Can AI Solve Its Own Energy Problem With Machine Learning?
By 2030, AI-related data centre energy use could consume up to 3% of all global power consumption, with some regions facing far steeper demands.
The International Organization for Standardization recently published guidance ISO/IEC 42005 to help companies assess AI’s broader impact on society, including environmental costs.
For example, Ireland, which hosts major digital service providers, AI-related data centre energy use could reach 35% of national power consumption.
However, Philippine de T’Serclaes, Chief Sustainability Officer (CSO) of French software company Dassault Systèmes maintains optimism about AI’s environmental potential.
The potential of machine learning for data centre sustainability
Recent studies show machine learning (ML) algorithms can improve grid efficiencies by 15% and boost battery storage efficiency by 10-20%.
AI could also reduce schedules for new clean energy projects by approximately 20%, potentially saving hundreds of billions of dollars by 2050.
Additionally, McKinsey estimates that AI and ML could accelerate 47% of initiatives required to achieve the global 1.5-degree pathway under the Paris Agreement.
This means that the technology shows promise for sustainable innovation. For instance, AI has developed lightweight packaging requiring less transport energy and advanced materials for efficient batteries.
Furthermore, recent research demonstrates AI-developed paint coatings that could reduce building temperatures by 20°C.
We have what we need to make the infrastructure and ecosystems that power AI start working more effectively today.
However, current deployment patterns raise concerns about energy sourcing.
Dassault Systèmes’ partnerships for efficiency gains
MIT’s Technology Review recently warns that rapid application growth means “data centers are expected to continue trending towards using dirtier, more carbon intensive forms of energy.”
Tackling this challenge, Dassault Systèmes partners with Quanta Cloud Technology (QCT), the Taiwanese server manufacturer, for efficiency gains – and Philippine advocates “Frugal AI” approaches that emphasise lightweight models and measurable impact.
Model pruning allows programmers to remove unnecessary neural network connections, reducing computational requirements while maintaining accuracy.
This technique cuts energy consumption without sacrificing performance.
Meanwhile, data centre optimisation offers immediate efficiency opportunities.
Cooling systems account for up to 40% of energy use and selecting effective systems can improve efficiency by 30%.
Philippine references work at Dassault Systèmes using virtual twins on the 3DEXPERIENCE platform to enable sustainable operations through partner collaboration.
Modelling scenarios perform trade-off analyses between technology configurations, helping teams optimise setups and accommodate future technologies.
Dassault Systèmes also collaborates with QCT to simulate data centre heat and airflow patterns for building effective air conditioning systems.
The company’s solutions also supports Bouygues Construction, the French building company, in modular construction processes, while also helping Olivier Naar design modular nuclear reactors.
“I don’t see isolated projects,” Philippine says, “I see interconnected, mutually enhancing nodes within a wider value network.
“I see how those same techniques can help build modular data centers. I see how we can power them with electricity that is low-carbon and convenient.”
The importance of broader thinking
Network effects emerge through initiatives like the Coalition for Sustainable AI, where Dassault Systèmes participates.
However, Philippine emphasises that technical fixes alone cannot address sustainability challenges. Processing power value depends on application purposes, regardless of data centre efficiency.
“AI will be what we make of it,” she says.
“Circular thinking needs to be embedded not just in our processes and products, but also in our methods and in the way we think about the world.”
She concludes: “We have what we need to make the infrastructure and ecosystems that power AI start working more effectively today.
“What we make of it? Well, that’s up to us.”
