The hidden environmental
and social costs of data centers

And why data center companies going beyond PUE and GoOs should be rewarded

Johanna Flood

For many years, the share of electricity going to data centers was fairly stable, around 1%. Even as computing demand grew exponentially, improvements in server efficiency led to decreased costs and energy use per unit of computing power. Not anymore.

Not too long ago, something changed. In 2023, ChatGPT and AI reached the broader public. By 2024, the share of global electricity used by data centers had increased to 1.5%. By 2030, it is expected to double to around 3%.

High in energy consumption

The energy demand of data centers is high. One hyperscale data center can have a power demand of 50 MW. An AI-focused data center can require as much as 100 MW.

For most of us, these numbers don’t mean much. To put a 100 MW data center into perspective, you would theoretically need around 33 wind turbines running at full capacity all the time to power it. But in reality you need many more.

The annual energy consumption of a data center campus can be up to 1 TWh for the largest facilities, which is equivalent to the electricity use of a city with 70,000 to 100,000 homes.

Water for cooling – but not everywhere

Some data centers, luckily not all and not everywhere, use water for cooling. If a data center relies on water cooling and is located in a warm climate, it can use large amounts of water. In warm places, the water use can reach hundreds of millions of liters per year. That is comparable to the annual water use of 1,000 to 3,000 households, only that in some cases the data centers get the water and not the households.

In colder regions, such as the global North, water cooling may only be needed during the hottest months, but with increasing draught because of climate change also up here in Scandinavia, that is when we are also suffering from low ground water levels. Some data centers do not use water at all for cooling; they use more energy.

Complex value chains

The value chain of data centers is complex and often begins in mines. Data centers are essentially buildings full of electronic equipment made from metals sourced through mining. They also require land, and construction can take years, creating local environmental and social impacts for nearby communities.

What is the embodied carbon of a data center?

The embodied carbon of a data center varies depending on size and materials, but a large facility can have around 10,000 tonnes of CO₂e from the building and another 10,000 tonnes from equipment (cooling, power, etc.). On top of this come the servers, which are often owned by clients in colocation facilities. A gigantic data center campus can house up to 100,000 servers, and one server can have an embodied carbon footprint of around 1 tonne of CO₂e (this of course varies too). This means the total embodied carbon of a data center can reach around 120,000 tonnes.

Even though the embodied carbon footprint is larger than the annual emissions of many companies, the majority of emissions over a data center’s lifetime come from operational energy use.

Renewable energy still comes with a carbon footprint

If the data center uses renewable energy, these emissions can be reduced to zero for Scope 2, according to Greenhouse Gas Protocol accounting principles. However, going back to the wind turbine example, we might theoretically need around 100 turbines to power such a facility. If one wind turbine has an embodied carbon footprint of roughly 1,000 tonnes, the total theoretical embodied carbon associated with powering one data center could be:

10,000 + 10,000 + 100,000 + (1,000 × 100) = 220,000 tonnes of CO₂e

The embodied carbon comes from the materials used in the data center. These materials will eventually become e-waste and construction waste after 15–20 years. What is often overlooked, however, is the upstream waste from electronics production. A smartphone weighing 200 grams can generate tens of kilograms of waste during production. For a computer, this can be up to 100 times its weight. I have not seen specific figures for servers, but given the similar materials, it is reasonable to assume the upstream waste is significant.

What can we do?

A text prompt has a relatively small impact, but still higher than a standard search. Generate an image, and the footprint grows. Generate a video, and it can jump dramatically, often many times higher in both energy and water use.

It is clear that we, as AI users, need to use these tools with care.

But we all have a role to play – not only data consumers.

Data center operators need to ensure that the industry grows responsibly. The equipment we buy today will become waste in 15–20 years. And if digitalization is to be part of the solution, it needs to replace and not add to other resource-intensive activities. Data center operators that take sustainability seriously – by that, I mean going way beyond PUE and GoOs – need to be rewarded.

Investors and banks need to understand the full sustainability impact and set stronger, more holistic ESG requirements for data centers.

Municipalities need to ask how data centers give back to society – not only through waste heat, but also from a biodiversity, water, and social perspective. Are they using local contractors, or flying people in from across the world? Are they minimizing disturbance and including local communities? Are they choosing materials and equipment with the lowest environmental impact, ideally supporting local companies?

And maybe we should aim a bit higher.

What if data centers were not only efficient, but also good neighbors?
What if they contributed to local communities and ecosystems, not just consumed resources?
What if they were places where people grow – not just professionally, but also as humans?

And we need to raise awareness in organizations to use AI and data with care—not enable AI plug-ins for everything.

Data is a resource like any other – let’s treat it that way

We need to start seeing data as a resource – just like water, energy, or food – and treat it with respect.

In a webinar I hosted not long ago for ISEP, someone in the audience suggested that we should pay for the data we use. I like that idea.

AI has enormous potential to bring valuable intelligence to the world. But we should use it for meaningful purposes – not just to create videos of ourselves doing backflips in the Alps or to rely on AI for everything instead of using our own brains.