The quick adoption of blockchain technology and generative AI is a major factor in the world's electricity use, which raises concerns about their long-term environmental impact. To save energy, the first thing you need to do is figure out how much energy you are already using. But because blockchain and generative AI are both cloud-based services, it's not easy to understand how much energy they use when they're not at your site. This makes it harder for companies and organisations that want to improve the accuracy of calculating Scope 3 emissions. This study determines the energy consumption of these technologies at both the system level and per-use basis, comparing them to traditional services such as payment networks and web search engines. For instance, Bitcoin, which uses a Proof of Work (PoW) blockchain, uses about 121 TWh, or 0.43% of all the electricity used in the world. It also uses 720,000 times more energy per transaction than the Visa payment system. When Ethereum switched to Proof of Stake (PoS) in 2022, it used 99.988% less energy, showing how much more efficient things can be.Generative AI models also use a lot of energy, especially when they are being trained and used to make predictions. For instance, it took about 9,450 MWh of energy to train GPT-4, and it took more than 500 MWh of energy to do inference work every day. Inference, which is always powered by user activity, is often more resource-intensive than the training process. The authors say that we need to learn more about and lessen the environmental effects of these technologies right away. Possible solutions include energy-efficient consensus mechanisms or giving AIs the ability to better optimise their own lifecycle. The report is meant to help businesses think about how to use technology in a way that is good for the environment as part of a better or more complete Scope 3 emissions strategy.
Dr.B.Swathi et al. (Wed,) studied this question.