« Infrastructure » est une journée d’étude sur les aspects souvent négligés des infrastructures qui sous-tendent le cloud computing des systèmes d’intelligence artificielle. La journée se concentre sur les impacts environnementaux et les enjeux socio-techniques générés. Cette journée vise à réunir des chercheurs des sciences informatiques, en Science and Technology Studies (STS), et des disciplines associées, pour étudier avec une perspective critique les fondations matérielles, environnementales et socio-techniques de nos systèmes informatiques. 

5 December 2025 (pre-workshop dinner on 4 December)

Location: Nantes, France

Argument

Over the past two decades, the cloud computing paradigm and its evolutions have become dominant—sometimes even treated as a utility service. This success stems from the ability of cloud systems to abstract complex computing infrastructures into manageable, higher-level “resources.”

However, this abstraction renders the physical and human infrastructure behind the cloud invisible. All cloud computing systems ultimately rely on servers, network equipment, cables, datacenters, distributed infrastructure software, and the skilled workers who design and maintain them. While abstraction simplifies usage, it introduces several fundamental challenges:

Environmental Impacts

The infrastructure supporting cloud services has rapidly grown in environmental footprint, especially with the rise of AI. Massive datacenters are being constructed, consuming power at the gigawatt scale, alongside a sharp increase in demand for GPU-class hardware.

The Illusion of Unlimited Resources

The elastic model of the cloud promotes the idea of infinite computing capacity. This disconnect between usage and physical infrastructure encourages overconsumption and obscures its material costs. As noted in the 2011 NIST definition of cloud computing:“The capabilities available for provisioning often appear to be unlimited and can be appropriated in any quantity at any time.” 

Governance and Power Asymmetries

Centralized infrastructure controlled by a few providers leads to uneven power dynamics—illustrated by cases like the VMware/Broadcom fallout. When infrastructure is fully outsourced for simplicity, clients lose autonomy and critical skills become concentrated. Governance structures deeply shape users’ rights and influence the control they have over infrastructure, including sensitive matters such as access by intelligence agencies.

Interdisciplinary Contributions

Submissions are encouraged from Computer Science, STS (Science and Technology Studies), and other humanities and social sciences. The objective is to foster a genuinely interdisciplinary dialogue on the hidden material, technical, and political aspects of today’s computing infrastructures.

Submission format

- Submissions should be short position papers of up to two pages (excluding references).

- Submissions must be in PDF format with all fonts embedded.

- Use the following template: A4 page, 11pt, one-column, 2 cm margins on all sides.

- Submissions are not anonymised.

- Text inside figures must be readable when printed.

- Figures should use colorblind-friendly palettes and remain interpretable when printed in black and white.

- By default, accepted contributions will be published on the open archive HAL, unless authors explicitly opt out.

- Generative AI policy: authors must clearly document any use of generative AI tools (text, figures, data). Responsibility for the content rests fully with the authors.

Important Dates

Paper-related 

• 24 October 2025, 23:59 AoE (Anywhere on Earth) – Paper submission deadline

• 31 October 2025 – Deadline for travel grant applications – Via registration form
• 7 November 2025 – Notification of acceptance

Participation-related

• 19 November 2025 – Deadline for mandatory registration – Via registration form
• 4 December 2025 – Pre-workshop dinner
• 5 December 2025 – Workshop

Scientific Committee

• Hélène Coullon, IMT Atlantique, France
• Adrian Friday, Lancaster University, UK
• Valérian Guillier, CNRS, France
• Pierre Jacquet, ÉTS Montréal, Canada
• Anne-Cécile Orgerie, CNRS, France
• Matthieu Simonin, Inria, France