“AI chips from Groq and Google's TPUs are designed to have deterministic latency.”