[qnleigh]

Last I heard, Cerebras chips were entire wafers and would be extremely expensive. How could OpenAI possibly have enough of these to serve a popular model at scale?