Companies are no longer using AI as an experiment, they are now at a point where they view AI spending as a serious financial risk. That is why Anthropic created a budget alert feature into Claude Enterprise . This new feature includes model access control, along with warning messages when the team will exceed their budget. Vendors never create such features for customers that have historically spent money wisely.

The release came at the end of a year of companies going over their budgets. Many executives are now asking themselves if they’re actually making money with all of their high-priced AI investments, according to Axios . An AI consultant told Axios that one client spent roughly half a billion dollars in a single month after failing to set usage limits on employee AI licenses. Uber burned through its entire 2026 budget for AI coding tools in four months . Its chief operating officer then said the costs are getting harder to justify.

Enterprise budgets contain "cushion" that no small business has. The impact of one bad quarter for an enterprise will be absorbed by a well-crafted memo (policy) and then the company moves forward. That is not true for a 12-employee firm. If the AI costs for small businesses were treated similarly to those for Uber's (a large company), they would create payroll pressures for a 12-employee firm.

Agentic AI Broke The Cost Model Everyone Was Using

A chat-based AI had an inherent "spending" governor. People could only spend time typing. Agents removed this governor. An AI agent (as opposed to a human) working on a multi-step process can use tokens while you are sleeping, if the first attempt fails, continue to try until the process has been completed. Most did not account for these costs as few companies were utilizing agents at scale. In its 2026 Enterprise AI Survey , Writer found that 59% of surveyed enterprises invest at least $1 million a year in AI and only 29% report significant ROI from generative AI. As such, the amount of money spent on AI technology was growing much quicker than the body of knowledge relative to how best to use it.

Small Businesses are operating at an even smaller scale of this identical pattern. Subscriptions build upon each other. Per-seat AI charges for your current tool(s) add quietly. A single line item charge builds over time with no single "I’m making that decision" moment. The agents (not necessarily from Microsoft or Google) accelerate the creeping nature of this process. Why? Because the cost of an AI Subscription is no longer simply monthly seats. It can be, and often includes costs such as, AI token costs, model pricing and agentic AI usage-based pricing etc.

Set The Cap Before The Tool Enters The Business

Before buying into an AI subscription again, determine how much money can be spent on AI by your company each month. It is being given to enterprise customers as part of their service agreement. Anthropic is giving Enterprise Administrators additional controls over who can use AI, how much they can use AI and how much they can spend. For those vendors that publish pricing based upon usage (i.e., for example ElevenLabs publishes price per minute for voice and agents), this reinforces the fact that there is a cost associated with every automated minute . If a vendor provides options to set an upper limit on the amount of money your company will pay for AI, implement one immediately when adopting an AI solution. If not, then your accounting department will need to create a new line item each month for AI costs that includes an upper limit (AI budget cap) and review it at least annually.

Match The Model To The Task

The biggest cost of all with your AI budget is to spend on a high-powered (the most powerful) model for low-level tasks. In their release notes Anthropic states that enterprise administrators can limit or restrict which models are accessible by users as well as what effort levels they may be used at. This has an impact on billing due to the fact that you will pay based upon model selection. You probably don't want to run your simple draft, summary, research pass and clean up processes through the high price model. Most low-cost AI tools have better performance when the appropriate model is selected for the specific process being performed. Only select and use the higher priced model if it generates income.

The Gap Between Spenders And Earners Keeps Widening

The firms that get their money’s worth out of artificial intelligence often do so by being intentional with their spending, even if they are not spending as much as other firms. According to data from PwC’s 2026 CEO Survey, only 12% of surveyed CEOs were able to identify increased revenues and cost reductions generated by AI during the last twelve months. The primary distinction between these two groups does not lie within their investment in technology or software applications. Rather, the distinguishing factor lies in the fact that one group has established clear expectations of desired results prior to investing its funds, whereas the second group has failed to define such expectations.

Most of the failure to use that check lies with the small business owner. There isn't anything wrong with being a small business owner. They have been implementing AI through subscription services for their business, one subscription service at a time. No-one ever came along and gave them an AI Budget Management System when they signed up for that "free trial." Now large enterprises are getting better spend control with each new product release. A small business owner could create something similar themselves on a Sunday afternoon.

Trace Every AI Dollar To An Hour Or An Outcome

A cap will limit spending a trace will tell you if your spending (investment) was worthwhile. Once per month, place two columns side-by-side. How much money did an AI tool spend ? And how many hours or deliverables was the AI tool able to produce with this amount spent? Most of the time saved is the true measurement of productivity when using AI tools. For the remainder of AI tools, deliverables are the true measurement. Any AI tool which can’t provide enough data for both columns over the course of two consecutive months is a subscription and not a system. The underlying reason why most businesses fail at leveraging ROI from their AI investments has not changed. The tools have changed the disciplines of ROI remain unchanged.

The owners of the cap and trace will have a better idea in 90 days as to what tools represent their money. Companies that write the big checks need to buy back this information at full price. Anthropic developed the alarm system based on the needs of larger clients since your budget is much smaller, your alarm can't wait until later.