Google just gave businesses a way to put a ceiling on their AI bill. The company added a pricing calculator, spend caps, and overage controls to Gemini Enterprise, its AI product for companies, plus a pay as you go payment option to sit alongside the existing per seat subscription. In plain terms, a finance team can now set a maximum monthly spend and get warned or cut off before it blows past that number.
This is not a small feature update. It is a signal that AI spending has become a real problem for the people who write the checks. Gartner expects worldwide spending on AI models and platforms to reach 64 billion dollars this year, up from 39 billion dollars last year. Costs are rising because companies are moving from simple chatbots to AI agents that take multiple steps to finish a task, and every step burns more computing power.
Google is late to this party, not early. Snowflake already added a feature that automatically sends a task to a cheaper AI model when a more expensive one is not needed. Oracle sells AI usage in fixed bundles instead of open ended pricing. Amazon has previewed a tool that hunts for unusual cost spikes. Anthropic already lets customers set a monthly spending limit that pauses usage once it is reached. Google is catching up to a trend its competitors started.
The more telling story is why costs are spiraling in the first place, and most cost calculators do not fix it. Research from McKinsey found that most companies already spend more on AI agents than they budgeted for, and a large share of that spending, roughly 60 percent, goes toward the AI checking, correcting, or redoing its own answers rather than the original task. A spend cap stops the bleeding once it happens. It does not stop an agent from taking three expensive attempts to get a simple answer right.
That gap is why a group of large companies, including IBM, Accenture, SAP, ServiceNow, and JPMorgan Chase, formed a new nonprofit called the Tokenomics Foundation. Its job is to build shared, vendor neutral rules for measuring what AI actually costs and what value it delivers, similar to what the FinOps movement did for cloud computing a decade ago. Right now, every vendor measures cost and value differently, which makes it hard to compare Google's pricing against Microsoft's or Amazon's in any honest way.
There is also a quieter path some companies are taking: building smaller, cheaper models in house for routine work and saving expensive frontier AI only for the hardest questions. Travelers Insurance does exactly this, routing everyday insurance questions to its own model and saving broader reasoning tasks for larger AI systems.
For any business using or considering AI agents, the lesson is straightforward. Treat the AI budget the way you treat cloud computing spend: set limits before you turn features on, not after the invoice arrives. Vendors are finally building the tools to do this, but the tools only work if someone actually uses them.