Key facts
- Capgemini surveyed 500 executives at enterprises with annual revenue above $1 billion across 10 countries.
- Some 53% of surveyed executives said they were worried about AI costs, up from 45% a year earlier, Capgemini said.
- Open-weight models handled 72% of traffic on OpenRouter in September but accounted for just 27% of AI spending, per Capgemini.
- Capgemini said open-weight tokens were on average 87% cheaper than closed ones.
- Half of respondents said they would retain closed models where task complexity warranted the additional cost, Capgemini said.
Nearly 6 in 10 senior executives at large companies plan to increase their use of cheaper open-weight AI models over the next year as AI bills overshoot budgets, according to a Capgemini study published on November 12. The IT consulting firm surveyed 500 executives at enterprises with annual revenue of more than $1 billion in 10 countries.
Some 59% of the surveyed executives said AI spending over the past year had been significantly above their expectations or budgets, by 24% on average. Another 53% said they were worried about AI costs, up from 45% a year earlier.
Open-weight models, such as those from DeepSeek or Mistral, can be downloaded and run anywhere, while closed options such as Claude, ChatGPT or Gemini models are only accessible from the cloud. Capgemini said open-weight tokens were on average 87% cheaper than closed ones.
Open-weight models handled 72% of traffic on OpenRouter, a multi-AI provider, in September, though that represented just 27% of AI spending, Capgemini said. Closed models broadly remain in the mix, with half the respondents saying they would retain them where the complexity of the task warranted the additional cost.
The trend echoes SAP, whose finance chief said last month it was replacing some of its costliest AI models with cheaper ones where performance held up. Anthropic, which launched its IPO roadshow this week, also warned in its prospectus that customers may shift to cheaper options, including open-weight models.
