When One Datacenter Is No Longer Enough
nextplatform.com Oct 6, 2026

When One Datacenter Is No Longer Enough

AI-summarised brief · reviewed before publication

Cisco’s Rakesh Chopra explains that training the world’s largest AI models now requires distributing workloads across multiple data centers, a shift that turns networking from simple transport into a core compute component. Traditional inter‑datacenter links cannot handle the synchronous, low‑loss GPU traffic needed; instead, Cisco’s Silicon One architecture and Intelligent Collective Networking use high‑speed coherent optics, buffering, and integrated silicon to coordinate bursts, reduce bottlenecks, and improve job completion. Power efficiency, security (MACsec/IPsec), and programmability are also critical as clusters grow.

💡 Why It Matters

  • · Distributed AI training forces datacenter operators to rethink network design, turning interconnects into performance‑critical, energy‑aware systems that directly influence model accuracy and deployment speed.