FinOps Under Siege: AI’s Token Economics Forces a Decade of Evolution into One Year

Breaking News: AI Costs Reshape Cloud Financial Discipline in Months

LAS VEGAS — The financial discipline known as FinOps, which emerged over the past decade to manage cloud costs, is now being forced to adapt to artificial intelligence—and it has roughly one year to do so, according to industry leaders speaking at Google Cloud Next.

Token economics, the pricing model for AI models based on tokens processed, is breaking the assumptions that underpinned FinOps in the cloud era. Roi Ravhon, co-founder and CEO of Finout, told The New Stack Makers podcast: “We need to do the same thing we did for cloud to AI, but we’re doing it in a year.”

The urgency stems from two key differences: AI costs are rising even as token prices fall, and the cost of the same prompt can vary wildly. Ravhon explained, “You ask the same question twice, and you get different token usage for everything. So how can this scale?”

Background: Cloud vs. AI FinOps

During the cloud era, FinOps had roughly a decade to mature. Companies learned to track compute, storage, and network usage, optimizing for predictable cost patterns. AI, however, introduces unpredictable token consumption.

CFOs initially approved unlimited budgets for AI innovation, but the conversation has shifted back to ROI. “CFOs started with ‘unlimited budgets, let’s be innovative,’” Ravhon said. “Now they’re circling back to ROI.”

What This Means: The New FinOps Playbook

Pathik Sharma, who leads cloud FinOps at Google Cloud, offered a solution: don’t overuse the most powerful AI models. He recalled a customer who used Gemini Pro for simple tasks like email summarization. “We started using Pro for everything,” the customer said. But most tasks are better served by Flash, Google’s cheaper model.

Sharma stressed the importance of an orchestration layer: “The FinOps discipline isn’t about asking every employee to memorize which model fits which task. It’s about building an orchestration layer beneath that routes each request to the cheapest model that can reliably answer it.”

He also warned against “reaching for Thor’s hammer when you don’t need it.” In other words, small models often suffice for routine queries, sparing GPU/TPU resources.

Additional Cost Drivers

AI costs extend beyond LLM API spend. Sharma noted that training compute, inference compute, and storage for training data all contribute. “GPUs and TPUs are still scarce,” he said, “so efficient orchestration is critical.”

Agentic FinOps and Deterministic Guardrails

The discussion also covered agentic FinOps—automated cost optimization tools. Ravhon and Sharma agreed these tools require deterministic guardrails to be useful. Without them, automated decisions can drive costs higher.

Both recommended that practitioners first learn from the FinOps Foundation before turning to vendors. “Start with the foundation’s training and framework,” Ravhon said. “Then apply it to your environment.”

Immediate Actions for Enterprises

• Audit AI usage: Identify which models are used for which tasks.
• Build an orchestration layer: Route requests to the cheapest capable model.
• Monitor token costs: Implement tracking for each prompt to detect anomalies.
• Set governance policies: Limit use of expensive models to essential tasks.

Looking Ahead

As AI models become more capable, the FinOps discipline will need to evolve continuously. The one-year timeline serves as a stark wake-up call for CFOs and engineering leaders alike.

“Token economics are on the clock,” Ravhon concluded. “We don’t have the luxury of a decade.”