In the popular imagination, artificial intelligence (AI) exists almost untethered from physical constraints: a disembodied entity summoned from some abstract digital ether. Yet this perception overlooks the crucial reality that AI’s operations do have a significant material dimension — and one with the potential to profoundly shape local geographies, economies and modes of social organisation. Each predictive text suggestion, each voice command interpretation, each facial recognition match relies on a complex physical substrate.
AI’s materiality can be considered from two angles: firstly, its consumption of physical resources and secondly, the infrastructure which sustains it. The former ‘resource-based’ aspect encompasses the raw physical inputs consumed in AI development and deployment, such as silicon and germanium for semiconductor fabrication, rare earth minerals for other computing components, and water for cooling processors and servers. The second ‘infrastructural’ facet includes the constructed physical environments that house and enable AI operations, such as data centres and AI factories. On the rare occasions when AI’s materiality has entered public dialogue, discussion has typically focused narrowly on resource constraints while neglecting the infrastructural dimension. Addressing this blind spot is crucial if we wish to understand the challenges and opportunities which AI poses to society – in particular, how AI’s material infrastructure entwines with urban planning, and so shapes life within the built environment. As AI infrastructures take root, they redraw boundaries between industrial and civic space, concentrate power and resources in new geographies, and inscribe algorithmic priorities into the very fabric of urban existence.
The vast network of physical infrastructure that AI requires cannot be deployed just anywhere. The geographic specificity of AI systems – their need for particular site conditions – results in a highly uneven spatial distribution that concentrates development in certain cities and regions while bypassing others entirely. AI infrastructure gravitates toward locations with abundant cheap electricity, extensive water resources for cooling, robust fibre connectivity, and access to specialised labour pools. Google’s decision to build major AI research facilities in Montreal stems not only from the city’s academic expertise in machine learning but also from Quebec’s surplus hydroelectric power, which provides both renewable energy and competitive pricing. Similarly, Microsoft’s selection of counties along Washington’s Columbia River for data centre sites leverages the region’s hydroelectric resources. In northern Virginia’s Loudoun County, over 70% of global internet traffic now flows through massive data centres that have transformed former farmland into a ‘Data Center Alley’, thanks to reasonably priced available land and low energy costs. The geographical concentration of this infrastructure creates what geographer Orit Halpern calls ‘intelligence corridors’ – regions where the accumulation of AI development facilities reshapes local landscapes according to the industry’s material requirements.
These intelligence corridors are not passive occupants of space but active agents of spatial transformation. As specialised territorial formations with their own operational logics and material requirements, these facilities operate as what might be called spatial products that fundamentally alter their surroundings in ways that traditional planning frameworks and governance structures are often ill-equipped to comprehend or regulate. What presents itself as a straightforward engineering exercise thus conceals profound spatial politics. The spatial politics at play extend far beyond visible interventions like security perimeters, massive water pipelines or widened highways, manifesting instead through fundamental reconfigurations of economic, ecological and political relations. The entire lifecycle of AI infrastructure projects – from site selection to construction, operation, and eventual decommissioning – represents not merely a technical decision but a profound intervention in urban and regional development with cascading effects that ripple through every aspect of community life.
Wealth Creation vs Inequality Tensions
The economic promises of AI’s material infrastructure manifest most visibly through direct regional investments that generate significant benefits for host communities. When established in areas seeking economic revitalisation, these facilities create immediate construction employment. Meta’s Eagle Mountain facility in Utah exemplifies this transformative potential, injecting over $1 billion into local economies through contracts with regional suppliers. Beyond direct employment, these developments produce significant fiscal windfalls through property taxation, reshaping municipal capacities in profound ways. DeKalb County, Illinois experienced this transformation firsthand when the opening of a Meta data centre in its boundaries increased its property tax base by over $1 million, enabling expanded public services without increased tax rates on existing residents. Equally significant is the transitional opportunities these facilities offer to post-industrial regions: the successful pivot from timber dependency to digital infrastructure, achieved by the city of Prineville in Oregon, demonstrates how communities can leverage AI infrastructure to navigate economic transitions. Municipal ownership models further amplify these benefits; Chattanooga’s municipal publicly-owned fibre network attracts AI development while ensuring returns flow to residents rather than external shareholders, presenting a model where infrastructure projects for technological advancement and community wealth generation operate in tandem.
