Technology · Cloud and Sustainability
The Cloud Is Not Made of Clouds
The digital world feels almost weightless, but every cloud service, AI workload, and effortless act of creation depends on physical infrastructure and finite ecological resources. Perhaps progress also requires knowing when not to build.
There was a time when creating something required us to think about the resources needed to create it. If we wanted to take a photograph, we needed a camera, film or storage, and some understanding of the process; if we wanted to build software, we needed to understand the machine on which it would run, the memory it would consume, and the infrastructure required to keep it operational. Today, however, we have entered an era in which the distance between imagining something and producing it has become extraordinarily small, largely because the internet, cloud computing, and artificial intelligence have made computational resources so accessible that we rarely stop to consider what happens beyond the interface through which we interact with them.
We can generate hundreds of images without owning a camera, produce videos without understanding video production, create applications without necessarily understanding their architecture, and store practically everything we have ever produced somewhere in the cloud, often without knowing where that storage physically exists. What makes this transformation particularly interesting is that, although our ability to create and consume digital products has expanded almost without friction, the physical resources required to support that expansion have not become unlimited; we have simply become increasingly disconnected from the systems that supply them.
The cloud is perhaps the most successful metaphor in modern computing, because it makes an enormous industrial infrastructure sound almost weightless, but the cloud is not made of clouds. It is made of buildings, silicon, copper, electricity, water, cooling systems, networking equipment, and millions of physical components that must be manufactured, powered, maintained, and eventually replaced, which means that almost everything we consider digital has a material existence somewhere in the physical world.
The Disappearance of the Cost of Creation
One of the most interesting consequences of modern technology is that we have made the production of digital objects so inexpensive and convenient that we are gradually losing our instinct to question whether those objects need to exist in the first place. A person can generate dozens of nearly identical images, publish automatically generated videos, deploy another website, or create an application simply because an AI assistant can produce the initial implementation within minutes, and although this accessibility is an extraordinary development in human creativity, it also introduces a question that deserves considerably more attention: when the effort required to create something approaches zero, what happens to our understanding of its actual value?
The rise of vibe coding makes this particularly relevant to software engineering, because we are increasingly seeing a separation between the ability to produce software and the ability to understand the systems being produced. Someone can ask an AI model to generate an application containing a frontend, backend, database, authentication system, background workers, and deployment configuration without necessarily understanding why those components exist or what operational responsibilities they introduce. The application may work, the interface may look convincing, and the initial demonstration may be impressive, but once it is deployed, its containers, scheduled processes, database connections, storage requirements, network activity, and background services become part of a physical computing infrastructure that must continue to operate.
Of course, this is not an argument against AI-assisted development, because the same technology can help experienced engineers build more efficiently and enable people with limited programming experience to solve problems that were previously inaccessible to them. The concern is what happens when the ease of producing software is mistaken for evidence that the software is useful, particularly in an environment where creating another application is often easier than understanding, maintaining, or retiring the applications we have already built.
Perhaps the more important engineering question in the age of AI is no longer simply how quickly we can build something, but whether the problem we are solving justifies the resources required to keep our solution alive.
We Are Building an Intelligence That Requires an Industrial Infrastructure
Artificial intelligence has made this relationship between digital convenience and physical consumption considerably more visible, although most users still experience it as an almost instantaneous exchange between a question and an answer. Behind that apparently simple interaction, specialised processors perform enormous numbers of mathematical operations, memory systems continuously move data, networking infrastructure connects distributed machines, and cooling systems remove the heat generated by hardware operating under sustained computational workloads.
According to the International Energy Agency's Energy and AI report, data centres consumed approximately 415 terawatt-hours of electricity in 2024, equivalent to around 1.5% of global electricity consumption, while its baseline projection suggests that this demand could rise to approximately 945 terawatt-hours by 2030 as artificial intelligence and other digital services expand. These numbers should not be interpreted as evidence that AI alone is responsible for all data-centre electricity consumption, since cloud services, conventional computing, storage, and networking also contribute, but they demonstrate the scale of the infrastructure supporting our increasingly digital economy.
