From recent purchase orders and analyst teardowns: ~$5.0M to buy a GB300 NVL72, ~$5.7M to deploy, $240–410K a year to run, and $2.6–2.9 per utilized GPU-hour at 80–90% utilization.
And to top it all off ... when Nvidia Feynman comes it'll require a completely different in terms of power and cooling and token outputs. Rendering much of this obsoleted.
And to top it all off ... when Nvidia Feynman comes it'll require a completely different in terms of power and cooling and token outputs. Rendering much of this obsoleted.
Extremely true!