Google acquired a company in 2007 which had expertise in multicore at the time multicore was really improving quickly. One of the acquired principal engineers wrote a manifesto saying that GPUs were a waste of time because they didn't solve problems like speeding up search (it's probably still around) or the other things Google was spending a lot of cycles on (ML training). It was a great example of a technical leader being completely wrong about one of the directions of the future and the technical solution to it. Now, I should say that multicore machines are still highly relevant and complement GPUs nicely- but the manifesto happened at a time to convince leadership to avoid making strategic deals with nvidia at a critical time.