What interests me more than the numbers themselves is the physical process behind them: we are building increasingly powerful computing systems, placing more processors into increasingly dense environments, and then constructing additional infrastructure to manage the thermal consequences of operating those processors continuously. A data centre is therefore not merely a building containing computers; it is an interconnected industrial system involving electricity distribution, thermal engineering, water management, backup power, semiconductor supply chains, and network infrastructure, all of which must function together to deliver the apparently effortless digital experiences we have become accustomed to.
There is an interesting philosophical connection here with Martin Heidegger's 1954 essay, The Question Concerning Technology, in which he explored how modern technology encourages us to perceive the natural world primarily through its availability as a resource. His concept of standing-reserve describes a way of revealing the world in which things increasingly appear as supplies waiting to be ordered and used, rather than as parts of a larger existence with their own significance.
Although Heidegger was not writing about artificial intelligence or cloud computing, his observation becomes surprisingly relevant when we consider how casually we discuss the availability of electricity, water, computing power, storage, and minerals, as though the only meaningful question were how much additional capacity we could make available whenever demand increased.
The Water Behind the Intelligence
Electricity is only one part of the environmental footprint of modern computing, because the processors performing these calculations continuously generate heat, and removing that heat requires thermal management systems whose resource requirements vary significantly depending on the design and geographical location of the infrastructure. Some data centres use water-based cooling systems, while others rely on different combinations of air cooling, liquid cooling, and heat-exchange technologies; additionally, the electricity used to operate these facilities may itself come from power-generation systems that consume water during production.
A 2025 study published in Resources, Conservation and Recycling, titled The Water Use of Data Center Workloads: A Review and Assessment of Key Determinants, examined the factors that determine how much water is associated with individual computing workloads and found that the results can vary by more than 10,000 times under different infrastructure and operating conditions. This is an important finding because it tells us that assigning one universal water-consumption figure to an AI request would be misleading: the actual environmental impact depends on server efficiency, utilisation, cooling architecture, electricity generation, geographical climate, and whether the infrastructure is being used effectively.
The geographical dimension makes this particularly important, because a facility operating in a region with abundant water resources and a relatively efficient energy system does not necessarily create the same environmental pressure as an equivalent facility operating in a water-stressed region. Consequently, the expansion of digital infrastructure cannot be evaluated exclusively through global electricity consumption or corporate efficiency targets; we also need to understand what happens to the communities and ecosystems surrounding the physical locations where that infrastructure is built.
The Things We Create Eventually Become Waste
There is another side of the digital economy that rarely appears in conversations about AI productivity or cloud scalability: the physical devices required to operate these systems have finite lifetimes, and replacing them creates an expanding stream of electronic waste.
The processors in our computers, the memory modules in our servers, the networking equipment connecting data centres, and the specialised accelerators used for AI all depend on complex manufacturing processes involving minerals, energy, water, chemicals, and international supply chains. When these components become obsolete or economically impractical to maintain, they enter a disposal and recycling system that is already struggling to keep pace with the rate at which electronic products are being produced.
The Global E-waste Monitor 2024, published by the International Telecommunication Union and UNITAR, estimated that the world generated approximately 62 million tonnes of electronic waste in 2022, of which only 22.3% was documented as formally collected and recycled in an environmentally sound manner. The report also projects that this amount could increase to 82 million tonnes by 2030, which means that the physical consequences of our growing dependence on electronic technology are accumulating alongside the benefits we receive from it.
This is where Rachel Carson's Silent Spring, published in 1962, becomes an interesting reference point. Carson demonstrated how introducing a seemingly useful technological intervention into nature could produce consequences that extended far beyond the original purpose of that intervention, because ecological systems are interconnected in ways that are not always immediately visible to the people making decisions about them. Her famous observation that "In nature nothing exists alone" was made in the context of environmental relationships and the consequences of chemical pollution, but it also offers a useful way of thinking about the digital infrastructure we are constructing today: a processor does not exist independently of the minerals used to manufacture it, a data centre does not exist independently of its electricity and water supplies, and a cloud service does not exist independently of the physical systems that sustain it.
The digital world may appear disconnected from nature, but its entire existence depends on nature.
