20.7x Token Surge Forces AI Cost Management Crackdown

AI token spend across Ramp customers has jumped 20.7x since June 2025, a scale shift that turns enterprise AI from a productivity experiment into a finance-control problem. The old habit of “tokenmaxxing,” pushing employees toward the biggest models and heaviest usage as proof of progress, is colliding with CFO scrutiny, according to PYMNTS.

The thesis is simple: AI cost management is becoming a required layer of enterprise software. AI usage is moving from experimental budget line to managed operating expense, and finance teams won’t tolerate opaque token burn for long.

20.7x token growth puts AI cost management on the CFO agenda

Ramp launched AI Token Spend Management on July 16, giving finance teams a dashboard to track, allocate and control AI spending across providers including OpenAI, Anthropic, Gemini and Cursor, per the company’s announcement cited by PYMNTS.

That matters because AI pricing doesn’t behave like the enterprise software finance teams already know. Traditional SaaS budgets were built around annual licenses, seat counts and contract tiers. Finance could forecast them with some confidence. AI, by contrast, runs through tokens, compute cycles and API calls. Usage can spike without a new purchase order.

Ramp said it built the product using usage data from more than 1,300 businesses and analysis of 110 trillion tokens. It identified three spending profiles based on model choice, caching behavior and spend-control discipline.

That last detail is the real story. Companies aren’t just buying AI tools anymore. They’re discovering that how employees use AI can matter as much as which tool they bought.

From SaaS seats to AI tokens, the budget model broke open

The old software budget had a rough ceiling. If a company bought 500 seats, the finance team knew the shape of the bill. AI usage has a different profile: each prompt, model call, workflow automation or API request can generate incremental cost.

This is why “tokenmaxxing” became a problem. In early enterprise AI adoption, heavy use looked like momentum. More prompts meant more adoption. Bigger models suggested better output. But when AI use spreads across teams and sits inside provider dashboards or invoices finance teams struggle to interpret, consumption stops looking like progress and starts looking like unmanaged exposure.

XOOMAR analysis: this resembles the broader finance-operations shift we track in areas like Real-Time Payments Invade Payroll, Checkout and B2B and Credit Card Installments Crush BNPL as Usage Hits 33%. Different markets, same management question: when usage accelerates, controls need to catch up.

12% potential savings show the gap between AI usage and AI value

Ramp found that, as of June, the average business could identify potential savings equal to 12% of its monthly AI spend. It also found that one in three businesses identified a lower-cost model alternative capable of doing the same work.

That finding cuts through the noise. The issue isn’t total AI spend alone. It’s whether expensive model usage produces enough incremental value over a cheaper alternative.

The Sansa Services example shows how quickly waste can hide in settings rather than strategy. Ramp reported that an employee left an expensive “Fast Mode” setting active without realizing it, pushing token spend sharply higher before the issue was caught.

“We flagged that somebody was using Fast Mode unnecessarily and that resulted in six times the token spend over a seven-day period. They’ve since turned it off. That is one story in which basically the product pays for itself,” said Neusha Sayadian, founder and fractional CFO at Sansa Services, according to Ramp.

That quote is the CFO case for AI cost management in one sentence. A single configuration choice created six times the token spend over seven days. The productivity value may or may not have changed. The bill did.

CFOs want AI agents that control budgets, not just create more work

PYMNTS Intelligence found that 43% of surveyed CFOs expect agentic AI for dynamic budget reallocation to have a significant impact. That was the highest-ranked use case cited in the source material.

These CFOs are not just asking for a cleaner invoice. They want systems that continuously scan spending patterns, flag overruns and shift funds toward higher-priority areas using real-time cost data.

That changes the role of agentic AI inside the enterprise. The early pitch centered on agents doing work. The finance pitch is now sharper: agents should also police the cost of doing that work.

XOOMAR analysis: the next internal debate won’t be “should employees use AI?” It will be “which work justifies premium AI?” That creates tension. Technology teams need room to test tools. Employees want strong outputs. Finance teams want forecasting, allocation and spend discipline before AI becomes another line item that expands faster than anyone can explain.

CloudZero is tying AI spend to customers, features and teams

Ramp is approaching the problem through token controls. CloudZero is attacking it through attribution. The company’s financial control platform connects AI spending to the customers, features and teams that generated it, according to its launch announcement cited by PYMNTS.

That is a different but related move. Counting tokens tells finance what was consumed. Attribution tries to show what that consumption produced.

“AI is becoming central to how companies build products, serve customers, and run the business,” said Scott Castle, CloudZero chief product officer. “But most companies still manage AI with token counts and monthly invoices, even as costs rise faster than expected and ROI remains unclear.”

That framing points to a new software category: AI spend management, or AI FinOps. The source material doesn’t prove the category’s eventual size. It does show why the need exists: token counts alone don’t answer whether a customer feature, internal workflow or product integration is worth its cost.

The cloud spending hangover is repeating faster with AI

The supplied context on cloud cost management makes the comparison hard to miss. Enterprises spent years building governance around cloud consumption: visibility, forecasting, optimization and alignment between finance and engineering. AI is reopening that problem because workloads behave less predictably than traditional applications.

AI projects can begin as pilots, then expand as teams test prompts, models and workflows. Costs can appear through cloud infrastructure, API consumption, third-party subscriptions, storage expansion or supporting data services. The source context also notes that leadership may see cloud costs rising without clearly identifying which AI initiatives are responsible.

The difference is speed. Cloud waste often required infrastructure provisioning. AI usage can grow through daily employee behavior and embedded workflows. Cutting spend blindly is risky too, because AI costs are tied to knowledge work and decision processes. A cheaper model may be enough for one task and inadequate for another.

That makes discipline more valuable than blunt cuts.

Smaller models and CFO-approved automation are the next test

The practical implication for enterprises is not “use less AI.” It’s use AI with cost signals attached.

Expect more scrutiny around:

• Model selection: Which jobs need premium models, and which can run on lower-cost alternatives?
• Caching: Ramp’s analysis explicitly included caching behavior as a spending-profile factor.
• Attribution: Which team, feature or customer generated the spend?
• Controls: Where should budgets, alerts and caps sit?
• Automation: Can agents manage budgets in real time, not just consume tokens?

The evidence that would confirm this thesis is already visible in the source material: Ramp’s 20.7x token-spend growth, its 12% average potential savings finding, the one in three lower-cost-model alternative figure and CFO interest in agentic budget reallocation at 43%.

The evidence that would weaken it would be different: if companies find that premium models consistently justify their cost, if spend-management tools fail to connect usage to value, or if finance teams accept AI invoices without demanding attribution.

For now, the direction is clear. The winners won’t be the companies that consume the most AI. They’ll be the ones that know when expensive AI is actually worth paying for.

The Bottom Line

• AI spending is shifting from experimental budgets to managed operating expenses.
• Finance teams need new tools because token-based pricing is harder to forecast than SaaS seats.
• Employee usage patterns can materially affect enterprise AI costs.