Cost Pressure
The management of infrastructure spending related to artificial intelligence is identified as the main concern for executives at large companies globally. According to a Capgemini study reported by the source, nearly six in ten executives expect to increase the use of open-weight AI models over the next year. The reason given is mainly economic: the costs associated with adopting these technologies are growing exponentially and, according to the source, exceed available budgets. The figure is part of a trend already observed in previous surveys conducted during 2026.
Open versus Proprietary: The Crux of the Spending Model
The difference between the two options concerns not only performance but also the cost structure. Open-weight models allow organizations to host the models independently, shifting spending toward more predictable infrastructure instead of paying for each individual interaction through proprietary APIs. In this reading, licensing costs fall because calls to external services are replaced by self-hosted inference. The source flags the opposite risk: companies could pay more than necessary by continuing to use costly proprietary API calls when cheaper alternatives exist.
On the technical side, the performance gap between open models and frontier proprietary models is said to have narrowed significantly in 2026, making the former suitable for a broader range of enterprise use cases. The distinction that emerges from the material is by task: smaller or open models would efficiently handle simple, high-volume activities, while larger proprietary models would remain preferable for complex reasoning and advanced coding. The source also cites advantages in terms of transparency and trust, without detailing them further.
The Resistance of Habits and the Numbers to Recalculate
Despite cost pressure, adoption of open models is not automatic. The source reports that users still choose closed models in 80% of cases, due to entrenched habits and the perception of greater ease of use. This is the point at which the distance between executives' stated intentions and actual behavior is measured, an aspect that the material does not quantify further.
The organizational issue is central: AI budgeting, according to the source, can no longer be treated as a secondary project and requires a structured approach aligned with business objectives. Companies that adopt these techniques can reduce expenses by 30% to 60%, a range that the source does not detail in terms of scope or methodology. It therefore remains to be seen how this saving is distributed among licenses, infrastructure, and personnel, and how much the revision of spending strategies attributed to major technology operators affects it.
The material also does not clarify the sample size of the Capgemini study or the questions asked, elements necessary to assess the robustness of the figure on six in ten executives. Nor is the reference to 80% use of closed models accompanied by indications of source, period, or calculation basis. The picture that emerges is one of a transition driven by the numbers rather than performance, but the available data do not allow determining how far advanced it already is.
