Blab AI
CIOs Grapple With Rising AI Token Costs Amid FinOps Shift
At the FinOps X 2026 conference in San Diego, CIOs were warned that the cost of the smallest unit of AI data—tokens—is outpacing every budget. Executive director of the FinOps Foundation, J.R. Storment, highlighted how token usage—the discrete units of data that large‑language models (LLMs) process—has pushed many firms to spend roughly three times their planned token budgets by mid‑2026.
Storment cited Uber as a cautionary example. Chief technology officer Praveen Neppalli Naga told The Information that the rideshare company’s adoption of Anthropic’s Claude Code had drained its entire AI budget within the first four months of the year. In June, Uber capped employee access to AI tools to stem the cost surge.
Token economics has become a core discipline for organizations that rely on cloud‑based AI services. The FinOps Foundation’s keynote underscored that token‑pricing models—charging per input and output token—translate directly into operational costs. Storment noted that newer, more capable models and the rise of agentic AI systems, which repeatedly query LLMs to achieve a goal, magnify token consumption. A Goldman Sachs study released earlier this year projects global token usage to multiply 24 times by 2030, reaching an estimated 120 quadrillion tokens per month. The study also points out that while the price per token has fallen, the sheer volume of usage is driving total spend upward.
Shutterstock, a global provider of licensed media, has integrated several generative‑AI tools to help customers create images, videos, and music. During the same FinOps event, chief technology officer and chief information security officer Courtney Totten explained that the company must balance innovation with cost. Totten said that AI models differ significantly in both cost and capability. Simple prompts, such as “image of a dog,” can be handled by less expensive models that use fewer tokens. More complex requests—like generating an image of a golden retriever baking cookies with a family in a kitchen—require larger, more advanced models and consequently consume more tokens. “More token use, higher cost,” Totten said, noting that output tokens make up 75 percent of Shutterstock’s total token consumption. She added that each customer interaction—search or image generation—creates tokens that translate into direct costs for the company.
Totten also highlighted a lack of cost visibility within the organization. Business leaders had limited information about which AI providers the company used and what models were in use, making it difficult to forecast spend. In response, Shutterstock partnered with its chief financial officer to route all AI expenses through a FinOps team that reports directly to Totten. The team now tracks both cloud and AI costs, aiming to prevent wasted commitments. Totten said the company identified $250,000 in unused vendor commitments that would have gone to waste if not monitored.
FinOps leaders are redefining their role in the AI era. Pooja Kumar, vice president of cloud strategy and transformation at Prudential Financial, said during the keynote that FinOps has become a strategic imperative. She argued that traditional FinOps practices, focused mainly on cloud spend, are insufficient when AI token economics dominate. Instead, teams must quantify the end‑to‑end cost of business outcomes and assess the responsibility of those outcomes. Kumar emphasized that the question is no longer how much is spent on cloud or AI, but what a business outcome costs when it is delivered responsibly.
Storment cited Uber as a cautionary example. Chief technology officer Praveen Neppalli Naga told The Information that the rideshare company’s adoption of Anthropic’s Claude Code had drained its entire AI budget within the first four months of the year. In June, Uber capped employee access to AI tools to stem the cost surge.
Token economics has become a core discipline for organizations that rely on cloud‑based AI services. The FinOps Foundation’s keynote underscored that token‑pricing models—charging per input and output token—translate directly into operational costs. Storment noted that newer, more capable models and the rise of agentic AI systems, which repeatedly query LLMs to achieve a goal, magnify token consumption. A Goldman Sachs study released earlier this year projects global token usage to multiply 24 times by 2030, reaching an estimated 120 quadrillion tokens per month. The study also points out that while the price per token has fallen, the sheer volume of usage is driving total spend upward.
Shutterstock, a global provider of licensed media, has integrated several generative‑AI tools to help customers create images, videos, and music. During the same FinOps event, chief technology officer and chief information security officer Courtney Totten explained that the company must balance innovation with cost. Totten said that AI models differ significantly in both cost and capability. Simple prompts, such as “image of a dog,” can be handled by less expensive models that use fewer tokens. More complex requests—like generating an image of a golden retriever baking cookies with a family in a kitchen—require larger, more advanced models and consequently consume more tokens. “More token use, higher cost,” Totten said, noting that output tokens make up 75 percent of Shutterstock’s total token consumption. She added that each customer interaction—search or image generation—creates tokens that translate into direct costs for the company.
Totten also highlighted a lack of cost visibility within the organization. Business leaders had limited information about which AI providers the company used and what models were in use, making it difficult to forecast spend. In response, Shutterstock partnered with its chief financial officer to route all AI expenses through a FinOps team that reports directly to Totten. The team now tracks both cloud and AI costs, aiming to prevent wasted commitments. Totten said the company identified $250,000 in unused vendor commitments that would have gone to waste if not monitored.
FinOps leaders are redefining their role in the AI era. Pooja Kumar, vice president of cloud strategy and transformation at Prudential Financial, said during the keynote that FinOps has become a strategic imperative. She argued that traditional FinOps practices, focused mainly on cloud spend, are insufficient when AI token economics dominate. Instead, teams must quantify the end‑to‑end cost of business outcomes and assess the responsibility of those outcomes. Kumar emphasized that the question is no longer how much is spent on cloud or AI, but what a business outcome costs when it is delivered responsibly.
