Raj Yavatkar, Juniper Networks | AI Infrastructure Silicon Valley - Executive Series
summary
Claims a working 800 gig switches fabric plus intent-based Apstra automation that runs auto-tuning application to adjust buffer thresholds and congestion controls in real time. Says telemetry from GPUs, compute, storage and network is correlated for application-aware assurance, enabling automated root-cause diagnostics and remediation across 64–10,000 GPU clusters. Cites ACG research that on‑prem training can cost about half of cloud and argues ethernet is reaching parity with InfiniBand, with an industry consortium extending Ethernet for training and inference workloads.
AI-written summary · about
