As the history rhymes, comparably we're at the mainframe era where limits of the 1st reach of fruitful capabilities could have been possible only with a big investment. Then came PC's and client compute. So, when we achieve mass RAM production and advance few more silicon nodes all within next couple of years, expect next phase of AI on the client with locally running AI (already some are doing it, but at a still prohibitive cost >$5K minimum upward to $20K).

Factor in inflation (which is unusually high and looks to be continuing until productivity effects materialize), so there could be market pricing adjustments/corrections (maybe short crash too, with repricing), although it is an interesting race dynamic where AI is progressing >2x than any previous cycle. In any case AI profits will materialize as they are already materializing at fastest pace than any tech cycle. There could be other re-pricing moments such as amortization/ROI of the AI datacenter cloud-s**. Near-mid term we'll overbuild power and memory production capacity, both harvestable in continuation and will drop price of KWh and GB, while cloud/datacenter buildings will not have such continuation.

So, it is actually easy to see >$30T TAM by ~2030 - buckle up, as that would shake many things and who survives might get to either paradise or hell.