In May, Axios shared that a client of theirs gave staff unlimited access to Claude licenses. In one month they racked up half a billion dollars in charges. Uber’s story is slightly less harrowing in that it took four whole months to exhaust the entirety of their 2026 AI budget.
These cases are outliers but are by no means isolated incidents. In fact, 49% of organizations have delayed or scaled back AI rollouts due to costs that are hugely disproportionate to return.
This trend has been called “AI sticker shock,” a high-tech example of learning the hard way.
Why So Many Organizations Overspend on AI
There’s no question that pricing models for generative AI tools are convoluted, especially if you’re used to software subscriptions with flat monthly cost per seat. Usage fees vary based on the:
- Amount and type of input data
- Context (previous inputs retrieved)
- Processing power required
- Complexity of the output requested
- Model chosen to complete the task
- Inputs and outputs cached for reuse
There are strategies to minimize cost such as “model routing” where you pair each task with the lowest-level model capable of successfully completing that task. However, most of us are not at all accustomed to working this way – especially if we aren’t even aware of all the variables at play.
As it turns out, only 35% of organizations have full visibility into AI operating costs, and only half have factored cost review into their AI approval processes at all. And while visibility doesn’t guarantee ROI, it does improve your odds by 400%.
Proactive Steps to Contain Costs
There are strategic, behavioral, and technical elements to consider.
- Revisit your strategy. In a study of law firms, those with a visible AI strategy were four times more likely to achieve ROI. Articulating your goals will help define an appropriate investment.
- Explore cost management settings. Set constraints within the tools themselves. As an example, Copilot allows you to limit credit usage by user, team, or project.
- Formalize usage guidelines. What happens when someone exhausts or exceeds credit limits? Will you require model routing as a technique?
- Expand employee training. Explain models and tokens and the impact they’ll have on the team’s experience. Provide reference material wherever possible.
We also suggest ongoing review of published rates. Frontier platforms are constantly evolving and constantly competing with each other, and this directly impacts pricing models.
If you’d rather have someone else manage your AI strategy, budget, governance, and training, our consultants are happy to step in! Contact us to learn more.