@2e2 Cintrini (as usual) has a nice take on the current state of things.
Overall, they are suggesting a correction and maybe a steep one, but that there is little data to support an AI bust.
The basic issue, which they describe in various ways, is that capex growth rate is declining, so all the momentum of that trade is lost.
My read of their comments is to not expect any kind of dramatic scaling back by the hyperscalers, but rather a bit of capex trimming to fit what can actually be completed.
As you mentioned, there is a problem on the revenue side and plenty of reason to expect the bulk of companies to adopt very slowly, but that is mostly an issue for ANT and OAI not the hyperscalers or their suppliers. The different AI sectors are operating under different economics and timelines.
NVDA could be the source of a steeper correction, as they will now almost certainly stop selling to the companies which were really just China conduits. The Jenson formerly know as CEO was the big pusher of open source, and was likely taken down in order to stop China from creating so much deflation.
For anyone who still doesn't get it, the Chinese companies have been using amazing amounts of AI agents to strip mine the weights from the frontier labs, then sell tokens at prices intended to economically attack those same labs. This is how Chinese companies compete in every industry. IP theft is their business model.
Capitalism can not exist alongside an economic system which is built on stolen IP.
The only answer to this nation-scale theft is to sanction, ban, and tariff.