The increase has already been exponential for years. With the AI hype, this demand for energy, cooling and water has increased dramatically.
What is known, what is to be expected and how an upcoming crisis be avoided? Can we reuse the energy? At least partially? Are there other concepts of integrating data centers into buildings and cities? Do we have non technical patterns driving the resource exhaustion?
The AI hype has increased the demand dramatically. The existing GPU based computing paradigm cuts hard into the standard design of data centers and demands other ways of cooling. Does the approach of modeling neurons really need floating point numbers? Which alternatives could be found?
This is an update of Thomas' previous talks at the #cccamp23 Camp[1] and at the Bits und Bäume conference [2].
[1] https://media.ccc.de/v/camp2023-57070-energy_consumption_of_data_centers [2] https://media.ccc.de/v/bitsundbaeume-19844-datenschutz-sparsamkeit-und-resourcenverbrauch-am-beispiel-einer-terminbuchungsanwendung