UBS says the majority of enterprise companies it's talked to recently are 'throttling AI spend'

Recent discussions with enterprise IT leaders have revealed a notable trend: a significant portion of companies are implementing measures to control their artificial intelligence expenditures. According to a report from UBS analysts Karl Keirstead, Timothy Arcuri, and Taylor McGinnis, approximately 60% of enterprises are now establishing some form of financial guardrails around their AI investments. The concern primarily stems from escalating costs associated with AI, which have caught the attention of CFOs and CTOs alike. Uber's Andrew Macdonald highlighted the challenges of justifying these rising expenses in light of disappointing returns on investment. The UBS analysts noted that their conversations, initiated in early June, pointed to a developing challenge for organizations, although the severity of this impact varies significantly. They observed that while some companies are experiencing a considerable slowdown in spending, others are either at the beginning of their AI journey or are already deeply invested and unwilling to restrict usage due to perceived returns on investment or a commitment to innovation. The analysts indicated that AI model developers, including major players like OpenAI and Anthropic, might be particularly vulnerable to these budget cuts. Interestingly, the report identified open-source and Chinese AI models, such as DeepSeek, as potential beneficiaries of this shift, especially for enterprises seeking solutions for tasks that do not involve coding. Despite these cutbacks, the analysts refrained from sounding alarm bells, framing the situation as a manageable issue rather than a crisis. They emphasized that optimizing AI spending is a normal and healthy response, asserting that the deployment of AI is not slowing down. Furthermore, they suggested that advancements in AI model efficiency and new technologies could lead to reduced costs. Notably, Google has introduced its Gemini 3.5 Flash model, while Anthropic launched Claude Sonnet 5, which reportedly operates at a level previously requiring more expensive systems. The UBS analysts concluded that the focus is shifting from merely using tokens to using them more effectively, with optimization becoming a standard engineering practice rather than a reaction to financial constraints. One company reported cutting back from five AI tools to two, illustrating the careful approach many are now taking in managing their AI resources.