Yet this economic transformation carries profound distributional concerns, threatening to exacerbate existing inequalities. The geographic concentration of AI infrastructure reinforces rather than disrupts established patterns of regional advantage, with already prosperous regions capturing disproportionate benefits while economically disadvantaged communities remain largely excluded from this new frontier of opportunity. Even within regions hosting major facilities, economic benefits distribute highly unevenly. The bifurcated nature of new employment generated by AI infrastructure – characterised by a small cadre of highly-compensated technical specialists alongside larger numbers of modestly-paid service workers – reproduces rather than remedies wage inequality. Housing markets reveal this stratification most clearly; in light of the massive Stargate data centre’s recent opening by OpenAI in Abilene, Texas, surrounding neighbourhoods expect to experience significant housing cost increases. Most concerning is how private ownership structures enable value extraction; regions providing substantial tax incentives for infrastructure development often discover that promised economic multipliers fail to materialise when profits flow to distant corporate headquarters rather than circulating locally, leaving communities with minimal returns.
Sustainable Solutions vs Resource Extraction
On the environmental front, AI’s material presence has produced unexpected ecological benefits that extend beyond the facilities themselves. As these massive structures insert themselves into local landscapes, they have paradoxically stimulated innovations in resource stewardship that benefit entire regions. The waste heat generated by computational processes – once considered a nuisance to be mitigated – has been transformed into a community asset in forward-thinking municipalities. Meta’s Odense data centre exemplifies this transformation; by redirecting thermal discharge into district heating networks serving nearly 7,000 Danish homes, the facility converts computational byproducts into residential warmth, reducing community carbon emissions by approximately 25% while diminishing reliance on dedicated heating plants. Similar community-level innovations appear in water systems; Google’s data centre in Douglas County, Georgia pioneered wastewater recycling technologies that now treat and reuse municipal effluent for cooling systems, reducing pressure on drinking water supplies while demonstrating circular water economy principles that neighbouring industries have begun adopting. The substantial electrical demands of these facilities have also accelerated regional grid transformation; when Microsoft committed to 100% renewable energy for its Irish data centres, the company catalysed the development of wind farms that now supply clean electricity to surrounding communities as well. In addition, there are landmark ecological restoration projects emerging alongside new facilities; Amazon’s commitment to water replenishment around its data centres has funded watershed restoration in Virginia’s Potomac River basin, returning groundwater supplies to levels not seen in decades while improving habitat for native species. These community-level environmental benefits represent not merely greenwashing corporate strategies but genuine improvements to local ecological systems that may prove more durable than the computational facilities that precipitated them.
Nonetheless, the ecological disruption triggered by AI infrastructure creates environmental challenges that reverberate through host communities in complex and often unanticipated ways. The concentrated water demands of these facilities – often exceeding millions of litres daily – are not always properly addressed and often end up stressing local hydrological systems beyond their sustainable capacity. In Oregon’s Dalles River watershed, Google’s data centre operations contributed to dropping aquifer levels, affecting agricultural users whose families had farmed the region for generations and triggering unprecedented water rationing measures during summer months. Similar conflicts have emerged around power infrastructure; despite commitments to renewable energy, the immediate electrical demands of AI facilities often exceed existing clean energy capacity, necessitating continued operation of fossil fuel plants that degrade local air quality and affect community health outcomes. The transmission infrastructure required to deliver this electricity creates additional landscape disruptions – new high-voltage corridors cutting through previously undeveloped areas, fragmenting habitats and altering viewsheds that once defined community identity. Thermal pollution presents subtler but equally significant concerns as the discharge of cooling water from data centre districts tends to measurably increase water temperatures, affecting marine ecosystems upon which local fishing communities depend. Perhaps most concerning are the emergent effects that accumulate over time; the combined heat island effects of multiple data centres in northern Virginia have measurably altered local microclimate conditions, increasing summer temperatures and changing precipitation patterns in ways that local infrastructure was never designed to accommodate.
