The cloud has weight

Data centers, networks and devices turn invisible clicks into material and electrical demand.

A click crosses real machines

The "cloud" is a physical industry, distributed across the world

When someone sends a message, runs a search or plays a video, that action crosses an entire chain of specific physical machines: the device itself, a home router, mobile-phone antennas or fiber-optic lines, and finally one or several data centers that process the request and return the answer, often also generating redundant backup copies for reliability. Every one of these links consumes electricity continuously, not only at the exact instant it is used. Building the servers and chips that make this whole process possible also requires extracting minerals from mines scattered around the world, manufacturing those components in extremely high-purity industrial plants that consume large amounts of energy and water, and running cooling systems that constantly remove the heat generated by the processing itself. None of this is metaphorically a "cloud": it is a physical industry distributed across the entire planet, as real as any factory.

Not every byte weighs the same

A generic figure like "so many grams of CO2 per email" or "so much energy per search" can sound precise, but it actually depends on such a large number of variables — the type of network used, the specific device's efficiency, the efficiency of the data center processing the request, the system's total load at that moment, and the mix of electricity sources feeding that region's data center — that any universal figure ages quickly and usually hides the specific assumptions it was calculated under. This became very clear with estimates of how much energy an AI assistant query consumes: early figures suggesting it used ten times more energy than a conventional search were based on comparing an old, already-outdated estimate of a search's cost with a more recent, more complete estimate of an AI query's cost, artificially exaggerating the real difference between the two.

The rebound effect: more efficient does not always mean less total consumption

When a technology becomes more efficient and, as a result, cheaper to use, the usual reaction is not to keep the same level of use and pocket the difference, but to take advantage of that greater efficiency to do far more operations than before. This phenomenon, known as the rebound effect, explains why data centers' and networks' technical efficiency has improved steadily for years while the sector's total electricity consumption kept growing: total demand for digital services — high-resolution video, cloud gaming, queries to increasingly powerful artificial intelligence systems — has grown faster than the technical efficiency gains achieved in parallel.

What happens at the end of a device's service life also counts

When a phone, computer or any other electronic device stops being used, its story does not end there: it becomes electronic waste, a category of waste that is growing rapidly worldwide and that, for the most part, is still not being collected or recycled in a documented, responsible way. That waste contains both valuable materials that could be recovered and reused, and potentially polluting substances that, without proper treatment, can end up leaching into soil or water. Extending a device's service life through careful use, repair and software updates, rather than replacing it prematurely, remains one of the individual decisions with the greatest real effect on digital technology's total material impact, precisely because manufacturing the device from scratch — extracting the minerals, manufacturing the components, assembling the unit — concentrates a very substantial share of its total lifecycle environmental cost.

Useful digital sufficiency

  • Extend the life of your phone and computer; repair when reasonable instead of replacing prematurely.
  • Avoid maximum-resolution video on small screens when it provides no perceptible real benefit.
  • Clear out redundant storage and duplicated backups with judgment, not compulsively.
  • Turn off equipment that is genuinely idle, without turning it into a ceremony if you'll use it again within minutes.
  • In organizations: measuring real server utilization, buying clean electricity, and reusing data centers' waste heat can outweigh thousands of symbolic individual gestures in real effect.

Don't substitute deleting emails for keeping hardware

Deleting old emails or duplicate photos has a real but modest effect on storage consumption; manufacturing the device itself usually concentrates a much larger share of its total lifecycle environmental cost. Use it well, repair it, and keep it longer before replacing it.

The numbers, with their source

Data centers, around 1.5% of world electricity, and growing fast

According to the International Energy Agency (IEA), data centers consumed around 415 terawatt-hours in 2024, about 1.5% of all electricity consumed worldwide. The United States accounts for 45% of that global consumption, China for 25%, and Europe for 15%. That consumption is growing very quickly — 17% in 2025 alone — driven especially by artificial intelligence: AI-dedicated servers accounted for 24% of server electricity demand and 15% of total data center consumption in 2024, and the IEA projects that global data center consumption will double to around 945 terawatt-hours by 2030, roughly 3% of that year's global electricity consumption.

Sources: IEA — Energy and AI, executive summary; IEA — Data center electricity use surged in 2025.

Why the "10 times more than a search" figure was inflated

Recent independent estimates put the consumption of a typical query to an artificial intelligence assistant at between 0.3 and 3 watt-hours, well below the initial reference figure of 3 watt-hours, which assumed the least efficient models. The "10 times more than a Google search" comparison was also based on a 2009 estimate of a search's cost (0.3 watt-hours); more recent estimates put that cost at around 0.04 watt-hours, an almost tenfold improvement left out of the original comparison, which artificially inflated the perceived difference.

Sources: Epoch AI — How much energy does ChatGPT use?; independent analyses collected on engineeringprompts.substack.com and zmescience.com.

E-waste is growing five times faster than its documented recycling

The Global E-waste Monitor 2024 estimates the world generated 62 million tonnes of electronic waste in 2022, 82% more than in 2010, and projects that figure will reach 82 million tonnes by 2030. Only 22.3% of that mass was documented as collected and recycled in 2022, a share the report itself expects to fall to 20% by 2030 if current trends do not change — e-waste is literally growing five times faster than the documented capacity to recycle it.

Source: UNITAR — Global E-waste Monitor 2024.

What this table is missing

None of these global figures replace evaluating a specific digital service: a particular data center's efficiency, the electricity mix of the region powering it, and the system's real load at the moment of use change the outcome from case to case far more than any global average can capture.

Sources