Nearly six in ten senior executives at large companies intend to increase their use of open-weight artificial intelligence models over the next 12 months as businesses seek to manage AI spending that has exceeded initial budgets, according to a study published by Capgemini on Thursday.
The research found that 59% of executives reported AI expenditure over the previous year had been significantly higher than anticipated, exceeding expectations or allocated budgets by an average of 24%.
The findings indicate a growing focus on the costs associated with deploying artificial intelligence across business operations, particularly as companies evaluate alternatives to proprietary AI systems.
Capgemini surveyed 500 senior executives at companies generating more than $1 billion in annual revenue across ten countries.
AI Spending Exceeds Corporate Budgets
The study identified rising expenditure as a factor influencing companies’ decisions about which AI models to deploy.
According to Capgemini, 53% of surveyed executives expressed concern about the cost of artificial intelligence, compared with 45% in the previous year’s research.
The increase suggests that spending considerations are becoming more prominent as organisations expand their use of AI applications.
Companies are examining whether less expensive models can perform certain tasks without materially affecting the quality of their results.
The findings reflect a broader effort to balance the operational benefits of artificial intelligence with the costs of running increasingly sophisticated models.
Open-Weight AI Models Offer Lower Operating Costs
Open-weight AI models, including those developed by DeepSeek and Mistral, allow organisations to download model weights and deploy the systems on infrastructure of their choosing.
This differs from proprietary models such as those offered through Claude, ChatGPT and Gemini, which are generally accessed through cloud-based services.
Capgemini reported that tokens processed using open-weight models were, on average, 87% less expensive than those handled by closed models.
Tokens are the units of text or other information processed by AI systems, and token-based pricing is commonly used to calculate the cost of model usage.
The difference in processing costs provides a financial incentive for companies to evaluate open-weight alternatives, particularly for applications involving large volumes of AI requests.
However, the suitability of individual models depends on the requirements of the tasks being performed.
Open-Weight Models Account for Majority of OpenRouter Traffic
Capgemini cited usage figures from OpenRouter, a platform providing access to multiple AI models, to illustrate the difference between usage volumes and expenditure.
In September, open-weight models accounted for 72% of traffic on the platform but represented only 27% of total AI spending.
The figures indicate that open-weight models handled a larger proportion of activity while accounting for a smaller share of expenditure.
This difference is consistent with the lower average token-processing costs identified in the study.
The OpenRouter figures provide an example of model usage and spending patterns on one platform rather than a measure of the entire AI market.
SAP and Anthropic Highlight Changes in AI Model Selection
The findings are consistent with developments reported by major technology companies.
SAP’s chief financial officer said last month that the enterprise software group was replacing some of its most expensive AI models with lower-cost alternatives where the performance of those alternatives remained suitable.
The approach reflects an effort to reduce expenditure without compromising the requirements of individual applications.
Anthropic also identified potential customer migration toward less expensive AI systems, including open-weight models, as a business risk in its prospectus.
The company began its initial public offering roadshow this week.
These developments illustrate how model pricing and performance are influencing purchasing decisions among enterprise AI users.
Companies Continue to Use Proprietary Models for Complex Tasks
Despite growing interest in open-weight systems, Capgemini’s research indicated that businesses do not necessarily intend to replace proprietary models entirely.
Half of the surveyed executives said they expected to retain closed AI models for applications where task complexity justified the additional expenditure.
This suggests that companies are increasingly considering a combination of different AI systems rather than relying exclusively on a single model provider.
Lower-cost models may be selected for tasks that can be completed within their performance capabilities, while more expensive proprietary systems may continue to be used for applications requiring additional functionality.
The research indicates that cost, performance and task requirements are becoming important considerations in enterprise AI deployment decisions.
Enterprise AI Spending Strategies Continue to Evolve
Capgemini, which provides consulting and technology services to companies integrating AI into existing systems, conducted the study among executives at large multinational businesses.
The findings show that a majority of respondents experienced higher-than-expected AI costs during the previous year, while concern about expenditure increased compared with the earlier survey.
The reported cost differences between open-weight and proprietary models provide one explanation for the growing interest in alternative deployment approaches.
At the same time, the continued use of closed models for more complex applications indicates that companies are evaluating AI systems according to both financial and operational requirements.
The results suggest that enterprise AI purchasing decisions are increasingly influenced by the relationship between model performance and the cost of processing workloads.