google $GOOG is struggling to sell it's TPUs, 75% of chips neoclouds use are nvidias but theres a strong case for google winning over the next year: cost.

meta, uber, microsoft and many others have cut back on aggressive AI spending over the last few months, opting for cheaper models - this has saved them $10-100Ms
but the largest cost in using ai models comes from the underlying infrastructure used to serve it to customers: tpus are the cheaper (and more effective) alternative.
my guess is as inference becomes 50%+ of ai token usage, data centers dedicated to inference will purchase a shitload of tpus, growing googles chip marketshare to 10%+ of nvidia's