The
AI systems people interact with directly, the chatbot, the recommendation
engine, the image generator, represent the visible surface of an
infrastructure that is orders of magnitude larger, more expensive, and more
consequential than what appears on a screen. Behind every AI interaction lies
a chain of physical dependencies: server hardware requiring continuous power
and cooling, fibre optic networks carrying data between facilities, logistics
systems supplying components and personnel, and energy infrastructure scaled
to demands that have grown faster than the planning systems governing it. The
IEA’s
Energy and AI report documented that global investment in data
centres nearly doubled between 2022 and 2024, reaching half a trillion
dollars. Electricity demand from data centres rose 17 per cent in 2025. The
IEA projects consumption will more than double to around 945 terawatt-hours
by 2030, equivalent to Japan’s current total electricity demand. This
infrastructure is not metaphorical. It is also, by design and by default,
largely invisible to the people whose daily lives it increasingly
shapes.

The invisibility is not accidental. Data centre planning
applications are frequently submitted under generic commercial or logistics
categories that do not indicate the scale of energy and water demand the
facilities will place on shared infrastructure. Communities in northern
Virginia, rural Ireland, and parts of the Netherlands have found themselves
hosting hyperscale AI infrastructure without having had planning processes
that adequately informed them of the cumulative resource implications. The
Oxford
Political Review’s analysis of AI data infrastructure as an urban planning
challenge found that the regulatory frameworks governing
conventional industrial development are systematically unprepared for the
pace and technical complexity of AI infrastructure expansion. The result is
that consequential decisions about AI’s physical footprint are being made in
planning offices operating with inadequate information and insufficient
jurisdiction.

The Geography of AI Infrastructure

Data centre location is determined by the intersection of cheap power,
tax incentives, land availability, and network connectivity. The regions
satisfying these criteria are not distributed evenly, and neither are the
consequences. Northern Virginia hosts the largest concentration of data
centre capacity in the world, to the extent that the region’s electricity
grid is under sustained pressure from the cumulative demand of hundreds of
facilities clustered in a relatively small area. Ireland, the Netherlands,
and Singapore have each attracted large data centre clusters that now
represent significant fractions of their national electricity consumption,
generating policy responses ranging from grid connection moratoriums to
tighter water use requirements for cooling systems.

The World Economic Forum has highlighted how AI-embedded urban
infrastructure, including the traffic management systems, environmental
sensors, and public safety networks visible in city environments, operates on
a backbone of cloud services and data connections whose governance is
multiple organisational layers removed from local accountability. Smart city
infrastructure that appears to be a local public service is frequently
dependent on systems operated by private companies under contracts whose
terms are not disclosed to the communities whose data those systems process.
The gap between the visible interface and the invisible infrastructure is not
only physical. It is also a governance gap, separating the accountability for
outcomes from the control over the systems producing them.

Water: The Underdisclosed Cost

Carbon emissions from AI data centres have received sustained
regulatory attention. Water consumption has not, despite being in many
locations a more immediately significant local impact. Evaporative cooling
systems used in large facilities consume water rather than returning it to
the source, releasing it as vapour in the cooling process. A UK Government
report on water use in data centres and AI projected that AI-driven demand
growth could push global facility water consumption to between 4.2 and 6.6
billion cubic metres annually by 2027, roughly half of the UK’s total annual
water consumption. In communities near large facilities drawing from shared
aquifers or river systems, this consumption creates direct competition with
agricultural and domestic water use that is rarely disclosed in planning
applications and has no standardised reporting framework comparable to the
energy disclosures now required in major jurisdictions.

The trajectory from unmeasured externality to regulatory concern
is being repeated for water, on a lag of several years behind the carbon
debate. Mandatory annual water consumption disclosures for large facilities
would create the information base needed for evidence-based planning
decisions. Without that information, local authorities are approving data
centre development without understanding the full resource implications of
what they are authorising. As our analysis of how
AI data centre demand is affecting local communities and energy
grids found, the gap between where AI value is captured and where
its physical costs fall is both real and poorly governed at every level from
local planning to international standard-setting.

Making the Invisible Governable

The governance challenge posed by AI’s physical infrastructure is
not primarily technical. The information needed to govern it, where
facilities are located, how much energy and water they consume, how they
connect to public systems, and what their cumulative effect on shared
resources is, could be collected and disclosed under existing regulatory
frameworks if those frameworks were applied with the same rigour they apply
to other heavy infrastructure users. The obstacle is political and
institutional: the technology sector has successfully maintained regulatory
treatment that does not reflect the scale of its physical footprint, and the
planning systems that apply to comparable industrial users have been slow to
close that gap.

The IEA’s finding that AI data centre capital expenditure is set
to increase by a further 75 per cent in 2026 indicates that the pace of
infrastructure expansion will intensify before governance frameworks catch up
with it. For the communities in the path of that expansion, and for the
energy and water systems serving those communities, the gap between the speed
of deployment and the speed of governance is not an abstract policy problem.
It is the condition in which the most consequential decisions about AI
infrastructure are currently being made, and the decisions that are made now
will shape the physical landscape of AI for the rest of the
decade.

The Democratic Deficit

The pace at which AI infrastructure decisions are being made
outstrips the capacity of democratic processes to shape them. Planning
applications for hyperscale facilities are approved in weeks; the
environmental and community effects of those facilities unfold over decades.
Environmental impact assessments designed for conventional industrial
development do not capture the cumulative grid, water, and community effects
of multiple AI facilities in the same region. The information asymmetry
between the companies building AI infrastructure and the communities and
regulatory bodies meant to govern it is substantial and growing, because the
technology and the associated resource demands are evolving faster than
assessment frameworks can be updated.

Addressing this democratic deficit requires both better
information frameworks and more deliberate community engagement in decisions
about AI infrastructure. Mandatory pre-application consultation requirements,
standardised resource impact disclosures, and independent technical
assessment of cumulative effects across multiple facilities in the same
region are achievable policy interventions that would substantially improve
the quality of planning decisions. As our analysis of AI
governance gaps in law enforcement found, the distance between
where AI decisions are made and where accountability for those decisions
falls is a structural feature of current deployments across multiple sectors.
In infrastructure planning, closing that distance requires regulatory
frameworks that treat AI infrastructure as a category requiring its own
assessment standards, rather than fitting it into frameworks designed for
warehouses and logistics parks. The investment in those frameworks is
considerably smaller than the cost of the governance failures that will
accumulate without them. Our analysis of how
AI infrastructure exclusion compounds across populations shows why
governance failures in one area consistently create pressure in
others.

About the Author

Stuart Kerr is the Technology Correspondent for LiveAIWire. He
writes about artificial intelligence, emerging technology, and the forces
reshaping work, business, and society.