The CFO’s guide to surviving the end of flat-rate AI

For the past two years, enterprise AI spending has behaved like classic SaaS procurement: predictable monthly subscriptions, negotiated seat pricing, and relatively stable forecasting. Finance leaders approved budgets, teams adopted tools, and everyone assumed costs would scale (more or less) linearly with headcount.

That assumption is already breaking down.

On June 1st, GitHub Copilot officially moved to usage-based AI Credits billing. Anthropic followed by restructuring Claude enterprise pricing into separate pools for interactive and agentic usage, with changes taking effect June 15. Meanwhile, companies like Uber and ServiceNow have publicly acknowledged blowing through annual AI budgets just months into the fiscal year.

KPMG recently reported that U.S. organisations expect to spend an average of $207 million on AI over the next 12 months – nearly double last year’s projection. AI isn’t simply becoming more expensive; as a cost, it’s becoming materially less predictable. And as many finance teams are discovering, the problem is the way in which the underlying billing model has changed.

Key takeaways

• Enterprise AI pricing is rapidly shifting from flat-rate subscriptions to usage-based billing, making costs far less predictable for finance teams.
• Major vendors, including GitHub and Anthropic, are already moving customers toward consumption-based pricing models tied to AI usage.
• Finance leaders now need cloud-style governance for AI spend: auditing tool sprawl, tracking usage in real time, and setting team-level guardrails before costs spiral.

The end of unlimited AI 

One of the early signals came from Anthropic. 

Its enterprise support documentation now states that legacy flat-rate enterprise seats are no longer available for new contracts, with customers transitioning toward token-based billing at renewal.

GitHub made a similar move with its Copilot plan, introducing AI Credits tied to usage consumption rather than fixed subscriptions.

And OpenAI executives have begun to frame unlimited AI pricing as fundamentally unsustainable. Nick Turley, OpenAI’s Head of ChatGPT, recently compared unlimited AI plans to an unlimited electricity plan, arguing that this model “just doesn’t make sense” as usage intensifies.

Jothan Webb, CFO at research consultancy Censuswide, echoes that sentiment: “You want to know where you’re sitting, in the same way you’d want to know where your gas bill or your electric bill is sitting… especially if it’s going to be a lot more expensive. You could be running tens of thousands or more in a month.” 

These changes don’t bode well for budgets, as Uber’s experience shows. CTO Praveen Neppalli Naga said the organisation exhausted its entire 2026 AI budget by April and was “back to the drawing board” after monthly API costs reportedly reached between $500 and $2,000 per engineer.

At ServiceNow, CIO Kellie Romack called rising AI costs a difficult management challenge after the company reportedly burned through its annual Anthropic allocation within the first months of the year.

Why AI spend is uniquely hard to forecast 

Finance leaders aren’t new to variable costs; cloud infrastructure, sales commissions, travel, and payment processing can and do fluctuate. What makes AI different is where the spend originates.

“It’s an underpriced resource right now. You’re getting a discounted rate, and you need to know that.” — Jothan Webb, CFO at Censuswide

Much of today’s AI adoption is happening from the bottom up:

• Engineering teams experimenting with coding agents
• Marketing teams generating content workflows
• Operations teams automating repetitive processes
• Customer support teams integrating AI assistance tools
• Individual employees expensing standalone AI subscriptions

The result is AI tool sprawl: dozens of vendors, overlapping contracts, fragmented ownership, and consumption patterns that often sit outside centralised procurement.

And unlike traditional software contracts, many AI platforms now scale using invisible units like tokens and prompts. Most business leaders don’t intuitively understand what drives those costs until invoices arrive.

Under flat-rate SaaS pricing, the key financial question was usually: “How many seats do we need?” Under consumption pricing, the question becomes: “What behaviours are driving usage, and who owns the budget when usage spikes?” That’s a fundamentally harder governance problem.

As Jothan from Censuswide explains, “[SaaS companies] started off by undercutting the market and gradually came up to it once everyone adopted it. AI is going to be the same.”

