Comment by nelox

> * Valuation of the sp500, the hyperscalers and Nvidia is (mostly) reasonable based on earnings

That is a hell of a statement to make (their earnings are mostly negative, after all, except nvidia). Would require exceptional evidence, which doesn't seem to be there.

> * Build out of infrastructure is demand-driven, hyperscalers are not building just for future demand that would not materialize

This does not reconcile with the large amount of empty datacenters and GPUs which have not been installed: https://www.wheresyoured.at/ais-economics-dont-make-sense-ad...

> * OpenAI, anthropic & co can be overvalued but that does not mean there's a systemic bubble

OK? It could also mean there is.

> I think this underestimates contagion effects and the fact that demand appears to be subsidized and may disappear quickly, but it's just MHO.

Even with subsidized demand Microsoft still ended up cancelling over a gigawatt(!) of planned datacenters already back in 2024. But yeah, their arguments are missing a lot.

I think the earnings are suspect and exaggerated. Hardware manufacturers are making real money now, but there is a big question if any of these AI companies can deliver profitability to match their current valuation let alone future valuations when they go public.

Hyperscalers are in big trouble if the build out suddenly stalls. Even Nvidia and Micron are going to see their value significantly trimmed if it looks like growth is stalling. With such concentration at the top of the S&P among tech companies and with SpaceX, Anthropic, and Open AI, three companies that probably burn a combined 50+ billion a year. The whole stock market will be a tinderbox.

The whole thing is so private capital can get their exit. Default rates of private capital are already at 6%. Banks are exposed so they are on board with the fraud.

Oh yes, this time it will be different, of course. (Like the last time.)