New white paper: Public AI as a democratic alternative to the concentration of private power

Gütersloh, May 20, 2025 The most powerful AI systems of our time were developed and are controlled by a small number of private companies – including OpenAI, Anthropic, Google DeepMind, Meta, and DeepSeek. These companies don’t just dominate the development of models, they control the basic infrastructure of the AI ecosystem: computing capacities, training data, and cloud services. This concentration of power is not just a technological reality – it is a political challenge. It raises a central question: Who designs the systems that are increasingly influencing our society?

To counteract this growing imbalance, the Bertelsmann Stiftung, with the support of Open Future, has presented a new white paper on Public AI. This publication outlines a strategic and actionable framework for an alternative approach to the development and application of AI, based on greater transparency, open access to critical infrastructure, and a stronger focus on the common good.

Societies should not just use AI, they should actively shape it

Public AI is not a rejection of private innovation. Rather, it is a proposal to rebalance the dynamics of power. Societies should be put in a position to not just use AI, but to actively shape it. It is not sufficient to ensure that AI is safe for the public to use. Rather, shared responsibility must be taken for AI by means of democratic structures, which can in turn be used to control it.

At the heart of the proposed strategy is an ambitious but necessary goal:
ensuring the permanent existence of at least one complete open source model with AI capabilities close to those of leading proprietary systems. Without this sort of AI model, public actors will remain limited in their ability to act. The white paper identifies three key recommendations for implementation:

• development and/or fortification of complete open source models and the open source ecosystem
• creation of public computing infrastructure to support the development and use of open models
• scaling investment in AI skills to attract sufficient skilled workers to develop and apply these models

The white paper also develops three policy intervention paths along the structure of the AI stack, which consists of computing power, data and models:

• Compute path – establishing public computing capacities, ensuring access for open projects and coordinating national and supranational infrastructure initiatives (e.g., EU AI factories)
• Data path – development of high-quality data sets as digital commons, managed by governance models oriented towards the common good with mechanisms to protect against misuse
• Model path – promotion of an ecosystem of open source models – both powerful “capstone” models and specialized smaller models – with long-term financial and infrastructure support

These paths are supplemented by cross-cutting measures – such as promoting talent development or the financing of open software.

Systematically evaluating AI initiatives

The white paper presents a strategic tool for evaluating and managing AI initiatives: the “Gradient of Publicness”. This framework can be used to systematically evaluate AI – based on openness, governance structures, and a focus on the common good. The Gradient of Publicness makes it possible for political decision-makers to classify existing and new projects along a spectrum from private to public and to derive concrete steps to increase the public benefit.

Public AI offers a realistic and necessary alternative to the AI dominance of private actors. But it will not happen by itself. It requires strategic investment, institutional coordination, and political will. This white paper is intended to provide the impetus for an international debate that sees AI as a public endeavor.