The obvious argument is that since we have not solved alignment, accelerating AGI is unsafe. I’d wager that is the general line of reasoning from Sutskever and the board.
“We are growing quickly enough and a GPU shortage gives everyone more time to catch up on safety research” seems like a logically consistent position to me.
If they really want to ensure that ensure that AI benefits all of humanity they have to keep it affordable to everyone.
Scaling out of expensive GPUs, and making the service cheap and therefore generally available to more humans sounds sensible.