Why Is My AI Project Suddenly So Expensive?

The price of AI tokens has been falling for two years. So why are enterprise AI budgets going up?

The answer is agentic AI — and it's changing the economics of enterprise technology in ways that most organizations aren't prepared for.

The Token Consumption Problem

Traditional AI use is relatively predictable. A user asks a question, the model generates a response, the interaction ends. Token consumption is roughly proportional to usage.

Agentic AI works differently. An AI agent doesn't just respond to a single prompt — it reasons through a problem, calls tools, retrieves data, generates intermediate outputs, evaluates those outputs, and iterates. A single agentic task can consume 10x, 50x, or 100x the tokens of a simple query.

When organizations move from AI assistants to AI agents — and most are moving in that direction — their token consumption doesn't scale linearly with the number of users or tasks. It scales with the complexity of what the agents are doing. And complexity is hard to predict.

Why Per-Seat Budgeting Doesn't Work

Most enterprise software is priced per seat. You know how many users you have, you multiply by the per-seat cost, and you have your budget.

AI doesn't work that way. The cost of an AI deployment depends on how much the AI is doing, not how many people have access to it. An organization with 100 users running complex agentic workflows can spend more than an organization with 1,000 users running simple queries.

This breaks the mental model that most finance and IT teams use to budget for software. And it creates a situation where AI costs can grow much faster than expected — not because of waste or misuse, but because the AI is working.

What CIOs and CFOs Need to Do Differently

Model consumption, not seats.

Before deploying agentic AI at scale, model the expected token consumption for each use case. This requires understanding what the agent will actually do — how many steps, how many tool calls, how much data retrieval. It's more work than per-seat budgeting, but it's the only way to forecast costs accurately.

Build in monitoring from day one.

Token consumption monitoring should be part of every AI deployment, not an afterthought. You need to know which agents are consuming the most tokens, which use cases are driving cost growth, and where consumption is higher than expected.

Negotiate consumption-based contracts carefully.

Many AI vendors offer consumption-based pricing with volume discounts. These contracts can be favorable — but only if you have a realistic model of your consumption. Signing a consumption-based contract without that model is a budget risk.

Treat AI cost management as an ongoing discipline.

AI costs aren't a one-time budget line. They're an ongoing operational expense that will grow as AI use expands. Organizations that build cost management into their AI operating model from the start will be better positioned than those that treat it as a problem to solve later.

The economics of enterprise AI are changing. The organizations that understand those economics — and build their AI programs accordingly — will have a significant advantage over those that don't.