Comment by m_ke

I assumed google would lean into the efficiency stuff more and try to eat the easy 80% of workloads, winning market share on volume instead of frontier if they were not able to produce frontier level models.

They're very well equipped to be the volume discount store of inference.