Artificial Intelligence (AI) is often hailed as the technology of the future. It is transforming industries, revolutionising healthcare, optimising logistics, and reshaping education.
Beneath the excitement lies a lesser-known reality: the considerable environmental cost of developing and deploying AI technologies.
While discussions around AI’s societal impact are widespread, few conversations address its ecological footprint. Training large AI models can require as much energy as hundreds of households consume in a year. This contributes significantly to carbon emissions.
To better understand how companies are tackling these hidden costs, I spoke with Dimitri Picasso, Head of Compliance at Infomaniak, a Swiss technology company and a leader in sustainable digital infrastructure. His insights reveal how organisations can measure, manage, and reduce the environmental impact of AI.
The computational demands of AI are rising fast. Every large language model, from training to deployment, requires vast amounts of data processing and energy. Training a model like GPT 3, for example, is estimated to emit as much carbon dioxide as five cars over their lifetimes. Yet, this environmental toll often remains invisible to users.
According to Dimitri Picasso, the problem is twofold: "Working with AI is a plus, but you need to be aware of the consumption," he says.
Public awareness of AI’s ecological cost is limited. Most people are amazed by AI’s capabilities but rarely think about the energy needed to power them. Picasso notes that even generating a single prompt involves significant computation. "If you talk to someone not working in the tech industry, they usually have no idea about the environmental impact generated by AI calculations," he explains.
This lack of transparency creates an urgent need for clearer communication. It also calls for responsible innovation. Big tech companies must lead efforts to educate the public and design more sustainable AI models.
Most companies are still searching for ways to make AI more sustainable. For Infomaniak, the first step is clear: measurement.
For Picasso, "Sustainable AI means measuring impacts — the consumption of energy and improving the efficiency and lifespan of machines."
At Infomaniak, sustainability starts by evaluating machine efficiency and carefully selecting hardware. By choosing efficient servers and extending their lifespan, they aim to lower the environmental toll of their infrastructure. They are also developing indicators to track key factors. This includes energy consumption, server efficiency, and hardware lifespan. This data will guide smarter technological investments.
Infomaniak’s approach is not limited to its own operations. The company encourages other organisations to adopt similar practices, highlighting a key principle: before improving your impact, you must first measure it.
While many companies are just beginning to explore sustainable AI, Infomaniak is already taking action.
One of their flagship initiatives is the D4 Project, focusing on energy reuse and sustainable datacenter management. This video link provides more information on how the D4 Project redefines sustainable data infrastructure
As Picasso explains, “The core idea is to use all the renewable electricity consumed by this data center twice — once to power computing, storage, and AI, and a second time to heat homes,” says Picasso. “Today, this energy is still mostly wasted into the atmosphere around the world.” In addition, the servers are exclusively cooled using the cold generated by heat pumps during the process of raising temperatures to meet the needs of the SIG district heating network. That’s not all — this data center has no visual impact on the landscape. It is installed underground, in an eco-district, placing it in direct proximity to the district heating network. It proves that today’s data centers no longer need to be built outside of cities.
This innovation goes beyond Infomaniak’s own needs. The D4 Project is open source, and its technology and methods are freely available for others to adopt via d4project.org. The main goal is to set a new standard for datacenter sustainability. To facilitate large-scale replication, institutions such as EPFL Lausanne, the University of Lausanne, and IMD have thoroughly documented the datacenter’s operation. Beyond the D4 Project, Infomaniak continues to integrate sustainability into all new infrastructure projects and collaborates actively with academic institutions to further research in this field.
“The D4 Project was developed and implemented by Infomaniak,” says Picasso. “EPFL, the University of Lausanne, and IMD collaborated closely with our technical teams to document the datacenter’s operation in detail — including the development of measurement indicators for energy efficiency, heat pump performance, and the full recovery of the datacenter’s energy consumption as usable heat.” Infomaniak shows that sustainability in AI is not just a vision. It is a reality, achievable through action.
As AI becomes more embedded in society, its environmental impact demands more attention. AI systems rely on energy intensive computations at every stage. From training to real time use, they contribute significantly to carbon emissions.
Despite growing awareness, the industry lacks clear standards for measuring AI’s environmental footprint. Metrics like carbon emissions per model and water usage for cooling exist but are not consistently disclosed.
What is needed, and what the world is calling for, are:
Some companies are responding by optimizing models through reduced complexity, designing more energy-efficient AI architectures, and adopting circular economy practices to extend hardware lifespans.
Beyond infrastructure projects like the D4 Project, Infomaniak is managing AI resource consumption internally. "We work with tokens for employees using our internal AI tools, and we are starting to measure the power consumption per core to track and optimize AI usage," says Picasso.
By limiting tokens and keeping a detailed inventory of AI hardware, they can closely monitor energy demands. It's a simple yet effective strategy that demonstrates how innovation can be aligned with ecological responsibility. In a sector where ethics and innovation often lead the conversation, Infomaniak’s example is a timely reminder that environmental sustainability must also be a core pillar of responsible AI.
Building a sustainable AI future will require more than small improvements. It demands a complete shift in how AI is developed and deployed.
One promising direction is frugal AI. This movement focuses on building smaller, more efficient models that consume less energy without sacrificing performance. Federated learning also shows promise. It keeps data local and trains models collaboratively, cutting down on energy usage.
Another emerging idea is eco design for AI. This approach integrates sustainability into model development from the start. Developers are challenged to think not just about speed and accuracy, but also about energy efficiency.
As always, collaboration is key. Universities, private companies, and governments must work together. They need to set shared standards, create green certifications, and form partnerships that drive change.
The road ahead is long, but companies like Infomaniak are setting an example. Through actions like recycling datacenter heat and monitoring AI energy consumption, they show that sustainable innovation is possible.
By promoting open-source innovation and collaborating with research institutions, they prove that progress and responsibility can go hand in hand.
Artificial Intelligence holds immense promise - but realizing that promise responsibly means addressing its often-overlooked environmental impact.
The future demands innovation with a conscience. We must design AI models that are not only powerful but also energy-efficient. We must build infrastructures that are both resilient and sustainable. And we must cultivate a culture grounded in transparency and accountability.
Early initiatives like those at Infomaniak show that a different, more responsible path is possible. They demonstrate that technological advancement and environmental stewardship can — and should - go hand in hand.
The future of AI isn’t just about creating smarter machines; it’s about ensuring that today’s breakthroughs support a greener, more sustainable tomorrow.
To learn more about the D4 Project, visit their website here.
By the same author:
How Academic Research Methods Help Companies
The Future of AI: Insights from GBN Talk Show
Images:
denisismagilov 652917768
SergeyNivens 10828497
PhD in telecommunications from King's College London, seeking transition to the private sector as a data scientist. Scientific Collaborator at HEG and UNIGE with expertise in data analysis and machine learning in maritime and healthcare sectors. Expert in Python and R, committed to transforming data into actionable insights for businesses.
