Critical Cartography of Frontier AI: A Metabolic Map of OpenAI
Name
ayub-xayub98-mcp-dusp-2026-thesis.pdf
Size
5.19 MB
Format
Adobe PDF
Checksum (MD5)
274879bb18fa3f879c7515a862bef7fd
Author(s)
Ayub, Xavier
Advisor(s)
Wendel, Delia
Date Issued
May 2026
Publisher
Massachusetts Institute of Technology
Abstract
Planning’s engagement with artificial intelligence has concentrated on harnessing the benefits and managing the risks of AI’s downstream applications, or on individual upstream nodes such as data centres, rather than on the production system as a whole. That system, spanning mineral extraction, semiconductor fabrication, hyperscale data centres, contracted annotation labour, and scraped training data across multiple jurisdictions, has gone largely unmapped. This thesis builds a digital metabolic map of OpenAI’s infrastructure as a connected planetary socio-technical-natural system. Extending the methodology Crawford and Joler developed in their 2018 Anatomy of an AI System which mapped the Amazon Echo, it is, to the best of the author’s knowledge, the first comprehensive cartography of a foundation-model AI firm’s production system at planetary scale.
The scaling logic governing frontier AI development since the emergence of foundation models in the early 2020s produces three structural dispositions in the firms operating under it: expansion, externalisation, and opacification. Companies obscure these behind what Keller Easterling calls infrastructure stories — narratives of universal benefit, military necessity, and free-market liberty that occupy the space where accountability claims should originate. The map renders all three dispositions visible, making it possible to read OpenAI’s own policy documents against what the system actually does.
The mapping intervention is informational. International AI governance is fragmented and concentrated downstream, leaving the upstream layer effectively governed by the firms themselves. Planning, permitting, environmental, and labour authorities, alongside civil society organisations and affected communities, hold partial standing over individual nodes but lack any shared picture of the system they are governing in pieces. Adapting Ananny and Crawford’s argument that algorithmic systems must be understood by looking across their components as an assemblage, the thesis names this missing operation systemic legibility. Without it, the proactive and coordinated bottom-up governance that frontier AI requires has no informational ground to stand on. The map opens avenues for planners, civil society, and a broader public engaged in critical AI literacy, and is offered as an analytical case and transferable framework alongside existing AI transparency, risk, and harm-mapping instruments.
MIT Department
Massachusetts Institute of Technology. Department of Urban Studies and Planning
Terms of Use
In Copyright - Educational Use Permitted
Copyright retained by author(s)
Persistent DSpace Link