Most large organizations plan to increase their use of open-weight AI models to keep costs under control, according to a new report by the Capgemini Research Institute, which found that nearly six in ten companies spent significantly more on artificial intelligence than they had expected or budgeted over the past year.
The report, based on a September 2026 survey of 500 senior executives across 10 countries and 13 industries, found that 59% of organizations spent significantly more on AI than expected. Among the organizations affected, expenditure was 24% higher than expected on average, and 21% of surveyed organizations exceeded their AI budgets by more than 30%. The study's central finding — that most large firms plan higher open-weight AI use to cap bills, as Reuters reported — points to enterprise AI strategies entering a cost-discipline phase. The findings land amid broader AI industry coverage of enterprises rethinking how they buy and deploy models.
Where the money is going
The report highlights a persistent gap in how businesses estimate the total cost of AI deployment. Beyond direct expenses such as model usage and tokens, organizations must account for indirect costs that are easy to underestimate: human oversight of AI outputs, correcting AI-generated errors, and governance work.
Half of the executives surveyed said AI spending had been approved without adequate visibility into its full cost, and only 56% said they were confident in their ability to forecast how AI costs would change as adoption scales. In other words, for most enterprises, AI budgets were set before anyone fully understood what running AI at scale actually costs.
"Cost discipline alone will not create a profitable AI estate, just as value creation without cost control will not be sustainable," said Volker Darius, CEO of Capgemini Invent, in a statement accompanying the report. "Organisations need to measure cost, quality and business value together, and use the most appropriate model for each task."
From single-vendor bets to multi-model strategies
The rising expenditure is prompting businesses to reconsider their AI strategies in structural ways. More than three in five executives expect their organizations to adopt a multi-model strategy over the next 12 months, selecting different AI models based on cost, performance, speed and business requirements.
That marks a shift away from the pattern of the last two years, in which many enterprises standardized on one or two frontier model providers for everything from chatbots to document processing. A multi-model approach lets companies route simple, high-volume tasks to cheaper models while reserving expensive frontier models for work that genuinely requires them.
Nearly three in five organizations also plan to expand their use of open-weight models — models whose parameters are publicly released and can be run on a company's own infrastructure — and smaller, specialized models tuned for specific tasks, according to the report.
Why open-weight models are having a moment
The open-weight segment has grown rapidly as an alternative to proprietary, API-only models. Because open-weight models can be self-hosted, organizations with steady, predictable workloads can avoid per-token billing and negotiate a different cost curve — trading infrastructure and operations costs for predictability. Industry analyses have estimated that tokens from open-weight models cost a fraction of comparable proprietary ones, though the total economics depend on utilization and in-house engineering capability.
The Capgemini findings suggest the calculus is tipping further toward open-weight and small models as AI spending grows faster than the value it demonstrably creates. Cost pressure, in this framing, is not a rejection of AI — 59% of firms overspent because they were using more AI than planned, not less — but a demand for precision about where each model dollar goes.
What it means for the AI market
The report's implications cut in several directions at once. For proprietary model vendors, it signals that enterprise buyers are increasingly willing to diversify away from premium-priced frontier APIs where a smaller or open model will do. For cloud and infrastructure providers, multi-model strategies promise more heterogeneous workloads — and more demand for the serving tools that make switching between models cheap.
For enterprises themselves, the message is that AI cost management is becoming a discipline of its own, closer to financial planning than to experimentation. The report's finding that half of AI spending was approved without full cost visibility suggests many organizations will need better measurement before they can optimize.
The Capgemini Research Institute report adds to a growing body of survey evidence that the enterprise AI market's next phase will be defined not by who adopts AI, but by who deploys it profitably — and that the model mix inside large organizations is about to get a lot more varied.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
