Send message to Author

Short abstract

The paper examines how, in the current frenzy for scale in the AI sector, specialized architectures and chips co-create a material-epistemic assemblage that raises major concerns over AI's ecological footprint. We trace contrasting framings and valuation registers in this field of tensions.

Long abstract

Recent STS scholarship has called for renewed critical engagement with the logics and politics of scaling, suggesting that contemporary innovation cultures are underpinned by a "scalability zeitgeist" (Pfotenhauer et al., 2022) that prioritizes expansion and exponential growth. This "imperative to scale" (MacKenzie, 2026) is also embedded in current AI culture, where – since the paradigm shift to massive parallel processing (see Vaswani et al., 2017) – a dominant logic has prevailed: "the more compute and data you put in, the more intelligence you get out" (Patel, 2025). Driven by expectations of vast future returns, this brute-force approach to AI has triggered an infrastructure arms race, with unprecedented capital investments in data centers promising ever more potent iterations of foundation models.

This paper problematizes the current frenzy for scale in the AI sector by investigating the environmental overflows and externalities (Callon, 1998; Felt, 2025) of the "bigger is better" doctrine. More specifically, we show how the co-evolution of a certain way of 'doing AI' (transformer architecture) and specialized high-performance AI chips has given rise to a material-epistemic assemblage whose production, reliance on critical minerals, energy- and water-intensive operation, and short service life generate significant ecological pressures. Drawing on extensive document analysis and stakeholder interviews, the paper highlights contrasting framings of this assemblage, unpacking the registers of valuation that underlie controversies over the AI industry's escalating environmental footprint and attendant residues.

This research is carried out in collaboration with Ulrike Felt in the framework of her ERC-funded project Innovation Residues (GA 101054580).