Geography Colloquia Series - All that is compute transforms into concrete: Monopoly rent and the fixity of AI's infrastructure

About

In the capitalist imaginary, artificial intelligence is narrated on dematerialization. In this telling, capital has escaped the weight of human labor via the frictionless circulation of data. This paper argues the opposite: the AI economy is a geography of fixity, where training large language models requires warehouse-scale data centers anchored to specific territories by the demands of energy, infrastructure, and available land. Two analytical categories theorize this fixity. The first concerns the monopoly control of location and network proximity: value is extracted from whoever sits closest to the data, controls the interconnection, and commands the lowest latency. The second concerns the monopoly control of computation itself: training frontier AI models is capital-intensive, technically scarce, and concentrated in a handful of firms who can extract a tribute that is not reducible to productivity but to infra/structural position. Both forms of extraction are rent. And both are responses to a profitability crisis in the AI sector driven by rising fixed capital requirements that outpace returns. We draw on financial filings, industry reports, and field observations in Northern Virginia's Data Center Alley to show that what appears as an agglomeration economy is better understood as rent extracted through the strategic fixity of capital in digital infrastructure. Dillon Mahmoudi (dillonm@umbc.edu) is an Associate Professor of Geography and Environmental Systems at the University of Maryland Baltimore County. and Alan Wiig (alanwiig@ufl.edu) is an Associate Professor of Geography at the University of Florida. They recently published a book chapter, Land for AI: Data Center Real Estate Markets in The Inner World of Artificial Intelligence (2026).