Yes, it boils down to using either a single machine with multiple TPUs (and therefore doing things "the easy way" and relatively quickly) or having to use 128 G

Yes, it boils down to using either a single machine with multiple TPUs (and therefore doing things "the easy way" and relatively quickly) or having to use 128 GPUs (up to 8 per machine) and working with a single sample per GPU, really, really slowly. Given that a single model often requires dozens, if not hundreds of training runs to figure out hyperparameters that achieve a SOTA result, this means that with TPU you can do this, and with GPU you aren't even going to bother training something this big because it will take forever, and someone with a TPU will figure out something better by the time you're done. Which is what happened in this case, it looks like.