Original scientific paper

Powering generative AI and electrification

John Wanjiku

Abstract

Generative AI is expected to boost the global GDP by 7% over the next decade, translating to about US$7 trillion [1]. Despite its potential, its immense computation power and memory make datacenters both energy- and water-intensive. For instance, training a transformer model (the "T" in ChatGPT) utilizes approximately 5 GHz of processor clock speed, 3 terabytes of memory, and 700,000 liters of cooling water – producing emissions equivalent to five times those of an average car [2], [3], [4], [5]. A ChatGPT query uses 8 times more power than a Google search, while an AI image uses energy equivalent to a smartphone.

Keywords

AI; generative AI; electrification, data centers; transformer model

Hrčak ID:

331608

URI

https://hrcak.srce.hr/331608

Publication date:

19.5.2025.

Visits: 596 *