I think we're in a bubble.
That said: In principle, if the tech did what the strongest proponents forecast it would do, it would change "the economy" in a manner for which we don't even have a suitable metaphor, let alone soap and locomotives.
Now, I do not believe the AI on the horizon at the moment[0] will do any of that. The current architectures are making up for being too stupid to live[1] by doing its signal processing faster than anything alive by the same degree to which jogging is faster than continental drift[2].
As for "minimum" needed for this to not be a bubble, it needs to provide economic value in the order of 0.1 trillion USD per year, but specifically in a way that the people currently investing the money can actually capture that as profit.
The first part of that, providing economic value, I can believe: Being half-arsed with basic software development, being first line customer support, supplying tourist-grade translation services, etc. I can easily belive this adds up to a single percentage point of the world GDP, 10x what I think it needs to be.
Capturing any significant fraction of that, though? Nah. I think this will be like spreadsheets (Microsoft may be able to charge for Office, but Google Docs and LibreOffice are free) or Wikipedia. Takes an actual business plan to turn any of this into income streams, and there's too many people all competing for the same space, so any profit margin is going to be ground to nothing until any/all of them actually differentiate themselves properly.
[0] Not that this says very much given how fast the space is moving; it's quite plausible a better architecture has already been found but it isn't famous yet and nobody's scaled it up enough to make it look shiny and interesting, after all Transformers took years before anyone cared about them.
[1] Literally: no organic brain could get away with being this slow vs. number of examples needed to learn anything
[2] Also literally