Dive Brief:
- Oracle introduced a limited rollout of token bundles to better align agentic AI pricing with derived value. As advanced AI models and agentic tools make their way into the enterprise, executives are contending with higher usage of tokens — a unit of measure often used to price AI services — and increased AI costs.
- Allowing customers to purchase token bundles up front to use across applications creates cost predictability for businesses, Oracle CEO Mike Sicilia told investors during the company’s Q4 2026 earnings call on Wednesday. Sicilia said 33 customers took advantage of the token bundles during the limited rollout.
- “All of this helps our customers control their costs and align their spending with the value being generated,” Sicilia said.
Dive Insight:
AI costs have also caught the attention of other vendors that want to help enterprises keep track of spending.
AWS on Tuesday announced the public preview of its AWS FinOps Agent, which is designed to investigate cost anomalies, and Flexera this week launched the Flexera One platform, a set of AI cost management capabilities creating transparency into AI spend across agents, models and compute.
While Oracle rolled out token bundles, SAP has continued to build out its AI units pricing mechanism for premium services, including agentic, Philipp Herzig, global CTO and chief AI officer at SAP, said during the Bank of America C-Suite TMT Conference this week.
“The big problem is people spend a lot of tokens,” he said, building AI agents without a clear goal or idea of the return on investment.
Token usage is changing business models, and enterprises are starting to see software companies shift their pricing strategies from fees and licenses to consumption, said J.R. Storment, executive director of the FinOps Foundation, speaking during the FinOps X conference keynote on Tuesday.
The shift reflects enterprise AI maturity, as vendors no longer need to offer AI subsidies so that enterprises try new models and tools, Storment said. Case in point: OpenAI CEO Sam Altman said during an event that the AI model provider’s top token spender burns through 100 million tokens monthly.
“It turns out some of you were reported to have been costing these labs upwards of tens of thousands of dollars a month as inference whales when you were running everything on the latest model,” Storment said during the keynote.
The problem is that technology leaders lack full understanding of their token consumption, which occurs across multiple layers of the tech stack from data centers to the edge, he said. Shutterstock, for example, is requiring all of their AI costs be funneled through the FinOps team in a bid to create greater spend transparency.
Additionally, the Linux Foundation together with the FinOps Foundation this week launched the Tokenomics Foundation, which will bring together frontier AI model developers, large token consumers, neoclouds and hyperscalers to create standards and frameworks for monitoring token usage and enterprise AI costs.
“The next question the boardroom is starting to ask is, ‘What is the value of all this intelligence, why is it so expensive and am I getting enough back from it,’” Storment said during the keynote.
