enterprises throttle – UBS analysts say conversations with enterprise IT executives over the past several weeks show about 60% of enterprises have started throttling AI spending by introducing guardrails as token costs rise. While the bank calls it an emerging headwind, it also argu
The AI push inside large companies is still moving, but the brakes are starting to appear—quietly, in settings and usage limits rather than press releases.
UBS analysts say executives they spoke with recently are increasingly putting “guardrails” around how much AI they use and how fast they burn through tokens. In a report. the analysts—Karl Keirstead. Timothy Arcuri. and Taylor McGinnis—write that. based on “a dozen+ conversations” with enterprise IT execs over the prior several weeks. about “~60% of enterprises were now in some manner throttling AI spend” by implementing some degree of control.
The motivation is straightforward: token spending has become a major concern. UBS links the worry to CFOs and CTOs watching AI bills rise.
One example came from Uber’s operations chief, Andrew Macdonald, who said in May that it was getting harder to justify rising costs given the “pretty meager ROI.”
UBS describes the shift as a “modest ‘emerging headwind.’” Their conversations began earlier in June and, they wrote, were later reaffirmed through more recent discussions—though the degree of impact varies from company to company.
In some organizations. UBS says. token optimization has become a “key issue” that creates a “big spending speed bump.” In others. it’s less disruptive. The analysts wrote that some companies are either “too early in their AI deployments” or are “far deeper” in the technology but “unwilling to throttle users” because they see offsetting ROI or because AI use is tied to an organizational priority to drive innovation.
For the biggest AI suppliers, the cost pressure may land unevenly. UBS analysts write that AI model makers—specifically naming OpenAI and Anthropic—are likely to be “most exposed to cost-cutting” in the near term.
They also point to open-sourced and Chinese models as potential beneficiaries, calling out DeepSeek as the biggest potential beneficiary for enterprises looking for models for non-coding-related tasks.
Even with the slowdown in spending, UBS says it doesn’t amount to panic. The analysts emphasize they are “not ringing the alarm bells. ” describing the trend as “a healthy problem.” They argue that some optimization is normal. that companies are “not hitting the brakes on AI deployment. ” and that the industry may soon benefit from new models trained on next-gen chips that could drive token costs down further.
That expectation is paired with concrete examples UBS says large AI companies have already used to sell efficiency. The bank points to Google’s Gemini 3.5 Flash model and Anthropic’s Claude Sonnet 5. which the company rolled out on Tuesday. Anthropic said Claude Sonnet 5 “runs autonomously at a level that just a few months ago required larger and more expensive models.”.
Inside enterprises, the way throttling shows up can differ. One company UBS analysts spoke with, they write, believes the industry is moving away from a phase of AI experimentation.
For that firm, the issue is no longer whether to use tokens, but how to use them efficiently. UBS quotes the company’s perspective in an excerpt: “The question isn’t whether to use tokens. it’s how to use them efficiently.” The analysts add that “optimization becomes an ongoing engineering discipline rather than a reaction to a budget crisis.”.
Another enterprise described a different adjustment: reducing the number of AI tools. UBS reports an excerpt from a conversation where a CTO “went all in on AI early on,” but now is cutting back.
“We have 5 AI tools internally and all of the LLM products. Like others, we ran into the issue where we have already used most of our token budget for the entire year,” the analysts write the excerpt says. “Now we’re only using 2 AI tools and being careful around usage.”
Taken together, the pattern UBS documents is hard to miss: as token costs rise and budgets come under scrutiny, enterprises are learning to treat AI usage as a resource to manage—not just a capability to turn on.
UBS Karl Keirstead Timothy Arcuri Taylor McGinnis AI spend token spending throttling guardrails enterprise IT CFO CTO Uber Andrew Macdonald OpenAI Anthropic DeepSeek Gemini 3.5 Flash Claude Sonnet 5 next-gen chips token efficiency