WEMO 2025 (complet) - Flipbook - Page 43
W E M O 202 5
O U T LO O K
LLM Large Language Model
https://www.weforum.org/stories/2024/07/generative-ai-energy-emissions/
316
https://www.bcg.com/publications/2023/how-ai-can-speedup-climateaction#:~:text=1.,related%20adaptation%20and%20resilience%20
initiatives.
317
https://www.cnbc.com/2024/05/15/microsofts-carbon-emissions-have-risen-30percent-since2020-due-to-data-center-expansion.html
314
315
• Quantum computing: Present prototype quantum
computers consume large amounts of electricity, although
they are in limited numbers so far. However, in the for
eseeable future, the first machine capable of using a
signi昀椀cant number of qubits (quantum bit, a basic unit of
quantum information) will be ready. They will be able to
crack almost all the current cybersecurity systems, posing
a huge threat to all critical defense and 昀椀nancial systems.
Today the only known non-quantic options use much more
energy-intensive algorithms and communications. This may
pose an additional stress on the electricity system.
WEMO 2025
For example, utilities – organizations with thousands of
miles of transmission lines, pipelines, and other remote
infrastructure – often spend millions of dollars on asset
integrity. Corrosion and predictive-maintenance models
can be retrained with the integration of many sources of
data, including traditional records, such as past damage,
visual inspections, and data from sensors on the asset itself.
Other sources of data, such as drone and satellite-based
images can also be integrated. Thus, AI and Gen AI can
significantly improve the effectiveness of this core function.
• Gen AI and emissions315: According to certain studies, CO2
emissions could potentially be reduced by 5-10% by 2030.316
However, the electricity consumption growth driven by LLM
development, raises concerns about reliance on fossil fuels
in regions with insu昀케cient low emission electricity capacity.
For example, Microsoft’s emissions are up 30% since 2020
due to data center expansion317. These dynamics underscore
the need for energy-e昀케cient AI algorithms, e昀케cient data
center cooling and low emission electricity availability for
data centers.
42
By contrast, the ambitious, groundbreaking use cases
are more innovative and consequently require more
customization – sometimes even a trained-from-scratch
LLM314. And although they may have the potential to
deliver signi昀椀cantly more value, they also require substantial
upfront investment in capabilities and infrastructure.
