The carbon footprint of AI is much higher than reported because we've been ignoring the 'waste' of failed experiments.
April 15, 2026
Original Paper
Environmental Footprint of GenAI Research: Insights from the Moshi Foundation Model
arXiv · 2604.11154
The Takeaway
By including the 'hidden' compute of failed runs, debugging, and ablation studies, researchers quantified the true environmental cost of a foundation model (Moshi). The industry typically only reports the final training run, but this 'waste' can double or triple the actual footprint. It reveals that the R&D phase is just as energy-intensive as the final production. For companies focused on ESG, this is a call to include the entire R&D lifecycle in their carbon accounting. It's a reality check for an industry that often claims to be 'green' while hiding the waste of the experimental process.
From the abstract
New multi-modal large language models (MLLMs) are continuously being trained and deployed, following rapid development cycles. This generative AI frenzy is driving steady increases in energy consumption, greenhouse gas emissions, and a plethora of other environmental impacts linked to datacenter construction and hardware manufacturing. Mitigating the environmental consequences of GenAI remains challenging due to an overall lack of transparency by the main actors in the field. Even when the envir