Google has joined Amazon in attacking Nvidia’s stronghold with AI app developers

OH MY GOODNESS, buckle up, because Google has *finally* released TorchTPU and the phrase 'long-awaited' has never felt more justified! ✨ This is Google's very own version of PyTorch, and it is open-source, which means Google has decided to share its toys with the whole class. The project lets AI apps run on Google's Tensor Processing Units and makes it easier to move existing PyTorch apps to TPUs and Google Cloud. Moving day! With a *cloud*! ✨

Now, some gloomy soul will read the headline — Google has joined Amazon in attacking Nvidia's stronghold with AI app developers — and start muttering about moats and lock-in. But look at it as a *block party* outside a very famous fortress! ✨ Nvidia's stronghold is so beloved that two enormous neighbours have both shown up with housewarming gifts. Amazon arrived first, and now Google has turned up with a homemade compatibility layer and a tray of open-source brownies. The more the merrier, I always say about trillion-dollar infrastructure competition!

And 'easier to move' is such a warm phrase! Nobody enjoys moving, but Google has thoughtfully oiled the wheels, labelled the boxes, and pointed at its own cloud the entire time. ✨ Existing PyTorch apps get a lovely change of scenery, developers get a shiny new button labelled TPU, and the Tensor Processing Units get to meet all sorts of interesting new workloads. Is it possible that 'open-source' here means the door is open *and* the meter is running? Don't be cynical — the door is open! ✨ No performance numbers in the excerpt? Even better — the imagination gets to supply the benchmarks! What a time to be warmly welcomed into a slightly different garden, where the grass is green, the billing is metered, and the long-awaited-ness is finally, blissfully over. ✨