With all of that having been said and now out of the way, it is important to note that a Meta executive by the name of Sergey Edunov is rather optimistic about what the future holds. As the director of engineering for Generative AI at Meta, Edunov has keen insights about what kind of power consumption the industry might require, and he clarified that just two new nuclear plants will be more than enough to get the job done.
While speaking at a panel at the Digital Workers Forum, the lead engineer on Meta’s Llama 2 open source LLM stated that the energy produced by two nuclear power plants will provide adequate energy for a year’s worth of AI usage. When he was asked about whether or not the world has enough energy to keep the ball rolling, he mentioned that AI is not quite as energy intensive as it might be in the near to distant future.
One thing that bears mentioning is that Edunov was referring to how much energy will be necessary for inference. This is the process by which AI can be utilized in an application to answer queries and provide recommendations based on what users ask, and the existence of nuclear energy seems adequate based on some rough calculations he made.
It will be essential to answer the question of AI’s power requirements because of the fact that this is the sort of thing that could potentially end up allowing the industry to grow. According to Nvidia’s report, the one to two million GPUs it released will require 100,000 LLM tokens per person, and while that might seem like a lot it’s actually far more sustainable with all things having been considered and taken into account. Each GPU needs 700 watts for cooling and other processes, and nuclear energy might be able to cover that.
Read next: 97% Embrace AI — The 2023 Martech Revolution Unveiled