Sustainable data centers: the environmental cost that became a public issue

The narrative that cloud computing is clean and virtually immaterial has held up for more than a decade — and it was too convenient to be true. What was happening behind the scenes was the silent growth of one of the most energy and water intensive sectors in the world, protected by the technical abstraction that separates user from hardware. This veil is being ripped away, and whoever decides on infrastructure needs to understand what is being revealed.

The scale of consumption that became invisible

A large data center consumes between 100 and 200 megawatts — equivalent to the electrical consumption of a city with 80 thousand inhabitants. The largest hyperscale facilities already exceed 500 MW, driven by AI models that require clusters of GPUs running non-stop. Electricity was never the secret: sustainability reports always mentioned it. What was hidden was the water.

Evaporative cooling consumes volumes that seem absurd when exposed. A typical data center uses between 1.5 and 5 liters of water for each kilowatt-hour processed — on a 100 MW campus, this exceeds 4 million liters per day, the consumption of more than 20 thousand people. This demand was left out of discussions because water for cooling was not included in carbon reports; the standard metric, PUE, measures energy but does not touch water.

The debate changed when the shortage reached the data centers themselves. The Netherlands restricted expansion permits in Amsterdam following water pressure in the Noord-Holland region. Arizona saw city councils questioning how much of the Colorado River — already under severe pressure — would supply the next Microsoft or Google campus. When the resource starts to run out, the tacit consensus of unrestricted consumption cracks.

Regulatory pressure that is here to stay

The EU AI Act requires transparency about energy consumption of high-impact AI systems, and the CSRD requires companies to disclose auditable environmental metrics that include the indirect energy and water consumption of cloud providers. For those who outsourced the infrastructure, this means that Scope 3 now has an IP address.

In the US, the SEC advanced climate risk disclosure rules that, even in legal dispute, created market pressure before they came into force. Institutional investors are already demanding consistency between ESG goals and cloud supply contracts. A company that announces net zero for 2035 and runs everything in regions supplied by coal has a credibility problem that auditors will explore.

What matters is not just complying with the letter of the regulation — it is realizing that this environment created accountability where there was none. Permits for new data centers in water-stressed areas are difficult to obtain, and restrictions have already paused expansions in regions that five years ago seemed like obvious markets.

What hyperscalers are actually doing

The response from large providers combines genuine moves with strategic communication. Renewable energy Power Purchase Agreements (PPAs) are real: Google, Microsoft and AWS have signed contracts with wind and solar farms totaling tens of gigawatts — although the correspondence between consumption and generation is still disputed methodologically.

Liquid cooling is the most substantive technical change. New designs circulate fluids directly over components instead of cold air, reducing water consumption per kilowatt-hour and enabling higher computing densities. Microsoft tested data centers submerged in seawater; Google uses waste heat from its servers in Finland to heat buildings in Hamina — a byproduct turned into a municipal service.

There is still a race for more efficient chips. Google's TPUs and AWS's Trainium and Inferentia chips squeeze out more compute per watt than general-purpose GPUs. What started as cost optimization became a sustainability argument — the effect is real regardless of the motivation.

The geography of data centers is redesigned

The cloud infrastructure map is being rewritten by climate and the availability of renewable energy. The Nordic countries — Sweden, Norway, Finland, Iceland — have become preferred destinations: cold air reduces cooling costs, hydroelectricity and geothermal energy provide abundant clean energy, submarine cables resolve latency for European loads. Iceland is the extreme case — low temperature all year round, 100% renewable energy, almost free cooling. What made it peripheral became an advantage when sustainability entered the spreadsheet.

Regions historically attractive for existing infrastructure — parts of the American Southwest, certain European industrial zones — are losing projects to locations previously considered unlikely. This reconfiguration affects latency, data compliance, and colocation options. Those who ignore these movements when planning where to run their workloads may end up stuck with infrastructure under regulatory pressure or with less future investment.

What changes for those who decide infrastructure

For CIOs and CTOs, sustainability is no longer a public relations attribute and has become an architectural variable. The choice of cloud region now goes beyond latency and price per hour: the carbon and water footprint of workloads needs to be traceable to feed auditable ESG reports. Providers that do not deliver granular environmental consumption metrics by region and service create increasing friction with compliance teams.

The placement decision requires new criteria. A five-year contract with a data center in a water-stressed region could become an operational risk if expansion licenses are suspended or water rates rise. Asking the supplier about power source and cooling method is not decorative due diligence — it is continuity risk assessment.