Sustainable IT: The hidden Environmental Costs of generative Artificial Intelligence
In a recent podcast, Ivan Mariblanca Flinch, founder of Swiss start-up MIKUJY and Green IT / Sustainable IT expert specializing in measuring the digital footprint of companies, warns of the environmental drifts of generative AI. He describes a little-known reality: AI is far from immaterial, and its use generates significant digital pollution, often invisible to the end user.
Real digital pollution
“When we talk about the cloud, we imagine something light, airy, almost invisible. But the reality is quite different. Digital technology involves emissions and the consumption of metals, fossil fuels and electricity.”
Ivan Mariblanca Flinch points out that digital technology is based on three material pillars: electronic devices, networks and data centers. There are now more than 34 billion connected devices in the world, with an average of 11 devices per inhabitant in Europe.
Digital infrastructures are also extremely resource-intensive:
- The water used to cool servers
- The electricity needed to power data centres,
- Rare metals extracted on a massive scale, sometimes under questionable ethical and environmental conditions.
Generative AI: an energy-hungry model
Training a large-scale generative AI model such as GPT-3 alone consumed over 700,000 liters of water. Routine use of these models, such as a simple question-and-answer session, also represents a sizeable footprint: 25 to 50 queries are enough to evaporate half a liter of water, just to cool the servers.
Then there’s the impact of the rare materials* needed to manufacture computer hardware. *Rare materials, often referred to as “rare earths” (e.g : europium, yttrium, terbium, galium, tungstène, indium, tantale…), are a group of chemical elements essential to modern technology . Used in batteries, electronic components and high-performance magnets, their extraction is particularly energy-intensive and polluting. The extraction of one kilogram of lutetium, for example, requires the processing of 120,000 tonnes of rock. These figures underline the growing gap between stated ecological ambitions and the industrial reality of AI. Indeed, ever since companies embarked on environmental and digital transitions, the exploitation of natural resources has never been so intense.
Measuring the digital footprint: a necessity
In light of these findings, Ivan insists on the importance of measuring the digital footprint in its entirety, integrating all stages of the equipment lifecycle: extraction, manufacture, transport, use, reuse and end-of-life.
In companies, 90% of the carbon footprint of personal equipment (computers, smartphones, screens) comes from its manufacture. Conversely, for data centers, it’s daily use that generates most of the impact. It is important to note that the environmental impact of data centers varies greatly depending on the country in which they are located.
The key indicators to monitor are :
- Energy consumption (and the PUE “Power Usage Effectiveness” ratio),
- Water use for cooling,
- Equipment lifetime.
Energy efficiency: the wrong solution?
One of the most striking points raised in the podcast is the paradox of energy efficiency. If equipment becomes more efficient, this often encourages an increase in usage: more data generated, more models driven, more storage. The result: a growing overall footprint, despite per-unit gains.
So the problem is not limited to the technology itself, but to the data society in which it is embedded. As Ivan points out: “We’re putting a luxury car in everyone’s hands, while asking them to take public transport.”
What responsibility do digital players have?
Public policies, companies and users all have a role to play. Regulations can set a framework, but without culture and awareness, they remain insufficient. For example, many large companies prefer to pay fines rather than change their business model.
At the same time, users have real power of influence. The simple act of changing browsers or choosing more sobriety tools can prompt the major web players to review their priorities.
The challenge is immense: in Ireland, a third of the electricity consumption forecast for 2030 could be attributed to data centers. In this context, every initiative aimed at reducing the digital footprint becomes strategic.
Towards sustainable IT
This podcast raises a central question: Can technological data center development be reconciled with digital sobriety? The answer lies in a structured approach based on measurement, transparency and strategy.
This is precisely the mission of MIKUJY, which supports organizations in the comprehensive analysis of their digital footprint, supports them on their Sustainable IT journey and helps them define concrete reduction actions thanks to our CircularIT SaaS platform.