The Cost We Do Not See
Henry David Thoreau explored an interesting relationship between consumption, material possessions, and the actual cost of acquiring something in his 1854 book Walden, where he suggested that the real cost of an object should include the amount of human life exchanged for it rather than merely its monetary price. Although Thoreau was writing about a very different economic and technological environment, the idea becomes relevant when we consider how digital services have changed our perception of consumption.
When we generate an image, the financial cost may be almost negligible; when we upload another collection of photographs, the storage may already be included in a monthly subscription; and when we deploy an application, the infrastructure may be available through a cloud provider with only a few clicks. However, the absence of a visible price does not mean that the underlying activity is without cost, because someone still has to manufacture the hardware, supply the electricity, maintain the infrastructure, and manage the environmental consequences of extracting and processing the materials required to operate it.
In this sense, we may have become accustomed to measuring digital consumption through the price presented by a service rather than the resources required to provide that service, which creates an interesting disconnect between what something costs us personally and what it costs the larger system supporting our convenience.
A Different Engineering Instinct
I have been thinking about this through a relatively small engineering project of my own, where I built a desk-presence system intended to ensure that my computer does not continue operating unnecessarily when I am not in the room. The idea is straightforward: rather than treating a powered-on computer as the default state simply because it is convenient, the system considers whether there is a meaningful reason for the machine to remain active, and adjusts its operation accordingly.
Obviously, one computer switching off when its owner leaves a room will not transform global electricity consumption, nor should we exaggerate the environmental significance of an individual automation project. What matters to me is the engineering philosophy behind it, because it encourages us to reconsider an assumption that has become deeply embedded in our relationship with technology: that machines should remain available and operational regardless of whether we currently need them.
The same reasoning can be applied to software, cloud infrastructure, AI workloads, and data storage. Not every problem requires a large language model, not every application needs another background service, not every dataset deserves indefinite storage, and not every computer needs to operate at full capacity simply because the infrastructure allows it. In many situations, a smaller model, a simpler implementation, an appropriately sized server, or a properly managed shutdown policy can deliver the required outcome without introducing unnecessary computational demand.
Researchers studying what they call digital sufficiency have explored this idea more systematically, arguing that sustainable digitalisation requires more than improving the efficiency of individual devices; it also requires reconsidering how much hardware, software, data, and computational capacity we actually need. This distinction matters because a system can become more efficient at performing individual tasks while the total number of tasks increases so dramatically that overall resource consumption continues to rise, a phenomenon commonly associated with the rebound effect.
Perhaps Progress Requires Knowing When to Stop
There is something fascinating about the direction in which human technology is developing: we are becoming increasingly capable of producing things that previously required enormous amounts of knowledge, time, and specialised equipment, yet the easier creation becomes, the more important it may be to understand the difference between what we can produce and what is actually worth producing.
I do not think the answer is to reject artificial intelligence, abandon cloud computing, or return to an earlier technological era, because these systems have enormous potential to improve scientific research, healthcare, education, industrial efficiency, environmental monitoring, and the way we solve problems that would otherwise remain difficult or inaccessible. The more important question is whether we can develop these capabilities without assuming that every increase in computational capacity, digital consumption, or software production represents progress in itself.
Perhaps we need to reconsider the way we evaluate engineering achievements, because a system that solves a problem using fewer resources may be more valuable than one that demonstrates how much computational power we can assemble, just as software that performs its intended function without unnecessary complexity may be more intelligently designed than an application containing dozens of services that nobody fully understands or maintains.
The next generation of technology should certainly be more capable, but it should also be more conscious of the physical systems on which its capabilities depend. We need to build software that respects hardware limitations, infrastructure that operates according to actual demand, AI systems whose benefits justify their resource requirements, and an economy that recognises the difference between making something possible and making it necessary.
The planet is not an abstract resource pool waiting for our next technological invention, and the infrastructure supporting our digital civilisation does not exist outside the ecological systems that make human life possible. We can continue expanding the cloud, increasing computational capacity, and making the production of digital objects almost effortless, but none of these achievements changes the physical boundaries within which they must operate.
Perhaps the most meaningful technological progress will not come from teaching machines how to perform an unlimited number of tasks, but from developing the intelligence to understand which tasks are worth performing in the first place.
After all, we may eventually build machines capable of computing almost anything, but we will still have to live on the same planet that supplies the resources required to run them.