The cloud is commonly pictured as a clean, instant space with no location. Yet every synchronised file, video call, backup or automated answer passes through physical equipment, networks, buildings and territories. The invisible question that changes everything is not whether a technology works, but what it needs in order to work and who bears its costs. Data centres, antennas, cables, cooling systems, batteries, minerals, water, electricity and specialised labour sustain the digital experience. Making that infrastructure visible does not require rejecting it. It makes it possible to judge technology with complete criteria, compare alternatives and repair harm before convenience is mistaken for an absence of impact.
Professional responsibility begins by leaving two simplifications behind. The first claims that every digital activity is automatically cleaner than its material equivalent. The second blames individual users for choices that belong to design, procurement and energy policy. A serious assessment distributes responsibility according to the capacity to act. A service user can moderate unnecessary use; a designer decides how much data is collected; a buyer sets requirements; an operator chooses equipment, energy and cooling; and public institutions establish transparency, environmental protection and the right to repair.
The cloud has an address, materials and a timetable#
A digital service occupies land and needs reliable connections. Its impact depends on place, time and the specific technology involved. The same task may rely on an electricity grid with a different carbon intensity at different hours; a cooling system may use more water in a basin under stress; and an expansion may compete with other local uses of energy or land. A single global figure is therefore rarely sufficient. An assessment should state scope, period, location, source and assumptions, and it should separate measured data from estimates.
A digital service occupies land and needs reliable connections. Its impact depends on place, time and the specific technology involved.
It should also examine the life cycle. The use phase consumes electricity, but manufacturing servers, chips, batteries and network equipment requires extraction, processing, transport and construction. Replacing hardware too early shifts environmental burdens to places that do not appear on the customer's dashboard. Extending useful life, repairing modules, reusing components and commissioning verifiable end-of-life management can reduce that pressure. Operational efficiency matters, although it does not automatically compensate for unlimited growth in capacity, storage or traffic.
Timing adds a dimension that annual labels conceal. An organisation may claim renewable electricity through contracts while demanding power when the grid relies on more intensive sources. Responsible management examines temporal matching, additional capacity, backup and local constraints. It does not use a certificate as permission to stop measuring. Nor does it present a future promise as a current result. The language should enable a community to distinguish direct reduction, burden shifting, compensation and mere expectation.
Measuring without turning an indicator into an excuse#
Total consumption provides an essential signal, but it needs context. Energy per task, equipment utilisation, useful life, embodied footprint, water use, waste generation and the proportion repaired or reused describe different dimensions. No metric can summarise a fair decision on its own. An efficiency improvement may coincide with an increase in absolute consumption if a service multiplies features, users or copies. That rebound effect does not invalidate efficiency; it requires limits and priorities alongside it.
The functional unit should correspond to a recognisable need. Comparing an in-person meeting with a video call requires considering avoided travel, participant numbers, devices already in use and necessary quality. Comparing storage requires distinguishing an essential copy from indefinite retention for uncertain purposes. Measurement loses rigour when it divides impact by a vast number of irrelevant operations to produce a tiny figure. It regains rigour when it connects resources to outcomes, identifies uncertainties and shows who receives the benefit.
Useful transparency is not an incomprehensible dashboard. It should explain which facilities enter the calculation, which periods are missing, how information is verified and which decision would change after an adverse result. A datum with no owner or threshold can decorate an annual report without altering any practice. A verifiable commitment instead establishes a baseline, a date, a stable method and an authority capable of stopping an expansion when agreed limits are exceeded.
Case 1: a school accumulates copies without knowing it#
In a hypothetical case, a school network adopts a platform for assignments, videos and backups. Each academic year creates new spaces, while previous ones are retained in case material becomes useful again. The provider replicates files across several regions and enables historical versions by default. The invoice appears affordable and environmental impact remains outside the conversation. Five years later, nobody knows what must be kept as a legal obligation, what has educational value or what contains unnecessary personal data.
The responsible response is not a mass deletion order. The network classifies information with teachers, archivists, data protection staff and students; defines periods according to purpose; preserves important work in portable formats; removes verified duplicates; and requires review before adding capacity. The provider explains regions, redundancy and storage choices. The school measures total volume, growth and successful recovery, not merely space released. Repair includes explaining the changes and providing a way to recover material deleted in error for a prudent period.
The case shows that digital sufficiency is a governance practice, not an exercise in guilt. A lesson video may have considerable value; an abandoned copy has none. The institution distinguishes the two, assigns responsibility and prevents indefinite retention from becoming the default decision. It also purchases repairable devices and aligns replacement with real needs, so that the platform does not force new equipment solely through cosmetic changes or disproportionate requirements.
Case 2: a municipality negotiates water, energy and trust#
In a second hypothetical case, a company proposes a data centre near a city seeking jobs and investment. The announcement highlights efficiency and renewable power, but does not detail water consumption on hot days, reserved grid capacity, backup generators, noise, land use or stable employment after construction. The annual average appears favourable while critical moments remain hidden. Public discussion becomes polarised between unconditional acceptance and rejection without sufficient information.
The municipality requests an independent assessment and scenario-based data. It examines the water basin, electricity grid, emergency plans, recoverable heat, substances used and equipment end of life. It publishes assumptions and legitimate commercial limits, organises accessible sessions and listens to water services, residents, workers, environmental specialists and energy authorities. Permission links expansion to measurable thresholds, audits and incident reporting. If the facility cannot comply during drought or grid stress, an authority can reduce activity.
