60% of Enterprises Now Throttling AI Spending, UBS Finds

A majority of large enterprises are pulling back on artificial intelligence spending as token costs mount faster than expected returns, according to new research from UBS analysts.

Based on conversations with more than a dozen enterprise IT executives in recent weeks, approximately 60% of companies have implemented spending guardrails around AI deployments, UBS analysts Karl Keirstead, Timothy Arcuri, and Taylor McGinnis reported. The trend represents what they characterize as a "modest emerging headwind" for AI adoption, though the severity varies significantly across organizations.

Token budgets running dry

Token spending—the unit of measurement for AI model usage—has emerged as a flashpoint for corporate finance teams. Some companies told UBS they've already exhausted most of their annual token budgets partway through the year, forcing immediate cutbacks.

One enterprise reported to the analysts that it had deployed five AI tools internally along with multiple large language model products, only to burn through its yearly token allocation prematurely. The company has since reduced its toolkit to just two AI applications and instituted strict usage controls.

The cost pressure isn't theoretical. Uber's operations chief Andrew Macdonald said in May that justifying rising AI expenses had become increasingly difficult given "pretty meager ROI," according to the UBS report.

Winners and losers in the pullback

The spending constraints are likely to hit proprietary model providers like OpenAI and Anthropic hardest in the near term, the analysts noted. Meanwhile, open-source alternatives and Chinese models such as DeepSeek stand to benefit, particularly for non-coding enterprise applications where cutting-edge performance may be less critical.

Some companies are avoiding throttling altogether—either because they're too early in deployment to face budget pressure, or because they've achieved sufficient ROI to justify continued investment. Organizations with explicit innovation mandates are also maintaining spending levels despite the costs.

A "healthy problem," not a crisis

Despite documenting the pullback, the UBS analysts emphasized they're "not ringing the alarm bells." They view the optimization trend as a natural maturation of enterprise AI adoption rather than a fundamental threat.

"Some measure of AI spend optimization is normal, no one is hitting the brakes on AI deployment," they wrote. The analysts also noted that next-generation chips and more efficient models could drive token costs down further.

Major AI providers are already responding to cost concerns. Google offers its Gemini 3.5 Flash model as a lower-cost option, while Anthropic released Claude Sonnet 5 this week with the promise of delivering performance that previously required "larger and more expensive models."

One executive told UBS that the industry is shifting from experimentation to optimization: "The question isn't whether to use tokens, it's how to use them efficiently. As a result, optimization becomes an ongoing engineering discipline rather than a reaction to a budget crisis."

Why it matters

The spending pullback signals that enterprise AI adoption is entering a more mature phase where financial discipline matters as much as innovation. For AI vendors, the shift means competing on cost efficiency and demonstrable ROI, not just capability. For enterprises, it marks the end of open-ended AI experimentation and the beginning of harder questions about which use cases actually justify their price tags.

These findings were first reported by Business Insider based on the UBS analyst report.