Khalatbari's 4-Year GPU Inquiry Exposes AI's Hidden Material Costs

Cyrus Khalatbari, in a paper presented at the 2026 LIMITS workshop, presents a four-year investigation (2022–2026) into the political economy of graphics card (GPU) miniaturization. The work draws on research-creation methods—combining social sciences with speculative, critical, and fictional approaches—to open the black box of the GPU. Khalatbari conducted fieldwork in two contrasting sites: among liquid nitrogen overclockers in Taiwan (who push GPUs to extreme performance) and urban miners in Ghana (who recover valuable materials from e-waste). Additionally, he performed hands-on experiments with over 50 acquired graphics cards to trace their physical life cycles. The research is structured around three themes: dismantle and dissolve, rebuild, and remix.

The study argues that while GPUs enable massive parallel processing for AI training and their miniaturization drives the Internet of Things, the development of these objects reinforces major environmental and social problems. Khalatbari shows how the sociotechnical phenomenon of GPU miniaturization obscures material supply chains, energy consumption, and e-waste generation. By employing research-creation as a rigorous epistemology, the paper demonstrates how to apprehend what seems opaque or inaccessible in AI's infrastructure. It contributes to the field of ICT for sustainability by affirming that creative, reflexive methods can disentangle the material and environmental dependencies that computational systems depend on but hide.