Repair is designed before harm occurs. A fund or guarantee covers restoration, pollution or abandonment; contracts clarify responsibility; and the community knows how to make a claim. Benefits are also made concrete: training with labour rights, local procurement where viable and use of waste heat only where nearby, stable demand exists. No advantage permits hidden costs. No cost alone proves the project unviable. The decision gains legitimacy because its conditions can be checked and corrected.
Case 3: artificial intelligence with a material budget#
In a third hypothetical case, a cultural organisation introduces artificial intelligence tools to translate catalogues, describe images and improve search. The team observes better access, but each department begins using larger models for simple tasks. Multiple variants are generated but never published, and every request is retained. The provider reports relative efficiency without disclosing enough information to compare options. The project risks calling every increase in computation innovation.
The organisation first defines its public outcome: accessibility, discovery of collections and support for editorial work. It tests models of different sizes, local processing where appropriate and human procedures for sensitive tasks. It records volume, quality, time, estimated energy, errors and review required. Responsible staff can select the least intensive option that meets the standard, turn features off and delete logs without losing the catalogue. Procurement requires comparable information, portability and notice of material service changes.
The material budget is not an arbitrary quota. It connects resources to priorities and requires exceptions to be explained. An accessible description may justify additional work; one hundred unused automated drafts may not. The organisation publishes limitations and avoids figures with false precision when the provider offers insufficient data. It also reviews labour conditions, rights and cultural quality, because a favourable environmental balance cannot make fair a practice that displaces tasks without recognition or reproduces exclusion.
Repair means preserving capability#
Technological repair begins long before equipment breaks. It requires documentation, parts, modular design, reasonable updates, interoperable copies and staff with time to maintain systems. A service that can continue only through the original provider creates dependency even if it operates efficiently. Contracts should therefore cover complete export, migration assistance, verifiable deletion, failure recovery and responsibilities at the end of the relationship. A practised exit is an environmental measure: it prevents rushed replacements and the loss of useful work.
Social and territorial repair also matters. If infrastructure causes disturbance, consumes a scarce resource or breaches a condition, the response does not end with a technical improvement. It must recognise affected people, investigate causes, communicate timescales and restore what can be restored. Financial compensation may be part of the response, but it cannot replace prevention or purchase silence. A grievance mechanism needs independence, accessibility and genuine authority.
Circularity requires traceability. A statement that equipment is recycled does not explain how much is reused, which materials are recovered or where the remainder goes. An organisation keeps inventories, prioritises repair and refurbishment, selects authorised operators and requests evidence of treatment. Information security is handled through certified erasure or proportionate destruction, not the automatic disposal of devices that remain useful. Protecting data and extending material life can reinforce each other.
Governance for decisions that can be stopped#
Competent governance brings together technical, environmental, legal, financial, labour and community knowledge. It does not leave sustainability to a peripheral team that reviews communications. Before procurement, it asks which problem is being solved, which alternative requires fewer resources, which groups receive benefits and which harm would require a stop. During use, it reviews changes of scale, versions, incidents and unforeseen effects. At the end, it records lessons and executes the exit.
Management retains an accountable person inside the organisation even when infrastructure is external. A supplier provides evidence; it does not replace the customer's judgement or public oversight. Product and environmental targets are examined together so that one department does not save resources while another multiplies demand. Operational teams receive time and authority to identify failures. Staff should not have to choose between meeting a commercial target and reporting an environmental risk.
Early participation improves information quality. A local community understands constraints involving water, transport and noise; maintenance staff recognise weaknesses omitted by a sales presentation; and users explain which functions create genuine value. Participation does not mean that every preference becomes a veto. It means decisions answer plural evidence, explain disagreements and provide review routes before difficult-to-reverse costs become fixed.
Four institutional sources to guide practice#
The International Energy Agency examines the relationship among artificial intelligence, data centres and electricity systems, helping to place digital demand within energy planning. The International Telecommunication Union, through the Global E-waste Monitor 2024 produced with institutional partners, provides a framework for understanding electronic waste generation, collection and treatment. Together, these sources show why digital infrastructure must be observed through both electricity and materials, rather than through a screen alone.
The European Commission describes reporting obligations and indicators concerning the energy performance and water footprint of certain data centres. The Commission's Joint Research Centre maintains best practices for energy efficiency in these facilities. These references do not provide an automatic moral label or replace local regulation. They support questions, metrics and scrutiny of claims. Each organisation still needs to document why it adopts a criterion and how it will respond if results contradict its promise.
The question that turns convenience into responsibility#
The cloud can provide collaboration, access, continuity and knowledge. Its material nature does not cancel those benefits; it prevents them from being treated as free. The decisive question accompanies every choice: which resources it needs, for how long, in which territory, under which labour conditions and with what prospect of repair. When the answer is uncertain, an institution does not invent precision. It declares the gap, limits scale and requests the necessary information.
A mature policy combines sufficiency, efficiency, cleaner energy, durable equipment and shared responsibility. It reduces what brings no value, optimises what is needed, avoids shifting impacts and maintains an exit. It measures absolute outcomes alongside intensities and listens to people who live near infrastructure or work within it. Above all, it turns transparency into the capacity to decide. The cloud weighs on Earth, but that weight is not predetermined: it depends on choices that can be made visible, debated and repaired.
