A Capgemini survey found that 58% of executives answering its model-portfolio question expect greater use of open-weight artificial intelligence (AI), while 59% of the full survey said AI spending was significantly above expectations or budget.
The Capgemini Research Institute surveyed 500 CXO-level executives in September 2026. They worked at organizations with annual revenue above $1 billion across 10 countries and 13 industries. The model-portfolio question went to 259 executives with responsibilities spanning AI cost management, strategy, technology or procurement.
Among those 259 respondents, 61% expected their organizations to adopt a multi-model strategy, and 50% anticipated greater use of small or specialized models. Capgemini said companies consider cost, capability, latency and risk when choosing models for different needs.
In its September summary of OpenRouter gateway data, Capgemini put open-weight models at 72% of tokens and 27% of spending. Vercel’s October index, based on its AI Gateway data through September, showed a similar gap: 71.5% of tokens and 23% of spending. Vercel estimates spending from published list prices, which can differ from actual bills.
Open-weight models have published weights that others can run or adapt. Their licenses vary, and operating them on a company’s own infrastructure carries costs. Those factors matter as organizations assess AI expenses beyond model use: half of executives said spending was approved without adequate visibility into total costs, including human oversight, rework and governance. Only 56% were fully confident they could forecast how costs would change as adoption grew.
Concern about AI costs reached 53% of executives, up from 45% a year earlier. Volker Darius, CEO of Capgemini Invent, said companies should manage AI as “a portfolio of economic outcomes.” Among organizations that exceeded their AI budgets, two-thirds faced negative financial consequences, including lower free cash flow or reduced discretionary spending.
