Large Firms Turn To Cheaper Open-Weight AI Models As Spending Exceeds Budgets

Large companies are increasingly looking towards cheaper artificial intelligence models as unexpectedly high operating costs put pressure on technology budgets. Nearly six in ten senior executives plan to expand their use of open-weight AI models over the next year, according to a new study by consulting firm Capgemini.

The research, released on Thursday, found that 59% of executives reported significantly higher AI spending than anticipated over the past year. Among companies experiencing cost overruns, expenditure exceeded expectations by an average of 24%.

The findings suggest that businesses are becoming more selective about which AI systems they use. Rather than paying for the most powerful models for every task, companies are examining whether cheaper alternatives can deliver comparable results.

FCRF Launches CP-FRM to Build India’s Next Generation of Fraud Risk Professionals

Rising AI Bills Force Companies To Rethink Spending

Capgemini surveyed 500 senior executives across 13 industries and 10 countries, including India, the United States, China and Germany. All respondents represented organisations generating more than $1 billion in annual revenue.

The study found that concern about AI expenditure has increased considerably. Around 53% of executives are now worried about costs, compared with 45% a year earlier.

Approximately one in five surveyed organisations reported exceeding their AI budgets by more than 30%.

The financial pressure extends beyond the money paid to AI providers. Companies must also spend on infrastructure, software integration, employee training and the human supervision required to check AI-generated results.

For example, an AI system may produce a financial report within seconds. However, employees may still need to verify its calculations, correct errors and ensure that confidential information has been handled appropriately.

These additional expenses can make an apparently inexpensive AI tool considerably more costly to operate.

Capgemini found that only about one-third of executives could confidently estimate indirect expenses such as human supervision, corrective work and governance.

What Are Open-Weight AI Models And Why Are They Cheaper?

Open-weight AI models are artificial intelligence systems whose trained numerical settings, known as weights, are made available for others to download and use.

These settings determine how a model processes information and generates responses. Access to them allows businesses to operate compatible models on their own computing infrastructure or through independent cloud providers, subject to licensing conditions.

Companies such as DeepSeek and Mistral offer examples of open-weight models.

This differs from many proprietary AI services, where businesses pay an external provider to access a model without receiving its underlying weights.

However, open-weight does not necessarily mean fully open-source. Developers may release a model’s weights while retaining restrictions on its use or withholding information about its training process.

Another important concept is the token, a small unit of information processed by an AI system. Providers often calculate usage charges according to the number of tokens processed or generated.

Capgemini estimates that open-weight tokens are approximately 87% cheaper on average than those from closed models.

The difference is already reflected in usage patterns. In September, open-weight models accounted for 72% of traffic on AI platform OpenRouter but represented only 27% of spending on that platform.

However, cheaper tokens do not automatically guarantee lower overall expenditure. Companies operating models independently must still pay for computing equipment, electricity, maintenance and security.

SAP’s Shift Shows How Businesses Are Responding

German enterprise software company SAP provides a recent example of this changing approach.

In October 2026, SAP disclosed that it was replacing some expensive AI models with cheaper alternatives where the quality of results remained satisfactory.

The company uses more than 70 models from different providers. Its strategy involves selecting a model according to the requirements of each task rather than relying on the same expensive system throughout its operations.

SAP expects its broader AI initiatives to help generate more than €2 billion in efficiency gains over two years.

However, its management has also indicated that overall AI spending could continue increasing as more employees and customers adopt the technology.

This distinction is important. Lower costs for individual tasks may improve profitability without necessarily reducing a company’s total technology budget.

Companies Are Not Abandoning Premium AI Models

The research does not suggest that businesses are preparing to replace proprietary AI systems entirely.

Around half of respondents indicated that they would retain more expensive closed models for complicated tasks where their capabilities justify the additional cost.

Instead, companies are increasingly considering combinations of different AI systems.

A business might use a cheaper model to classify routine customer enquiries while reserving a more capable system for complicated financial analysis or software development.

For Indian enterprises, this approach could provide greater flexibility over technology expenditure and the handling of sensitive business information.

Nevertheless, operating downloadable models independently transfers additional responsibility to the organisation. Businesses must manage software updates, access controls, licensing requirements and cybersecurity protections.

The emerging challenge is therefore not simply identifying the cheapest AI model. It is finding the system that delivers reliable results at an acceptable overall cost.