Post

I also wonder if the shift by makers of memory and other hardware away from supplying the consumer market towards AI datacenters will lead to a "pandemic toilet paper" supply chain issue when the bubble does pop, where they don't have product where the demand is, and a glut where it dried up

I used to use a workstation graphics card for 3d rendering before I moved to just doing it all with 24 threads of CPU. The CUDA cores on these cards are useful for such tasks. But I suspect that most of them will wind up in landfills before they ever get used extensively.

Ed, thank you for your coverage. Re the desperate need for more revenue esp by OpenAI - did you mention advertising? It seems inevitable that the model builders will find a way to support paid content placement in LLM outputs. I assume this is the offramp they're gunning for, Evel Knievel style.

They are trying, the issue is the LLM stochastic unreliability becomes a problem for advertisers. You can get ads showing up in horrific contexts that will make advertisers think twice. They will probably manage to juice some revenue out of it but I don't think it will save them

Fully agree that advertising-in-LLMs is a huge gamble. But I think we've seen that SamA needs the money and doesn't have other options with large enough payouts to cover his debts. I'm not saying his odds seem good. Nor that I'll like the results.

The AI bubble is slowly turning cash-rich, margin-heavy hyperscalers into something far worse. Far from being the asset-light, high-margin software companies they once were, they’re now bloated, filled with tens of billions of dollars in GPUs and data centers. www.wheresyoured.at/premium-timebomb/

This all has a bit of a vibe of the whole AJP Taylor "WWI happened on train timetables" about it, like everything is a predetermined catastrophe, but no one can do anything about it...