporate cards for tools, then ask IT to “make it compliant” after the fact. Sound familiar? Remember all those Slack instances in your company? Cloud and infrastructure costs land in shared accounts, vendor overlap accumulates as multiple teams buy similar tools, and no one so far has a single view of total AI spend.
This isn’t a technology problem. It’s a governance and operating model problem.
The SMB trap is worse
Large enterprises at least have IT departments asking the right questions, even if they’re not always getting good answers. For SMBs, the picture is darker.
Without a dedicated procurement function, AI tool adoption tends to happen entirely at the team level — a marketer expenses a content tool, a developer adds a coding assistant, a sales rep subscribes to an AI prospecting platform. Each decision is defensible in isolation. Collectively, they create a cost structure that no one in the organization has mapped, governed, or negotiated.
And unlike a traditional SaaS subscription, these tools don’t just sit dormant if underused — some actively process data in the background, generate outputs, and rack up inference costs that show up as a surprise line item weeks (and sometimes months or quarters) later.
What a real AI cost model looks like
The organizations that have gotten this right have stopped treating AI as a collection of individual tool purchases and have started treating it as an operating capability with its own financial architecture. A few things that actually move the needle:
Build a single AI spend view. Consolidate across procurement, cloud billing, and headcount. Manage AI costs as a portfolio, not a series of isolated line items. Most organizations don’t have this picture. Building it is the prerequisite for everything else.
Push vendors toward hybrid pricing. IT leaders who’ve been through a consumption billing shock are pushing for hybrid pricing – fixed base fees with variable usage charges. The only model that balances flexibility with predictability. It can be a negotiable contract term, but only if you ask for it before signing.
Stage funding with kill/continue gates. Rather than funding a full AI transformation upfront, run smaller commitments with defined evaluation criteria. Adoption metrics, unit economics, and security review should all be checkpoints before the next tranche flows. This is basic product thinking applied to AI investment. Experienced product leaders are well-positioned to drive.
Inventory before you audit. You can’t govern what you can’t see. Before any policy or cost-optimization effort, organizations need to understand what’s actually running — including tools IT didn’t approve and infrastructure that got stood up outside normal procurement channels.
The financial reality of AI in 2026 is this: the chaotic, bottom-up phase of adoption, where anyone with a corporate card could spin up an AI tool, should move towards extinction. Budgets are being centralized under CIOs, who are now creating fixed pools of compute spend for entire organizations. The companies that get ahead of this transition will have leverage. The ones that wait will face a very uncomfortable reconciliation between what they thought they were spending and what they actually spent.
Shadow AI is a data risk. It’s also a 3–5x budget overrun waiting to happen. The governance conversation needs to include the CFO — not just the CISO.
Part 1: From Shadow IT to Shadow AI
Part 3: Risks of Agentic AI in Your Workflow
