UBS: 60% of Enterprises Are Tightening AI Spending
Alina Collins
A UBS survey finds roughly 60% of enterprises are capping or curbing AI consumption. Analysts call it a mild "emerging headwind" — nobody is hitting the brakes, but the money is getting pickier.
What exactly are 60% of enterprises tightening?
UBS analysts spoke with over a dozen enterprise IT leaders since June. About 60% are imposing some form of AI-spending control.
The most common lever: token-usage caps — not slashing budgets, but drawing a line on how much each team can consume.
This means → companies still value AI, but the mode has shifted from "use freely" to "use deliberately."
Is this bad news?
The analysts are explicit: "nobody is hitting the brakes." They frame the trend as a "healthy issue" — a mild, emerging headwind.
Severity varies: for some firms it is a "meaningful downshift," while others — still in early deployment or already seeing positive ROI — feel less impact.
One enterprise put it plainly: "The question isn't whether to use tokens, but how to use them efficiently." Optimization is an engineering discipline, not a budget crisis.
Another firm's CTO went all-in on AI early, but has since cut internal AI tools from five to two — the annual token budget was nearly exhausted.
Who gets hurt, and who benefits?
Closed-source model providers face the most pressure — OpenAI and Anthropic are first in line, because the token bills enterprises are trimming go directly to them.
Open-source models and Chinese competitors like DeepSeek stand to gain most, especially for non-coding tasks where enterprises are more willing to swap in cheaper alternatives.
In plain terms = enterprises are not quitting AI — they are shopping for a cheaper way to use it. Whoever is cheapest moves up.
What to watch next?
Next-generation models trained on new chips could push token costs even lower. Google's Gemini 3.5 Flash and Anthropic's Claude Sonnet 5 are already positioned on the cost-reduction curve.
This means → the supply side is cutting prices proactively, which could accelerate the enterprise shift toward efficiency.
This reflects a larger question: whether the inflection point from "scale-first" to "efficiency-first" AI spending has arrived — and that answer will shape demand across the entire AI infrastructure chain.
Content is for reference only, not financial advice.