Companies Are Cutting AI Budgets as Costs Rise and Results Disappoint

AI Budgets: After a year of pushing employees to use AI everywhere, a growing number of large companies are putting limits on it. Bills are higher than forecast and the savings are smaller than promised. This is less a retreat from AI than a shift from “use it everywhere” to “use it where it pays.”

What is actually happening

Reports say Uber set a $1,500 monthly token cap per employee after executives questioned whether the productivity gains justified the cost, and that Amazon and Meta dropped AI usage leaderboards. Bloomberg reported that Walmart is also placing a token cap on employee AI usage after demand surged.

Other examples include:

Uber’s Chief Operating Officer said it is getting harder to justify AI costs, and that there is no clear link between token spending and new customer features.
Microsoft began canceling employee Claude Code licenses in favor of its own GitHub Copilot CLI tool, a move that appears at least partly cost-driven.
Reuters reported that Starbucks dropped an AI inventory tool after only nine months.

These are individual cases, so they don’t prove every company is doing the same. But the survey data points the same way.

Why AI costs keep rising

1. Running AI costs more than building it. The FinOps Foundation says inference, the cost of running models to produce answers, makes up roughly 85% of enterprise AI budgets. The more people use AI, the more the bill grows.

2. Budgets are underestimated. Research cited by one report found close to eight in ten enterprises overshot their AI costs in the past year. Another analysis found more than two-thirds of AI projects exceed their budgets, by over 40% on average, mainly because of data quality and legacy-system integration.

3. Companies priced the model, not the system. The subscription or token price is only part of the cost. Review, security, integration, and training add more.

4. “Use more AI” mandates drive up usage. If staff are rewarded for heavy use, they will use it heavily, whether or not it helps. Once finance teams add caps, AI becomes a cost to manage like any other.

Why results are disappointing

Savings are smaller than expected. A Bain & Company survey of nearly 1,000 large companies found about 40% saw AI cost reductions of 10% or less, and only 4% saw savings above 30%.
Cutting staff doesn’t guarantee returns. Gartner surveyed 350 businesses with over $1 billion in revenue. About 80% cut staff after deploying automation, yet they were no more likely to see positive ROI than firms that kept their workforce. Gartner said the organizations seeing results invested in new skills and roles around human-guided systems.
Pilots don’t always scale. One analysis argues that many early AI layoffs were based on pilot-stage results, not sustained real-world performance.

Is the AI boom over? Not quite

This is where the headline needs care. Gartner projects worldwide AI spending of roughly $2.5 trillion in 2026, up from about $1.5 trillion the year before. Agentic AI software spending is forecast to reach $376.3 billion by 2027.

So the picture is mixed. Total spending is growing while individual companies tighten control and cut low-value uses. Companies are limiting high-volume token usage, setting hard spending caps, and prioritizing uses with clear returns. Some analysts expect tools that cut inference costs and manage AI spending to benefit from this shift.

What this means for you

If you run or manage a business

Start with one measurable problem, such as support response time or invoice processing, not a company-wide rollout.
Set a baseline first. Record current cost, time, and error rate so you can prove a change.
Assign cost ownership. One person or team should track AI spend monthly.
Set usage limits early and review them regularly.
Match the model to the task. Don’t use your most expensive model for simple jobs.
Keep humans reviewing important output. One report found a majority of companies have no formal process for checking AI-generated work before it reaches a customer.
Stop what doesn’t work. Dropping a failed project is a sign of discipline, not failure.

If you’re an employee: expect more focus on proving value. Skills in judging, checking, and improving AI output matter more than simply using it often.

If you’re an investor: note that the pullback is about how AI is used, not whether it is used. Be careful with one-sided headlines in either direction.

Frequently asked questions

Are companies really cutting AI budgets?
Some are, and many more are capping or redirecting spending. Overall AI spending is still forecast to grow.

Which companies are limiting AI use?
Reports name Uber, Walmart, Amazon, Meta, and Microsoft in different ways. Check each original report, since the actions differ.

Why is AI so expensive to run?
Most of the cost is inference, the ongoing cost of generating answers, and it grows with usage.

Does replacing staff with AI improve ROI?
Gartner’s survey found no clear advantage over keeping staff. Results came from investing in skills and new roles.

Should small businesses avoid AI?
No. Start small, measure results, and set a budget cap.

How can I control AI costs?
Set limits, pick cheaper models for simple tasks, track spend monthly, and prioritize uses with proven returns.