CIOs are grappling with rising AI costs as token usage eats up budgets.
Costs often center around the usage of tokens, the manageable fragments of data that AI models process and how AI model providers price inputs and outputs, said J.R. Storment, executive director of the FinOps Foundation, speaking during the keynote address at the FinOps X 2026 conference Tuesday.
Many companies have shared stories about already being three times over their token budgets, just shy of midway through 2026, Storment said. In April, Uber CTO Praveen Neppalli Naga told The Information how employee use of Anthropic’s Claude Code maxed out his AI budget four months into the year. The rideshare company in June capped employee use of AI tools to reduce costs.
Newer models and the rise of agentic AI have contributed to greater token usage and, as a result, higher costs, Storment said. Global token usage is expected to multiply 24 times by 2030 to 120 quadrillion tokens per month, according to research from Goldman Sachs. CEOs are pointing at rising AI costs as the biggest challenge for companies today, Storment said.
“Agentic hit the scene and exploded,” he said during the event. “Adding reasoning and loops and retries and corrections, and all this insanity where a nonhuman — that does not give up, does not rest, and does not tire — keeps hitting the LLMs, keeps retrying, keeps failing. This created a lot of loops and an extravagant amount of token usage.”
Shutterstock, a global provider of licensed images, video and music, has onboarded multiple generative AI tools to help customers create content, Shutterstock CTO and CISO Courtney Totten said during the FinOps event keynote. As the company embraced AI, Totten said she has had to balance innovation with value and cost.
“Understanding your AI costs is no longer optional, it is foundational to business strategy,” she said.
AI costs and FinOps
A few months ago, Totten said Shutterstock executives asked why not give customers access to the best AI models every time they generate an image. The answer: cost.
AI models vary significantly in cost and capability, Totten said. Less advanced models — which use less tokens — can handle basic prompts, such as asking for an image of a dog. However, larger, more specific queries such as an image of a golden retriever with a family baking cookies in a kitchen, requires a more advanced model and thus more tokens.
“More token use, higher cost,” she said. “Our output tokens make up 75% of our total token consumption. Every time somebody comes to our site and searches for something or creates a generative image, that’s creating tokens, that’s creating costs for us. It’s our primary cost driver.”
Totten said she realized she had an education problem, recognizing that business leaders didn’t have the information they needed to make strategic decisions about what AI models to use and what it would cost.
“We jumped from 50 user licenses of ChatGPT to 750 in a month,” she said. “That is significant, that’s changing our costs. End products, adding features, all of these things are creating costs, yet our leaders didn’t have what they needed.”
Totten said it remains a significant challenge to quantify costs, as different models are used across varying use cases. Teams lack a true understanding of and visibility into overall AI costs, she said. Tech leaders are also facing pressure from executives to do more with less and not blow budgets.
Business leaders need to know which AI providers the company works with and what models the business currently has access to. Showing signals for strategy to enable cost forecasting is also critical, as is understanding where the company has made usage commitments to vendors.
Shutterstock identified $250,000 in unused commitment that would’ve gone to waste, she said.
Technology executives should also make AI costs a priority with a top-down mandate, Totten said. She partnered with the CFO to require all AI costs to come through the FinOps team, which reports to her. As AI evolves business operations, Totten said Shutterstock’s FinOps team now focuses on both cloud and AI costs.
Traditional FinOps was dying before the rise of AI and has resulted in a shift, said Pooja Kumar, VP of cloud strategy and transformation at Prudential Financial. FinOps is becoming “a true strategic imperative in the age of AI,” she said during the conference keynote.
Instead of building dashboards or taking on other tasks that are easily automated, she said FinOps teams need to lean into value generation and token economics to become an indispensable part of the organization.
“The questions that I’m getting and executives are asking these days are not about how much I’m spending on cloud, AI or other things, that’s not the question anymore,” she said. “It’s, ‘what does one business outcome cost end-to-end and what is the cost of doing it responsibly?’ And that’s where we are shifting.”