A practical framework for finance leaders 

Finance teams don’t need to stop AI adoption to regain control, but they do need a different operating model. "Whatever you try to budget around AI, it will be wrong, either directionally in terms of how much you spend, or in the impact it has organisationally,” says James Cleave, CFO at Everflow Utilities.

“My focus for next year's budget is aligning the senior team on AI strategy and governance, and how we stay agile when the landscape changes. The world is fundamentally different to what it was even six months ago, so being responsive is more critical than ever." — James Cleave, CFO at Everflow Utilities 

This three-step framework is a useful starting point.

1. Audit AI tool sprawl before costs compound 

Most finance leaders are already familiar with approved enterprise AI vendors. Fewer have visibility into departmental or individual adoption.

Start by identifying:

• Which teams are purchasing AI tools directly
• Which vendors are reimbursed through expense systems
• Which APIs connect to engineering or cloud environments
• Which “pilot” projects have become everyday production workflows
• Where duplicate tools exist across functions

The goal is simply to establish a baseline.

Many organisations are currently managing AI budgets based on top-down assumptions rather than actual usage data. That worked when AI adoption was modest and subscription-based, but it becomes risky now that consumption pricing has entered the picture.

Pay close attention to shadow AI spend embedded inside existing software contracts. Increasingly, vendors are bundling AI features with separate metered components that sit outside core seat pricing.

2. Identify which vendors are moving to consumption pricing

Not every AI vendor has fully transitioned at this point, but the direction of travel is clear. Finance leaders can proactively classify vendors into three groups:

• Fully usage-based today
• Hybrid subscription plus consumption pricing
• Flat-rate now, but likely to transition at renewal

Anthropic and GitHub have already moved. Others are likely evaluating similar changes as model usage increases and infrastructure costs rise.

3. Set team-level guardrails before surprises emerge

Effective AI governance models have clear financial guardrails, which means establishing:

• Team-level budgets for AI experimentation
• Usage alerts tied to thresholds
• Clear ownership for API consumption
• Monthly review cadences with department leaders
• Policies around approved versus unapproved tools

Pleo, for example, offers Vendor cards (essentially a virtual payment card) that help manage your recurring payments and digital spending by clearly separating them from employee expenses. Simply set aside a fixed amount for each AI tool; your other payments stay protected, and you can see exactly what each tool is costing you.

Importantly, these controls shouldn’t frame AI as a simple cost problem to suppress. “Spend on AI requires fundamentally different thinking", notes James from Everflow Utilities. “I think about it more like marketing spend than IT licensing: more token spend isn't inherently good or bad, it depends entirely on the outcome. We're at the stage where I want people to experiment with these tools. If we add too many guardrails, people will either stop using them or use personal licences, which is a much greater risk."

The companies seeing the fastest AI cost growth are often the ones generating the strongest productivity gains. For example, Uber’s challenge was its unexpectedly successful adoption, which drove up usage and costs, rather than failed adoption.

Finance leaders, therefore, need governance systems that preserve experimentation whilst restoring predictability. “We need to make sure that stopping AI isn’t a business killer,” says Jothan. “As we move toward the point where it would be, you really need to understand where you could pare back, and where it’s genuinely critical.”

The broader shift that’s redefining software spend

AI is the headline story today, but the underlying trend is larger.

Enterprise spend is becoming increasingly distributed, increasingly usage-driven, and increasingly difficult to forecast with traditional procurement models. AI simply accelerates the pattern because consumption scales so quickly.

For CFOs, that changes the role of spend management. It’s no longer just about approving annual software budgets, but maintaining real-time visibility, applying dynamic controls, and creating clearer accountability across teams.

For some finance leaders, that's already reshaping how they work with technical teams. James Cleave again: "The lines are blurring as to who does what. Our focus is on increasing the pace of delivery while getting the governance right. Those conversations were always happening, but the environment is more dynamic now, given the pace of change."

Understanding exposure early helps finance teams respond proactively, rather than discovering unexpected usage and costs at the end of the quarter. And for many companies, that next surprise invoice is probably already accumulating.

Pleo helps finance teams set team-level budgets, track real-time spend, and catch cost surprises before they become P&L problems. Learn more about our spend management solution.