https://en.wikipedia.org/wiki/Jevons_paradox

I'm not going to speculate about what might be ahead in regards to Oracle's forecasting of data center demand, but regarding the idea of efficiency gains leading to lower demand, don't you think something like Jevons paradox might apply here?

People said the same thing for deepseek-r1, and nothing changed.

If you come up with a way to make the current generation of models 10x more efficient, then everyone just moves to train a 10x bigger model. There isn’t a size of model where the players are going to be satisfied at and not go 10x bigger. Not as long as scaling still pays off (and it does today).

Absolutely not; the trends have proven that people will just pay for the best quality they can get, and keep paying roughly the same money.

Every time a new model is released, people abandon the old, lower quality model (even when it’s priced less), and instead prefer to pay the same for a better model.

The same will happen with this.

If someone had to bet on an AI crash which I imagine would led to unused datacentres and cheap GPUs how would they invest their winnings to exploit these resources?

No. The gains in inference and training efficiency are going to be absorbed by frontier LLM labs being more willing to push more demanding and capable models to the end users, increase reasoning token budgets, etc.

For the last 2 years, despite all efficiency gains, I am literally watching characters appear on my screen, as if this was a hacker movie. Lately, I am also waiting for at least 60s for anything to appear at all.

If that happened at 10x the speed, I would still be slow in computer terms, and that increasingly matter, because I will not be the one reading the stuff – it will be other computers. I think looking back a few years from now, every single piece of silicon that is planned right will look like a laudable but laughable drop in the ocean.

The real quality demand needs is not there, so more processing is very probably needed, so efficiency gains may allow the extra processing.

(A string example read today of Real quality demand needs: the administration of Albania wants some sort of automated Cabinet Minister. Not just an impartial and incorruptible algorithm (what we normally try to do with deterministic computation): a "minister". Good luck with that.)