GPU capacity bust more likely than a bank run, argues AI compute cycle thread
bookwormengr · x · 2026-10-28
In the synthesis of a 16-part thread, RamaswmySridhar maps lessons from other industries onto AI infrastructure: banks teach funding contagion, commodity merchants teach spread risk, airlines teach utilization, utilities teach offtake finance, and CSPs teach pooling. He says the signals to watch are paid GPU usage, price per unit, and firm obligations — and concludes a capacity bust is more likely than a bank run.
More from Infra
- Celestica guides 2027 revenue to $40B, up 90%, with margins still expanding — TiernanRayTech · 2026-10-28
- RTX 5090 runs a 33B Q4 model at ~86 tokens/s: buy memory bandwidth, not TFLOPS — AIFlow_ML · 2026-10-28
- Model response speed is mostly memory bandwidth, not compute — and Rubin regressed — AIFlow_ML · 2026-10-28
- NVIDIA Vera Rubin NVL72 page goes live: 72 Rubin GPUs, rack-scale agentic AI supercomputer — AIFlow_ML · 2026-10-28
- SK Hynix scraps US IPO plans for NAND subsidiary Solidigm, WSJ reports — Beth_Kindig · 2026-10-28
- NVIDIA quietly cut Vera Rubin's HBM4 bandwidth 13% per GPU — the spec that matters for inference — AIFlow_ML · 2026-10-28