Just look at tensorflow. It masquerades as enabling open source ai, while really pushing people to pricey, proprietary google cloud compute. How these ppl can keep a straight face, I have no idea.
I currently run a 120 dollar tensorflow device all locally that does real time object detection. I dont understand your criticism.
I'm talking about Google sequestering tensor cores and writing the distributed training code for g-cloud pods. Even your nice embedded inference device was designed with the hopes you'd grab some high performance compute to train something high quality. (These embedded devices are useless for training)
Fair enough, but with everything google, the free tier good enough for a prototype and personal use. If I need to scale it, there are other options. For me, I appreciate the option of this model, it makes high tech accessible to low capital ventures. And sometimes buying locked in services is the rational decision taking into account the costs of learning other platforms and maintaining a
vender unlocked platform.
I run tensorflow on my laptop for small jobs, and AWS for big ones or in production. Not sure what the "push" is supposed to be.