Can we make AI less power-hungry? These researchers are working on it.
Briefly

In November 2024, the FERC rejected Amazon's bid to directly purchase 180 megawatts from a nearby nuclear plant, citing fairness to other energy users. This decision highlights the impact of surging power demands driven by data centers, especially as they utilize complex AI models. The historical backdrop includes the landmark development of AlexNet in 2012, which illustrated the need for advanced computational resources in AI research. This shift not only underscores the technological evolution but also the regulatory and operational challenges faced by major entities in the energy sector.
Mark Christie, a FERC commissioner, emphasized that there is a significant surge in demand for power, especially with data centers utilizing advanced AI models.
The rejection of Amazon's request by FERC demonstrates the regulatory challenges and competitive concerns surrounding direct energy purchases specific to data centers.
AlexNet's development marked a pivotal shift in AI capabilities, showcasing that larger neural networks could drive advances in image recognition technologies.
The growing appetite for power from data centers highlights a crucial junction in energy distribution and AI development, affecting both industry practices and regulations.
Read at Ars Technica
[
|
]