A platform works for millions of people, a data centre meets its annual average, and a phone receives updates for the advertised period. Then a town faces water restrictions, a battery cannot be replaced, or a family loses access to school materials when an application changes. Such cases are often called exceptions, as though they existed outside the system. Yet they expose its rules with unusual clarity: which resources it consumes, which dependencies it creates, who absorbs the cost, and who retains the capacity to repair. The exception does not break the rule; it explains it. Applied to the cloud that also weighs on the Earth, this idea makes it possible to assess technology by its complete behaviour, not merely by the convenience of its interface.
Digital life appears immaterial because its operations arrive as icons, instant messages, and files that occupy no visible space in a room. Behind them stand mines, factories, cables, aerials, data centres, cooling systems, electricity grids, packaging, transport, and waste. The International Energy Agency places data-centre electricity consumption at about 1.5 per cent of the global total in 2024 and warns that local impacts are far more concentrated. That figure does not make every digital use abusive, but it prevents the cloud from being described as a place without land or a material bill.
An exception is information about design#
An exception gains analytical value when it is no longer treated as an inconvenient anecdote. If one device fails early, it may have an individual defect; if many devices fail in the same way under heat, dust, or intensive use, the supposedly rare case reveals the narrow conditions for which the product was designed. The same applies to a digital service that works well only where bandwidth, recent devices, and technical support are plentiful. Its average success can coexist with a systematic pattern of exclusion and additional consumption.
An exception gains analytical value when it is no longer treated as an inconvenient anecdote.
The professional question is not whether a technology is good or bad in the abstract. It is what the purpose, unit of analysis, and consequence are. A service may reduce travel while increasing server use, device replacement, or dependence on one supplier. Responsible measurement records both movements and distinguishes global impacts from territorial ones. Efficiency per operation does not guarantee an absolute reduction either: if the cost falls while the number of operations grows faster, total consumption may rise. The exception helps locate where that rebound occurs.
It also reveals the distribution of power. A person who can buy another device experiences failure as an inconvenience; someone who depends on a single terminal may experience it as a loss of education, work, or access to government. A large company can negotiate energy contracts and redundancy; a small municipality may face pressure on its grid or water supply without taking part in the decision. The same technical failure produces different consequences according to the resources available. Equity is therefore not a later moral addition but a necessary dimension of any material assessment.
The cloud has a life cycle, a territory, and a timetable#
Analysis begins before a device is switched on. Mineral extraction, semiconductor and battery manufacturing, assembly, and transport concentrate impacts that never appear on the user's electricity bill. The use phase, maintenance, updates, and end of life then follow. UNITAR and the International Telecommunication Union calculated that the world generated 62 million tonnes of electronic waste in 2022 and that this waste was growing far faster than documented recycling. The figure is a reminder that replacing an object is not the same as making it disappear.
Territory matters because a kilowatt-hour, a litre of water, or a tonne of waste does not have the same effect everywhere. Electricity may come from different mixes; cooling may compete with other uses in a stressed watershed; a recycling plant may operate under robust controls or transfer occupational and environmental risks to less visible communities. An aggregate balance can be correct while concealing significant local harm. Assessment needs maps, time bands, material origins, and final destinations in addition to an annual sum.
Timing also changes the result. An efficiency improvement may save energy for years if it extends a system's life, but it may create a manufacturing spike if thousands of terminals must be replaced simultaneously. A temporary solution may become permanent infrastructure without review. A repair delivered after six weeks may be recorded as resolved even though a person has lost a course, income, or an administrative deadline. Measuring duration, reversibility, and the moment of consequence keeps the average from turning lived experience into a footnote.
Case 1: an efficient data centre in a stressed watershed#
A technology company announces a data centre with efficient servers and contracted renewable electricity. Corporate indicators show less energy per computation than at previous facilities. During a dry summer, the municipality restricts certain water uses and residents ask how much cooling consumes. The company responds with an annual average. The case is presented as an exceptional coincidence, but that coincidence exposes a shortcoming: the average does not explain demand on critical days or which resources are competing within the same territory.
A complete assessment would separate water withdrawn, consumed, and returned; distinguish drinking, reclaimed, and rainwater sources; publish seasonal data; and describe behaviour during heatwaves. It would also compare evaporative cooling, closed systems, and other alternatives without making one metric the verdict. Saving water can increase electricity use, while saving electricity can shift pressure towards water. A mature decision makes that relationship visible, establishes thresholds, and determines who has authority to reduce load when the territory enters a critical situation.
Institutional repair is not limited to financial compensation after the event. It includes local participation before capacity is expanded, understandable information, independent auditing, and a public protocol for periods of scarcity. If the infrastructure provides jobs, connectivity, or recoverable heat, those benefits should also be measured and distributed according to clear criteria. The case does not establish that no data centre can operate in a vulnerable watershed. It establishes that a foreseeable climatic exception belongs in design, contracts, and permits instead of being classified as a surprise.
Case 2: the phone that works except for one inaccessible part#
A three-year-old phone retains sufficient processing power, an intact screen, and software support, but its battery loses capacity. The manufacturer offers replacement at a price close to that of a refurbished handset and restricts access to parts, manuals, or tools. For most customers with resources, the quick solution is to purchase another device. The person who needs to keep it appears to represent a special case. In fact, that case exposes an economic rule: design and after-sales service have turned a consumable component into a decision about the product's entire life.
