EUVIMEDEUVIMED entdecken
Crossref

Carbon Footprint of Training Large AI Models: Measuring and Proposing Mitigation Strategies for the Environmental Cost of AI

Urmi Gosaliya, Meet Dasalaniya, Aastha Jani, Vivek Gajjar

INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT · 25.04.2026 · 10 (04) · 1-9

DOI: 10.55041/ijsrem61165 ↗

Kurz gefasst

Fachartikel in der Zeitschrift INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT, online erschienen am 25.04.2026. Verfasst von Urmi Gosaliya und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Over the past decade, artificial intelligence has quietly shifted from a research curiosity to infrastructure that billions of people depend on every day. Large language models write code, assist doctors, translate languages, and hold conversations in ways that once seemed firmly out of reach. But behind every chatbot interaction and AI-assisted search result, there is a data center consuming electricity and in most parts of the world, that electricity still comes largely from fossil fuels. The carbon cost of building and running these systems is real, growing, and almost entirely invisible to the people using them. This paper investigates that hidden cost in depth. We examine how energy demands have grown alongside model size from BERT’s modest 652 kg of CO2 in 2018 to independent estimates placing GPT-4’s training footprint in the thousands of tonnes. We survey the measurement tools researchers use to quantify these emissions and expose the significant gaps that remain. We then present a comprehensive set of mitigation strategies spanning algorithmic efficiency, hardware improvements, geographic scheduling, renewable energy, and transparency standards. Finally, we honestly confront the structural barriers competitive scaling incentives, the Jevons Paradox, and the near-total absence of embodied carbon from current estimates that make progress harder than the technical solutions alone would suggest. The central argument is straightforward: sustainable AI is technically achievable. What it requires is the collective will to treat it as a genuine priority. Keywords: Artificial Intelligence, Carbon Footprint, Large Language Models, Energy Consumption, Green AI, Sustainable Computing, CO2 Emissions, Machine Learning, Inference Efficiency, Embodied Carbon.

Zitieren

APA

Gosaliya, U., Dasalaniya, M., Jani, A., & Gajjar, V. (2026). Carbon Footprint of Training Large AI Models: Measuring and Proposing Mitigation Strategies for the Environmental Cost of AI. INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT, 10(04), 1-9. https://doi.org/10.55041/ijsrem61165

Vancouver

Gosaliya U, Dasalaniya M, Jani A, Gajjar V. Carbon Footprint of Training Large AI Models: Measuring and Proposing Mitigation Strategies for the Environmental Cost of AI. INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT. 2026;10(04):1-9. doi:10.55041/ijsrem61165

BibTeX
@article{Gosaliya2026,
  author = {Gosaliya, Urmi and Dasalaniya, Meet and Jani, Aastha and Gajjar, Vivek},
  title = {Carbon Footprint of Training Large AI Models: Measuring and Proposing Mitigation Strategies for the Environmental Cost of AI},
  journal = {INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND MANAGEMENT},
  year = {2026},
  volume = {10},
  number = {04},
  pages = {1--9},
  doi = {10.55041/ijsrem61165}
}
Kostenlos starten →