AI Consumption Billing Breaks Traditional IT Budget Models

The predictability problem

For decades, enterprise IT budgeting relied on a simple formula: count users, multiply by per-seat license costs, add infrastructure and support. The model worked because technology costs were fixed—a Microsoft 365 license cost the same whether an employee sent ten emails or ten thousand.

AI consumption billing has shattered that predictability. Organizations now face variable costs that scale with usage intensity, not headcount, creating budget forecasts built on data that simply doesn't exist.

The shift affects vendors across the board. Microsoft's E7 license bundles a $99 per-user base price with additional consumption charges for agent execution and Security Compute Units. OpenAI, Anthropic, and Google bill by tokens processed. Zoom's ZoomMate uses AI credits. Salesforce Agentforce charges per agent action, while ServiceNow meters "assists."

Why it matters

This isn't just a procurement headache—it's a structural misalignment between how organizations budget for labor and how they're paying for AI that replaces or augments it. When Finance deploys an AI agent to process invoices, IT typically carries the consumption cost despite having no control over usage intensity. The business unit gains productivity while IT defends unpredictable bills it didn't generate.

The forecasting vacuum

Early evidence shows even sophisticated technology companies struggle with AI budget forecasts. Uber exhausted its entire 2026 AI coding budget by April after giving 5,000 engineers AI tools in December, according to reporting first published by The Information. The company responded by capping token spend at $1,500 per engineer per month.

Amazon reportedly spent $1.8 million on a failed Claude Sonnet project—an 860 percent budget overrun that took five months to detect, the Financial Times reported. When GitHub Copilot moved to usage-based credits in June 2026, some heavy users saw bills jump from $29 to nearly $750.

"In April and May, I started hearing from companies: 'Oh my god, we are 3x over our entire 2026 token budget and it's only April,'" FinOps Foundation Executive Director J.R. Storment told TechCrunch.

Consumption depends on task complexity, model selection, context length, agent steps, and user behavior—variables with no historical baseline. Unit token costs have fallen, but total spending climbs as organizations deploy more ambitious autonomous workflows.

The cap creates new problems

Spending caps prevent surprise invoices but convert financial problems into operational ones. When an employee's credits run out mid-month, do they miss deadlines set assuming AI assistance? File IT tickets and wait? Expense personal AI subscriptions, creating shadow AI?

The comparison to human labor is instructive: organizations don't send productive employees home halfway through the month. Hard credit caps do exactly that to AI-augmented workflows.

Practical steps forward

IT leaders entering 2027 planning cycles face budgeting without benchmarks. Several approaches can help:

Run instrumented pilots with detailed per-user metering over 90 days to generate consumption data before broad deployment. Budget from measured usage, not vendor calculators.

Present AI spending as ranges—base case, expected case, high-adoption case—rather than single numbers that imply false precision.

Negotiate contractual protections including caps on credit ratio changes, credit rollover provisions, and fixed pricing for business-critical workflows.

Assign consumption ownership before deployment. Decide whether AI costs are centrally held or charged back to business units, and document the process explicitly.

Define exhaustion policies in advance: what happens when users or teams run out of credits mid-cycle, and who has authority to approve additional capacity?

Treat quarterly re-forecasting as mandatory, not optional, in a market where usage patterns and vendor pricing both change rapidly.

These details were first reported by No Jitter in a comprehensive analysis of AI budgeting challenges facing enterprise IT organizations.