The European Directive on the repair of goods seeks to extend use through repair obligations for products covered by reparability requirements and measures that make offers easier to compare. Its principle matters beyond one regulation: a repairable product needs safe opening, available spares, documentation, diagnostics, reasonable prices, and updates that do not disable the intervention. Safety and intellectual property can justify specific limits, but they should not become general excuses for preventing all independent repair or erasing the reuse market.
The most informative measurement compares years of service, return-to-use rates after repair, total cost, avoided embodied emissions, and the destination of the removed battery. It also observes who can access service and how long each person remains without a device. A network of local repair shops can reduce travel and develop knowledge, provided it has training and safe conditions. When a repair fails, the intervention needs a warranty and the part needs traceability. The right to repair then ceases to be a slogan and becomes everyday infrastructure for trust.
Case 3: a school migrates to a lighter application#
A school replaces its learning platform with an application that promises lower data consumption and better performance on older equipment. Migration appears beneficial: it avoids replacing tablets and reduces traffic. However, three families cannot open downloaded materials because the new format depends on an active account; a student using a screen reader encounters unlabelled buttons; and the archive of previous courses loses some links. These exceptions are numerically small, but they explain hidden rules about ownership, accessibility, and preservation.
The school responds responsibly when it retains an equivalent route, exports files in open formats, tests accessibility with real people, and documents future withdrawal. It must also verify whether the improvement on the device shifts work or consumption to external servers. The footprint cannot be reduced by comparing transferred megabytes alone. Device life, duplicated storage, backups, support, training, and the capacity to change provider all matter. A lightweight tool that locks the archive can create deferred material and cultural debt.
Repair in this case includes restoring access, correcting labels, rebuilding links, and making up for lost time without blaming families. A monitoring committee can record how many people abandon each step, not only how many complete the task. Continuation should depend on published criteria: verifiable savings, accessibility, preservation, security, and satisfaction separated by group. If the service fails, the contract needs an operational exit and a complete export. The exception thereby becomes a quality test rather than a reason to discredit the person who reports it.
A method for measuring without hiding anyone#
The first step is to frame a narrow, verifiable question. Instead of asking whether the cloud is sustainable, an organisation can ask how much energy, water, material, and equipment replacement one function requires over a specific period and in a defined territory. It then establishes the baseline: what happened before, which equivalent service existed, and which costs were shifted to users, suppliers, or government. Without a baseline, an efficient figure may look like a saving even when consumption has merely moved beyond the measured boundary.
The second step builds a life-cycle inventory proportionate to the decision. Not every initiative needs a complete academic study, but every initiative should state what it includes and excludes. Operational energy, manufacturing, transport, maintenance, water, parts, waste, and induced use form an initial list. The European Environment Agency stresses that digitalisation creates pressures across the life cycle, from extraction and production to use and disposal. This view prevents a clean screen or renewable contract from hiding less visible phases.
The third step deliberately seeks three kinds of exception: the day of highest demand, the person with the least margin, and the failure that forces use of an alternative. Each test needs a stopping threshold, an accountable owner, and a repair deadline. Qualitative accounts explain mechanisms that statistics may detect late; quantitative data indicate whether an account is isolated or repeated. Both are reviewed together, without turning testimony into decoration or a metric into absolute authority.
The fourth step publishes a short record covering purpose, scope, data sources, assumptions, territorial differences, risks, accountable owner, review date, and exit plan. It distinguishes measurement from estimation and identifies uncertainty. It also states the complaint channel and maximum response time. When scale, supplier, or climate context changes, the assessment reopens. An old authorisation is not an indefinite licence, because material conditions and available alternatives evolve.
Responsibility, repair, and public decisions#
Real responsibility requires authority. A sustainability officer cannot be limited to writing a report while lacking access to contracts, consumption data, or purchasing decisions. A technical team cannot repair when the supplier blocks diagnostics and parts. A community does not participate effectively when it receives information after infrastructure has been approved. Coherent governance gives someone the power to pause, investigate, correct, and communicate, and it protects those who raise a reasonable problem.
Repair takes different forms: restoring access, replacing a component, reducing a load, recovering a file, funding a local improvement, or changing a clause. It must always respond to the specific harm and verify whether the measure worked. Compensation does not replace prevention when harm is irreversible, and an apology does not remove a dependency that continues. Incident records should inform later purchasing, design, and training. Each exception then leaves knowledge and reduces the chance that the same cost will recur under another name.
Transparency does not require the publication of trade secrets or personal data. It requires enough information to understand the purpose, resources involved, consequence, and review route. Indicators may be reported as ranges when an exact figure is sensitive, and an independent audit may check reserved details. The essential point is to prevent technical complexity from becoming immunity. An environmental claim needs evidence, a boundary, and a date; a repair promise needs parts, deadlines, and accountable people.
A rule made more complete by its exceptions#
The cloud that weighs on the Earth is not corrected through individual guilt or indiscriminate rejection of technology. It is governed through decisions that recognise life cycles, territory, time, and the unequal distribution of resources. The three cases display the same structure: a favourable average concealed a critical condition; an affected person made that condition visible; and an adequate response required changing the rule rather than silencing the case. That sequence turns an incident into institutional learning.
When understood rigorously, the exception also protects innovation. It demands more resilient services, durable products, reversible contracts, and measurements that survive a difficult question. An organisation that knows its limits can improve; one that celebrates only the average remains exposed to the next summer, failure, or migration. The final criterion is simple and demanding: a digital solution is responsible when its benefits do not depend on hiding matter, transferring harm, or abandoning whoever needs repair. The cloud then ceases to be an evasive metaphor and becomes infrastructure that can be explained, debated, and cared for.
