Material Infrastructures

Big data reaches users in seemingly immaterial forms, but its production, storage, and availability at scale depend on an extensive physical infrastructure. Storage, computation, and transmission rely on hyperscale data centres, electrical grids, cooling and water systems, and fibre networks. Decomposing these systems leads further to processors and storage hardware, semiconductor fabrication, critical minerals, manufacturing, and global supply chains. Material Infrastructures treats these dependencies as conditions of digital information itself, placing the infrastructures that sustain computation and the institutions that make them possible within a common field of inquiry. Source criticism at this scale extends accordingly to the conditions under which information can be stored, processed, moved, and retrieved.

Energy makes one dimension of these dependencies measurable. BDSL’s work on data energetics traces the relationship among computational scale, efficiency, and energy demand, where growth in one does not translate proportionately into another. Between 2010 and 2018, global data-centre storage capacity increased twenty-fivefold while watts per unit of storage fell by a factor of nine. Such efficiency gains acquire meaning only in relation to expanding demand, more intensive forms of computation, and the continuing growth of hyperscale infrastructure. Generative AI has sharpened this problem as rapidly rising demand for computation, accelerators, memory, power, and data-centre capacity strains supply even amid concerted efforts to improve efficiency and expand production. The pattern raises an older question associated with the Jevons paradox, namely under what conditions efficiency gains reduce aggregate resource consumption and when they enable still greater use.

The concentration of storage and computational capacity among a small number of firms brings ownership and political economy directly into the inquiry. This concentration began well before generative AI, as Web 2.0 and cloud computing consolidated infrastructure and computational resources among a relatively small group of technology companies. Large-scale AI is intensifying both the concentration and the demand for specialized processors, electricity, cooling, data-centre construction, critical minerals, and the supply chains that connect them. BDSL studies ownership, capital, industrial policy, digital sovereignty, and geopolitics as parts of the same infrastructural arrangements as servers, fibre, electricity, water, and hardware. Following these dependencies opens computation onto a much larger material geography, connecting what happens at an interface to resources, facilities, firms, states, and communities distributed across the world.