Cerebras WSE-3 die is ~57x an H100 die, so TDDB failure odds could make the chips depreciate fast
teortaxesTex · x · 2026-10-29
A circulating technical analysis argues Cerebras' wafer-scale engine can indeed solve the yield problem via redundant interconnects and routing, but the real uncertainty is how long the wafer lasts in operation. All the extra BEOL wiring needed to route around fab defects greatly increases the effective area susceptible to TDDB (time-dependent dielectric breakdown) — a weakest-link failure mode whose odds grow with dielectric area, and a WSE-3 is roughly 57x an H100 die. Shorts from TDDB are more dangerous than ordinary faults, so the chips could depreciate extremely quickly unless field shorts can be mapped and routed around like fab defects.
More from Infra
- Agents push compute back to CPUs: $221B server CPU market seen by 2030 — BenBajarin · 2026-10-29
- AI server revenue forecast to hit $920B in 2027, pushing market past $1T — Beth_Kindig · 2026-10-29
- Oracle deploys Google TPUs for the first time and scales Vera Rubin NVL72 clusters to multi-gigawatt scale with Nvidia — Beth_Kindig · 2026-10-29
- Google Cloud expected to post 95.1% Q3 growth, outpacing AWS and Azure — Beth_Kindig · 2026-10-29
- lium rents NVIDIA B300s at $4/GPU-hour, H200 prices for Blackwell Ultra — bittingthembits · 2026-10-29
- Microsoft Expected to Generate $8B+ Cash Flow in Q3 While Meta, Google, Amazon Burn AI Capex — Beth_Kindig · 2026-10-29