WEMO 2025 (complet) - Flipbook - Page 41
W E M O 202 5
O U T LO O K
√ Industrial Processes redesign: Simulate and redesign
energy-intensive
processes,
like
cement
or
steel production, to enable lower consumption.
Example: Gen AI has been used to analyze a factory’s
energy usage patterns and generate a retro昀椀t plan to
cut consumption by 20%.
• Despite promising Gen AI driven improvements,
challenges must be overcome including:
√ Data Availability and Quality: Generative AI relies on highquality well-organized datasets which is still a challenge
for many companies.
√ AI models understanding: AI and generative AI relies on
large and complex models that are not fully understood
yet, leading to a huge number of untested cases with no
direct guarantee of their relevance.
304
https://www.iea.org/news/ai-is-set-to-drive-surging-electricity-demand-from-data-centreswhile-o昀昀ering-the-potential-to-transform-how-the-energy-sector-works
305
https://encoradvisors.com/data-center-cooling-companies/#:~:text=What%20are%20the%20
latest%20innovations,for%20昀氀exible%20data%20center%20designs
√ Computational Costs: Training and running generative
models can be energy-intensive, requiring careful
management to avoid negating energy savings.
√ Organizational changes and people: New agile
organizations and procedures must be implemented.
Moreover, there is a need for employee training and
job de昀椀nition shift, in particular for junior tasks. Also,
recruiting scarce AI talent may prove to be di昀케cult.
√ Electricity consumption: In 2024, generative AI
signi昀椀cantly increased electricity consumption in
Western countries, primarily due to the energyintensive nature of training and deploying largescale AI models in data centers304 (see above).
Direct-to-chip liquid cooling and oil immersion cooling
are key advancements, enabling higher chip density.
Furthermore, AI-powered systems optimize cooling
performance based on real-time data analysis. 305
WEMO 2025
• Energy Savings: Gen AI can identify opportunities for energy
conservation Applications include:
√ Self-ful昀椀lling loops: with successful usage of AI models to
predict failures, usual failures should be avoided and will
progressively disappear from the training of the next AI
models. Thus the AI trained agent can misunderstand
critical situations.
40
• New Materials: Gen AI can accelerate the discovery and
design of advanced materials for energy applications, such
as batteries, solar cells, and superconductors.
