I think this is really great perspective.
So if the backlog truly is just a wait for compute to be built...how exactly is that backlog going to be cleared? Where is the electricity going to come from to power these data centers? What is going to be done about the massive backlash against their construction in the first place? Do we think that using natural gas to power these data centers is scalable?
In the meantime, anyone who makes heavy use of the AI models (myself included -- like 75% of my job is interacting with an LLM now) knows that the current state of the technology is enough to replace cheap, junior employees or other low paid employees that do repetitive tasks. They can't understand complex problems and they never will because the context window required to do complex work (and corresponding energy requirements) are not feasible, at least not until we figure out quantum computing or build them in space next to some device that harnesses the energy of the sun.
And since it's mostly cheap employees being replaced, if the technology gets any more expensive than it currently is, companies will just go back to hiring more headcount since there are tons of laid off people out there who will gladly take a low paying job to get back in the workforce.
So all that says to me that like the dot com bubble, we're due for a massive correction to account for the realistic scale and timeline of the adaptation of this new technology. And all of the circular financing done to fund the capex buildout means that this could get really painful if the ROI is not there by X date. Yes, the ROI might come eventually, but if the promises are delayed to the 2030s when we have the compute, that will still be a big problem for the market.
So, similar to the dot com bubble, the technology will be transformational in many ways, but this massive gold rush is going to collapse hard.