> OpenAI staff reportedly attributed some of Codex’s performance limitations to Nvidia’s GPU-based hardware.

They should design their own hardware, then. Somehow the other companies seem to be able to produce fast-enough models.