Nearly 6 in 10 plan cheaper AI
Published 2026-11-19
Nearly 6 in 10 senior executives at large companies plan to use more open-weight AI models over the next year.
Open-weight tokens were 87% cheaper than closed ones, on average.
You pick the models your team pays for. What follows is one workflow to check, without dropping closed models where the task is complex.
Reuters reported that nearly 6 in 10 senior executives at large companies plan to increase use of cheaper open-weight AI models over the next year, as AI bills overshoot budgets. The study is Capgemini's.
The bill already blew
Capgemini surveyed 500 executives at enterprises with annual revenue of more than $1 billion, in 10 countries.
59% said AI spending over the past year was significantly above expectations or budgets, by 24% on average (Charlotte Observer). CFO Tech reported the same survey of 500 C-suite executives: 59% said AI spending over the past 12 months was significantly higher than expected.
53% were worried about AI costs, up from 45% last year.
The response in that Reuters report is open-weight. Nearly 6 in 10 plan to use more of it over the next year.
SAP's finance chief said last month it was replacing some of its costliest AI models with cheaper ones where performance held up. Anthropic warned in its prospectus that customers may shift to cheaper options, including open-weight models.
The money
An open-weight model is one whose weights you can run without staying inside a closed vendor's product. A closed model is the vendor's own.
Capgemini says open-weight tokens were on average 87% cheaper than closed ones.
Cheaper tokens did not mean most of the money. For traffic against spending, Capgemini cited OpenRouter, a popular multi-AI provider: open-weight models handled 72% of its traffic in September but that was just 27% of AI spending.
| Measure | Share |
|---|---|
| Open-weight traffic on OpenRouter in September | 72% |
| Open-weight share of OpenRouter AI spending | 27% |
On OpenRouter, most of the traffic was open-weight. Most of the AI spending was not.
The miss already hit the accounts. Two-thirds of organizations that overshot their AI budget reported negative financial consequences, such as lower free cash flow or cuts to discretionary spending.
A lower token rate only caps a bill someone can see. Six in ten organizations lack a consistent, enterprise-wide view of their AI costs. Among executives responsible for AI cost management, only 51% measure the cost of human oversight, skills and workflow change. Just 48% measure ongoing operations, retries, failures and rework.
So what?
If you choose what the team buys, the closed-model premium is now a line someone above you can question.
Half the respondents said they would retain closed models where the complexity of the task warranted the additional cost. So the plan keeps closed models for complex work and questions them on the rest.
Move the routine work
This is for the operator or product lead who picks the models the team keeps, and pays for that choice in budget and lock-in.
The job: sort tasks that still need a closed model from tasks that might not.
- Split the team's AI work into complex tasks and routine tasks.
- Keep closed models on the complex list.
- Move routine volume toward open-weight, and put the 87% token gap next to those tasks.
- Ask which closed-model jobs still earn that gap.
If you are building what those buyers use, price the hard task. Nearly 6 in 10 plan more open-weight, and half would keep closed models where the task is complex.
The catch
More open-weight traffic does not by itself shrink the bill: on OpenRouter, open-weight was 72% of traffic and 27% of spending.
What the executives stated is a plan to use more open-weight, not a measured cut. Among the 59% who said spending was above expectations or budgets, the miss was 24% on average. The sources do not show whether the plan will lower the total bill.
Two-thirds of organizations that already overshot reported lower free cash flow or cuts to discretionary spending.
Try one workflow
Pick one workflow the team already pays for.
Write down whether the task is complex enough to keep a closed model, and whether you can name that workflow's share of AI spend.
It worked if both answers are written down. If you cannot name the spend share, leave the model where it is until finance can.
Nearly 6 in 10 senior executives plan more open-weight as AI bills overshoot budgets. The check is one workflow, not the whole company.
Figures are from Capgemini's survey of 500 executives, reported by Reuters via the Charlotte Observer and by CFO Tech, and from Capgemini findings summarized by Olakai. Not a recommendation to buy or sell any vendor.
All source links
- cfotech.co.uk/story/capgemini-warns-ai-spending-is-running-well-over-budget checked 13 November 2026.
- charlotteobserver.com/news/business/article318042264.html checked 13 November 2026.
- olakai.ai/blog/capgemini-ai-cost-optimization-ai-yield checked 13 November 2026.
