The Hidden Power in Digital Infrastructure

In the contemporary debate surrounding artificial intelligence and Large Language Models (LLMs), attention often focuses on algorithms, models, and their capabilities. However, as Britt Paris, a critical informatics scholar and Associate Professor at Rutgers University, points out, true power often resides at a more fundamental level: the physical infrastructure that underpins the entire network. In her recent book, Radical Infrastructure: Imagining the Internet from the Ground Up, Paris invites us to reflect on the miles of cables, the acres of land occupied by data centers, and all the tangible components that make web browsing possible and, by extension, the execution of complex AI workloads.

This perspective is particularly relevant for technical decision-makers evaluating on-premise deployment options for their LLMs. Understanding the 'physical politics' of the internet means recognizing that control over infrastructure is not just a technical matter, but a strategic one, with direct implications for data sovereignty, compliance, and long-term Total Cost of Ownership (TCO). Paris's vision suggests that the current configuration, dominated by a few large players, is not the only one possible, and that alternatives exist that can better serve the interests of users and organizations.

Beyond Monopolies: The Cooperative Model

Britt Paris's work does not merely critique the status quo but actively explores alternative models. A significant example cited is that of telecommunications cooperatives, such as the one started by her great-great-great uncle in rural Missouri, long before major cities had network access. These initiatives demonstrate how it is possible to build and manage network infrastructures that directly respond to community needs, rather than being dictated by monopolistic profit logics.

Models like NEMR, though not specifically detailed in the source, are presented as concrete examples of how people can actively decide how their internet connection functions. This decentralized, community-oriented approach strongly resonates with the principles guiding many on-premise deployment decisions in the AI sector. The choice to host LLMs and data locally, in self-hosted or air-gapped environments, is often driven by the desire to regain control, ensure privacy and security, and reduce reliance on external cloud providers, replicating at an enterprise scale the logic of autonomy and sovereignty proposed by cooperatives.

Implications for LLM Deployment and Data Sovereignty

The discussion about physical infrastructure and its control has direct implications for the LLM deployment landscape. For CTOs, DevOps leads, and infrastructure architects, the choice between cloud and on-premise is not just a matter of cost or performance, but also of philosophy and strategic control. Paris's approach highlights that the ability to physically manage hardware, network, and data centers offers a level of data sovereignty and resilience that cloud solutions cannot always guarantee, especially in contexts requiring stringent regulatory compliance or high security standards.

Considering infrastructure as a common good, or at least as a strategic asset to be directly controlled, can profoundly influence investment decisions in dedicated hardware (such as GPUs with high VRAM for inference or fine-tuning), the design of local data pipelines, and the management of overall TCO. The ability to optimize resource utilization, customize the environment for specific workloads, and keep data within corporate or national boundaries are all aspects that benefit from a more conscious and 'radical' view of infrastructure.

Future Perspectives: Control and Autonomy in the AI Era

Britt Paris's reflection on infrastructure as the hub of power offers a valuable lens for interpreting the challenges and opportunities of the artificial intelligence era. As LLMs become increasingly pervasive, the question of who controls the underlying infrastructure becomes even more critical. Whether it involves AI-generated information objects, civic data, or digital labor, the physical foundation upon which these technologies operate ultimately determines who benefits and who holds control.

For companies and organizations aiming to leverage the potential of LLMs while maintaining autonomy and security, investing in self-hosted solutions and a deep understanding of infrastructural dynamics are fundamental steps. Paris's approach encourages us to imagine and build a future where technological infrastructure, including that for AI, is designed to serve the interests of a broader audience, rather than consolidating power in the hands of a few. This is a key message for anyone planning their path in the deployment of advanced AI technologies.