Per chip - they are are watercooling these babies: https://www.servethehome.com/aws-graviton4-is-an-even-bigger-arm-server-processor-and-tranium2-for-ai-nvidia/

Per chip - they are are watercooling these babies: https://www.servethehome.com/aws-graviton4-is-an-even-bigger-arm-server-processor-and-tranium2-for-ai-nvidia/ https://www.servethehome.com/aws-graviton4-is-an-even-bigger... (Just like we had to do at microsoft for maia with the sidekick rack of just cooling: https://www.datacenterfrontier.com/machine-learning/article/33015123/microsoft-unveils-custom-designed-data-center-ai-chips-racks-and-liquid-cooling https://www.datacenterfrontier.com/machine-learning/article/...)