The popular narrative of the 'cloud' as a disembodied, frictionless digital realm is being dismantled by new academic research that exposes the heavy material costs of AI infrastructure. A recent article published in the journal Geo argues that the expansion of hyperscale data centers for generative AI is driving 'hydrosocial dispossession,' a process where water resources, governance authority, and local knowledge systems are stripped from communities to serve computational demand. This framework challenges the tech industry’s tendency to view water merely as a fungible commodity, highlighting instead how AI’s resource appetite reinforces colonial-capitalist logics of extraction and inequality.

The Myth of Weightlessness and Material Reality

The research posits that the vocabulary of digital expansion cultivates a dangerous mythology of weightlessness, obscuring the physical reality of data centers, submarine cables, and semiconductor fabrication plants. These facilities rely on the extraction of finite resources, specifically water, energy, and minerals. The article cites estimates from de Vries-Gao (2026) suggesting the global water footprint of AI systems in 2025 ranges between 312.5 and 764.6 billion liters. However, the study notes that current disclosure practices prevent this massive figure from being resolved to specific facilities or watersheds, creating a significant blind spot in environmental accountability.

Defining Hydrosocial Dispossession

The core contribution of the paper is the concept of hydrosocial dispossession, which operates across three distinct registers: material, governance, and epistemic. Material dispossession involves the physical diversion or contamination of water for cooling and energy, severing the relations between households, farmers, and ecosystems. Governance dispossession occurs when corporate operators use non-disclosure agreements (NDAs) with local utilities to hide water and energy usage data from elected officials and residents. Epistemic dispossession denies the validity of Indigenous and local knowledge systems, reducing water to a technical resource management challenge and ignoring its sacred or relational dimensions.

The Role of NDAs and Informational Opacity

A critical mechanism identified in the study is the use of NDAs by hyperscale cloud service providers as a condition for facility siting. These agreements prevent the disclosure of water and energy use data to the public, creating a 'private informational regime' where decisions about public water allocation are made by corporate executives and contracted administrators. This opacity ensures that corporate volumetric accounting becomes the only authoritative record of water use, effectively silencing community concerns and making it difficult for affected populations to contest the environmental impact of data centers.

Key Takeaways

  • The 'cloud' is not weightless; AI infrastructure consumes an estimated 312.5 to 764.6 billion liters of water annually.
  • Hyperscale providers use NDAs to hide water usage data from public scrutiny and elected officials.
  • The concept of 'hydrosocial dispossession' links physical water loss with the erasure of local governance and knowledge.
  • Water conflicts around data centers are increasingly framed as colonial-capitalist struggles over resources and authority.

The Bottom Line

For builders and infra engineers, this is a wake-up call: you cannot optimize for compute efficiency while ignoring the hydrological and social costs of your stack. The era of hiding behind the 'cloud' metaphor is ending, and transparency regarding water usage will likely become a critical compliance and ethical requirement for AI infrastructure.