Most large firms plan higher open-weight AI use to cap bills, study shows

Nearly 60% of senior executives from large corporations plan to increase their usage of cheaper open-weight AI models within the next year, as reported by a study conducted by Capgemini. This shift comes as AI expenses are surpassing company budgets, a trend that mirrors SAP's recent decision to replace some of its most expensive AI models with cheaper alternatives when performance remains satisfactory.

Anthropic, a company that recently launched its IPO roadshow, has also cautioned in its prospectus that customers may opt for less expensive options, including open-weight models. These models, accessible for download and execution on any device, contrast with closed models such as Claude, ChatGPT, or Gemini, which are exclusively available through cloud services.

According to the survey, 59% of the executives indicated that AI expenditures during the previous year exceeded their expectations or budgets by an average of 24%. Simultaneously, 53% of the respondents expressed concerns over AI costs, a sentiment that rose to 45% last year. Open-weight models have dominated 72% of the traffic on OpenRouter, a widely used platform for multi-AI, although they only accounted for 27% of AI spending.

The study revealed that open-weight tokens are, on average, 87% more cost-effective than their closed counterparts.

Despite the cost advantages, half of the surveyed executives stated they would retain closed models when the complexity of a task necessitates the higher expenditure. Capgemini, a firm specializing in assisting companies in integrating AI into their existing platforms, polled 500 executives at enterprises with annual revenues exceeding $1 billion across ten countries.