That's a ways off. Investments, even if realized to have been "wrong", will be rationalized to delay that reckoning. Goals will be "scaled back" (and, this pitched as a "wise investment decision")
Eventually, the markets will abandon the idea of AGI as "within their grasp" and will turn to more "specialized/directed" applications of general AI. The LLM solution will lose much of its hype and be relegated to its ORIGINAL goals instead of magically assuming that it can be a panacea.
As more leaf processing is done, you'll see specialized AIs deployed to handle specific problem spaces; solutions to which they are more suited and which can deliver measurable returns, in the short term.
[E.g., using AI to control irrigation schedules is a no-brainer... compare the canned schedule with the actual schedule reified by the AI and you can note the net gains afforded... within DAYS. And, with far less resources "consumed" by the AI! Or "hoopla".]