Energy consumption should be considered from the outset of system design. Too often, new algorithms and AI systems are developed without sufficiently considering their computational and energy requirements, leaving efficiency challenges to be addressed later at the infrastructure or hardware level. Energy efficiency should instead become a core design constraint guiding the development of Data and AI technologies. BDVA highlighted this aspect in previous Strategic Agenda (see Pillar V: “Sustainable data and AI: enhancing efficiency and resilience while reducing resource demands”).
This direction should be reinforced by these major guidelines:
- Resource-efficient AI and computing, including efficiency-by-design and frugal approaches
- Energy-aware data and AI architectures, infrastructures and frameworks. To effectively reduce energy consumption across the entire AI lifecycle, it is essential to implement stakeholder awareness mechanisms. These mechanisms should target all relevant parties—from data center operators to end users—by promoting best practices such as optimising hardware utilisation, adopting renewable energy sources, and designing algorithms with energy efficiency in mind.
- Emerging energy-efficient computing paradigms, including the combination of neuromorphic approaches and spiking neural networks