Democratic Possibilities vs Institutional Erosion
The infrastructural presence of AI also opens up a politically complex terrain. By manifesting in specific locations rather than remaining abstracted in digital space, AI infrastructure establishes concrete sites where communities can engage with and shape technological development. Rather than remaining hidden in algorithms beyond public scrutiny, technology corporations acquire a concrete presence through material infrastructure; these become visible, physical points of intervention where citizens can assert democratic claims through regulatory processes, public consultations, and direct action. This material visibility does not in itself guarantee democratic oversight – companies may still evade scrutiny – yet under certain conditions, footholds for contestation are created. For example, planning processes and procurement regulations tied to infrastructure can become sites where citizens, civil society, and local governments assert democratic claims. Barcelona’s Digital City initiative exemplifies this possibility for democratic engagement: its public procurement requirements and its digital platform for citizen consultation made an attempt at shifting the development of AI infrastructure from closed corporate processes to open, participatory ones where citizens help determine how they operate in shared urban spaces. Unlike broad public consultations that dilute attention across numerous urban concerns, Barcelona’s process was thematically anchored to the infrastructural conditions of digitalisation, concentrating civic attention and accountability expectations. Similar civic innovations emerged in Amsterdam, where the Tada initiative establishes democratic principles for digital infrastructure governance that prioritise transparency, inclusion, and human dignity. Perhaps most promising is how these material sites enable diverse stakeholders to participate in technological governance; from environmental advocates concerned with resource consumption to housing advocates addressing displacement, AI’s physical manifestation creates entry points for democratic engagement that purely abstract technologies cannot provide.
While these civic experiments demonstrate how AI’s materiality creates opportunities for democratic deliberation about technological futures, governance challenges persist, often overshadowing the aforementioned participatory initiatives. When facilities become dominant economic actors in their host communities, their operational requirements often reshape governance priorities and resource allocations in ways that bypass normal democratic processes. For instance, after Google established its primary data centre in The Dalles, Oregon, the company negotiated water rights that fundamentally altered community resource allocation without transparent public deliberation, establishing concerning precedents for corporate influence over essential public resources. Most concerning are emerging governance models that formalise corporate authority; Nevada’s Innovation Zones legislation enables technology companies developing advanced computing infrastructure to establish autonomous governmental entities with powers traditionally reserved for elected county governments, privatising governance in unprecedented ways. These arrangements establish concerning precedents where technological development supersedes democratic governance rather than operating within it. The classified nature of many AI systems further complicates democratic oversight; when facilities develop military applications or security-related technologies, normal processes of public review and environmental assessment are bypassed through national security exemptions.
AI Infrastructure in Our (Urban) Backyard: Between Promise and Precarity
The materialisation of AI through vast infrastructural networks transcends conventional urban planning paradigms, operating simultaneously as industrial facilities, public utilities, and civic institutions. These computational hubs, with their distinctive architectural requirements and resource demands, have emerged as powerful spatial actors that simultaneously shape and are shaped by their host communities. Their mixed implications – creating economic opportunity while exacerbating inequalities, pioneering environmental innovations while straining local resources, bolstering democratic participation while also attempting to bypass it – defies simplistic narratives. Rather than being categorically beneficial or detrimental, AI infrastructures unfold through layered and often conflicting entanglements with local economies, ecologies, and governance structures, making them sites of both promise and precarity.
As AI development evolves at breakneck speed, urban planners need to design infrastructural landscapes serving the needs of today while anticipating the breakthroughs of tomorrow. The future infrastructural footprint of AI hangs in uncertain balance: it may expand with growing model complexity, or contract through algorithmic optimisation, forcing cities to hedge against opposing spatial trajectories. The indeterminacy of AI’s futures also forces the question of legacy as AI infrastructure needs to be rethought not as fixed assets – that turn into relics resistant to repurposing due to their highly specialised configurations – but as transitional spaces in the urban fabric, capable of evolving alongside the very technologies they initially house.
What emerges from this analysis is a recognition that AI’s infrastructure requires governance frameworks as sophisticated as the technology itself. Cities and regions that develop regulatory approaches acknowledging the unique features of these facilities will harness their transformative potential while ensuring community interests remain paramount. AI will only ever be as good as the infrastructure that supports it. We must reimagine these infrastructure projects neither as corporate colonial territories nor as temples to technological salvation, but as pragmatic civic assets where algorithms and communities coexist under thoughtful stewardship that balances innovation with equitable, sustainable development.