This sounds like it makes sense, but falls apart when you start to scratch the surface. ML practitioners can - by and large - build affordable GPU rigs. That’s

This sounds like it makes sense, but falls apart when you start to scratch the surface. ML practitioners can - by and large - build affordable GPU rigs. That’s why Google is heavily pushing cloud TPUs instead - they significantly outperform GPUs and you can’t buy them (retail or wholesale). You might get some spillover but I can’t see them making an investment as large as Stadia purely to soak up extra ML GPU capacity. If anything, it would be the other way around.