To be clear I'm only suggesting that hardware is a factor here, it's far from the only reason. The parent commenter corrected their comment that it was actually

To be clear I'm only suggesting that hardware is a factor here, it's far from the only reason. The parent commenter corrected their comment that it was actually Groq not Grok that they were thinking of, and I believe they are correct about that as Groq is doing something similar to TPUs to accelerate inference.