More special purpose AI chips being put into practice, this time for inference (whereas TPUs seem mostly for training?). Can't bode well for NVidia unless they have some specialized chip designs being cooked up. https://openai.com/index/introducing-gpt-5-3-codex-spark/