One of many much less mentioned implications of the AI push is the quantity of vitality wanted to energy the general public the systemic infrastructure wanted to energy these elaborate methods.
Based on studies, TThe raining course of for OpenAI’s GPT-4, powered by roughly 25,000 NVIDIA A100 GPUs, requires as much as 62,000 MW-hours. That is equal to the vitality wants of 1,000 US households over 5 years.
And that is only one venture. Meta will embrace new AI superclusters 350,000 NVIDIA H100 GPUs, whereas X and Google, amongst others, are constructing large {hardware} initiatives to energy their very own fashions.
It is a enormous useful resource burden, which would require important funding to ease.
And it’ll have an effect on the surroundings.
To supply some perspective on this, the group at Visible Capitalist has put collectively an outline of Microsoft’s rising electrical energy wants because it continues to work with OpenAI on its initiatives.
