The fundamental difficulty with “AI as a Service” is that it’s a solution to a problem almost nobody has.
Machine learning problems that require more hardware than the typical modern consumer laptop do exist, but they are exceedingly rare.
As alluded to in the article, the real problem with AI isn’t a lack of processing power; it’s the lack of expertise required to frame the problem such that it is amenable to being solved by a supervised learning algorithm.
It’s akin to navigating a minefield, and there are too many AI “experts” who don’t know what they don’t know about this due to limited experience “in the trenches.” Unfortunately, watching a few Andrew Ng lectures on YouTube isn’t sufficient.
But people would much rather build “AI as a Service” businesses because products are much easier to scale than expertise. Unfortunately for them, this is irrelevant if nobody needs your product.
(For background, I built my first neural network 24 years ago at the age of 16, and have been “in the trenches” ever since, across a wide variety of industries).