Runaway AI Spending Forces a Return to Cloud Controls

AI spending is becoming harder to govern because the bill now moves with every token, prompt, and agent action rather than a fixed software seat. That shift is forcing companies to borrow controls from the cloud era, then rework them for a product category where ordinary employees can trigger costs in real time.

Businesses are using cost controls developed for cloud computing to manage rising AI bills, according to PYMNTS, citing a Wall Street Journal report from Tuesday, June 30. The core problem is simple: as AI moves from pilots into daily workflows, usage-based pricing turns adoption into a moving target for finance teams.

AI Spending Discipline Has Moved From Pilot Cleanup to Budget Control

The sharpest warning came from Chris Reed, senior director of IT finance at Priceline. His point was not about AI quality. It was about who now has spending power.

“With AI, you’re putting the credit card in the hands of the end user. If you have no control over that, or if the end user is not educated enough, they’re going to run up that tab,” Reed told the WSJ.

That quote captures the change. Traditional enterprise software procurement usually starts with contracts, seats, permissions, and renewal dates. AI consumption can start with a prompt. If the company uses agentic systems, the cost can extend beyond one employee request because “every step an agent takes runs a meter,” as PYMNTS wrote in a related report cited in the source.

XOOMAR analysis: this makes AI spending a governance issue, not a cleanup task for IT finance after the invoice lands. The budget risk sits inside adoption itself. The more useful the tools become, the more often employees use them, and the more finance teams need visibility before month-end.

That also explains why cloud spending history matters. Companies already learned that variable infrastructure bills need tagging, monitoring, and accountability. AI brings the same lesson, but with a different trigger: human behavior and automated model actions, not just provisioned servers.

Token Volatility Turns Small Usage Changes Into Invoice Risk

The PYMNTS report identifies tokens as the basic unit behind AI billing and says companies face volatile pricing around them. It also notes that AI providers initially priced access through flat subscriptions, before usage-based billing became standard as agentic models moved into coding, customer service, research, and procurement.

That shift matters because the cost driver is no longer only access. It is activity.

A support chatbot, an internal research assistant, a coding helper, or a procurement agent can all create repeated model calls as usage grows. The source does not disclose invoice sizes or per-token rates for the companies named, but it does show why executives are trying to control AI spending earlier in the workflow.

BCG’s supplied survey context reinforces the budget pressure. Its latest IT buyers survey covered 602 IT buyers across North America and Europe. Respondents expected 2025 IT spending to grow 4.6% year over year, up from 3.5% in 2024. AI and GenAI ranked among the areas with the largest expected spending increases, while four out of five responding companies had adopted some level of GenAI.

That does not mean every AI dollar is wasteful. It means AI has moved into the part of the budget where finance teams will demand proof. As we noted in Billions Ride on AWS Public Sector AI's Cloud Grab, cloud AI spending is increasingly tied to large institutional use cases, which raises the stakes for governance when usage expands.

Cloud FinOps Playbooks Are Being Rewritten for Generative AI

Principal Financial Group is applying cloud-style cost management to AI. Kathy Kay, the company’s chief information officer, said financial services firms are “putting governance and optimization practices in place, similar to what companies have done with cloud, to manage costs as we scale.”

Principal’s cited tactic is model fit. The company is concentrating on using the right AI model for the right task so that “higher usage doesn’t necessarily translate into higher costs,” Kay said.

She also described the need to design for change: “Given how quickly pricing and capabilities are evolving, we’re designing for flexibility so we can adapt over time and continue to deploy AI efficiently.”

That is the most important operating principle in the report. AI cost control cannot depend on one static vendor decision, because the source says pricing and capabilities are changing quickly.

Smartsheet offers the clearest operational example. Ravi Soin, its CIO and chief information security officer, said the company’s FinOps team, described as a mix of financing, engineering, and product, tracks overall AI spend. Smartsheet has also created automated alerts that notify employees when they are about to reach token limits.

“We have user dashboards available to the entire company, by department, by manager, so you have real-time visibility on how often and what your costs are, so it isn’t a surprise at the end of the month,” Soin said.

XOOMAR analysis: dashboards and alerts change the politics of AI spending. They make usage visible to teams before finance has to intervene. That is less blunt than cutting access after costs spike, and it gives managers a way to distinguish productive consumption from careless usage.

CFOs, CIOs, Vendors, and Employees Are Pulling AI Budgets in Different Directions

The source does not give direct comments from CFOs, but the tension is visible. CIOs want adoption with guardrails. Finance teams need spend that can be tracked, allocated, and explained. Employees want access to tools that help them work faster. Vendors benefit when usage rises.

That is why Ramp appears in the PYMNTS piece. PYMNTS reported earlier this month that Ramp raised $750 million at a $44 billion valuation, almost tripling its worth in a year. The company is betting that AI consumption, billed by the token and shifting with every prompt and agent action, has become a cost category most enterprise finance teams cannot track, allocate, or control.

That is a direct signal that AI spending control is becoming a software market of its own. It is not only an internal accounting headache.

There is also a cultural problem. If employees view AI as a productivity tool, limits can feel like bureaucracy. If finance cannot see usage clearly, open-ended access can look reckless. Companies will need governance that is visible enough to change behavior but not so heavy that it kills experimentation.

The same trust issue shows up in other AI-adjacent enterprise decisions. Our coverage of Shopify Trustpilot Deal Puts AI-Era Trust on the Line looked at how companies are trying to make AI-era workflows credible to customers. Inside the enterprise, cost transparency plays a similar role for executives.

AI ROI Will Be Judged by Unit Economics, Not Demo Quality

The next phase of AI evaluation will not be won by impressive demos alone. It will be won by use cases that can survive cost attribution.

The source supports one clear direction: companies are trying to make higher usage less directly tied to higher cost through model selection, dashboards, token limits, and FinOps oversight. XOOMAR analysis: that points to a unit economics test. If a customer service AI tool costs more as usage rises, finance will ask whether it reduces workload elsewhere. If a coding assistant consumes more tokens, managers will need to know whether it saves enough engineering time to justify the spend.

Without attribution, companies risk making the wrong cuts. They could throttle valuable AI use because the total bill looks high, while missing low-value consumption that keeps growing quietly. Smartsheet’s department and manager dashboards are designed to avoid exactly that kind of month-end surprise.

BCG’s survey context also shows why this tension will intensify. Tech budgets are growing, but companies are still reallocating funds toward AI, cloud services, security infrastructure, and analytics while reducing spend in some mature categories. AI has to compete for budget even when it is a priority.

Cheaper Model Choices and Tougher Contracts Are the Next Test

The practical effect for enterprise buyers is already visible: AI procurement is becoming more controlled, more measured, and more tied to usage data. Companies that can route work to the appropriate model, monitor token limits, and show costs by department will have more room to scale AI adoption without budget shocks.

Vendors should expect tougher questions. If AI features are billed through usage, customers will want clearer visibility into what drives the invoice and where savings show up. Vague AI charges will become harder to defend once buyers build internal dashboards and FinOps teams around this category.

The next evidence to watch is specific. More companies should disclose whether AI governance reduces surprise bills, whether model selection lowers cost growth as usage rises, and whether finance teams can connect token consumption to measurable business results. If those controls work, AI spending can keep expanding with discipline. If they fail, the same usage-based model that makes AI easy to adopt could become the reason deployments stall.

The Bottom Line

• AI adoption can create fast-moving costs that traditional software controls were not designed to manage.
• Finance and IT teams need real-time visibility as employees and agents trigger usage-based charges.
• Cloud-era cost governance is becoming a template for keeping AI spending predictable.