My suspicion is that there's a there there, but it doesn't align with the predictions. This is supported by the tension between AI doom articles and the leading models experiencing diminishing performance gains while remaining error-prone. This is to speak nothing of the apparent LLM convergence limit of a ketamine-addled junior developer. Which is a boundary the models seem destined to approach indefinitely without ever breaching.

The "bust" in this scenario would hit the valuations (P/E ratio) of both the labs and their enterprise customers, and AI businesses dependant on exponential cost/performance growth curves with the models. The correction would shake the dummies (poorly capitalized or scoped businesses) out of the tree, leaving only the viable business and pricing models still standing.

That's my personal prediction as of writing.