Concerns are being raised over the economic and environmental costs of the increasingly vast computational demands associated with continue to push advances in AI and deep learning. 

The computational scale of training the size of model now required by cutting-edge deep learning techniques has been estimated to incur costs of up to US$100 billion and result in levels of carbon emissions equivalent to those emitted by New York City across a one month period. 

As Thompson et al. (2021) understatedly put it, “the cost of [AI] improvement is becoming unsustainable”.



Thompson, N. Greenewald, K., Lee, K., Manso, G. (2021).  Deep learning’s diminishing returns.  IEEE Spectrum, 24th September,