Where firms can create value with AI and why open-weight models matter

Europe’s AI strategy has focused heavily on expanding the supply of computing capacity and advanced models. This, however, often leaves the demand-side question open. How can value be created with AI? Which capabilities are necessary to create this value? 

In a new study written with the support of the think tank Leitmotiv, we want to add to the demand-side perspective and examine these questions in three areas: software production, consumer-facing banking services and private equity.  

Drawing on desk research and interviews, we distinguish two mechanisms of value creation. Cost advantage arises when AI reduces the resources required to perform an activity, for example through shorter processing times or fewer errors. Differentiation arises when it improves products or services in ways that customers are willing to pay more for, such as higher quality or additional functionality.  

Across the three cases, the evidence is considerably stronger for cost advantage than for differentiation. In software production and private equity, AI increases productivity by reducing the time required for coding, research, screening and other recurring activities. In consumer banking, automation of routine interactions provides a plausible route to lower service costs. 

By contrast, there is little evidence so far that AI enables firms to charge higher prices or achieve durable differentiation. 

Differentiation could, however, offer greater potential. By enabling firms to offer products and services that customers value more highly, AI could increase their willingness to pay, allowing firms to charge higher prices and potentially expand the market by attracting new demand. The cases point to this potential but provide little evidence that it has yet been realised. 

The activities through which firms create value also determine which model capabilities they need. The study compares benchmarks and shows that leading open-weight models perform close to leading proprietary models in software engineering, consumer-facing banking services and financial analysis. They therefore offer credible alternatives for these applications. 

The selected open-weight models are also substantially cheaper to access. In software engineering and financial analysis, leading open-weight alternatives cost less than one-tenth as much per output token as the strongest proprietary models. 

For Europe, open-weight models are particularly important because they can help firms retain more of the value they create with AI. Where they provide the required capabilities at lower inference prices, European firms can spend less on model access and retain a larger share of the value created through higher productivity.  

Open-weight models can also increase competition among AI inference providers. Where licences permit independent hosting, several providers can serve the same model, creating opportunities for European cloud and inference providers and application developers. This gives firms greater choice among suppliers and more control over how models are adapted and deployed. 

The current combination of capable models, low prices and independent provision may, however, not endure. Revenue earned by independent inference providers does not automatically finance development of the models they serve. Developers seeking revenue may tighten commercial licences and restrict independent hosting. Insufficient funding could also reduce the availability of future capable open-weight models. 

Ensuring the long-term availability of widely accessible open-weight models should therefore be a priority for European policymakers. This includes continued access to capable open-weight models under commercially usable licences, competition among inference providers and sufficient financing for future open model development. Whether current funding and revenue models can sustain this market structure over the long term remains an open question, which we will examine in a forthcoming study.