Presentation: Data Center Interconnect for the AI Era: Enabling Distributed AI Infrastructure#
Capacity Europe 2026 digest Program Core pick 12–15 October 2026
Nino Shaptoshvili, VP of Sales - PacketLight Networks Ltd
- Why: DCI is squarely in the PM's core territory — optical transport between data centres is becoming the fabric that binds distributed GPU clusters into one AI system. This is the one session in the program that directly addresses the DC/WAN interconnect problem for AI workloads.
- What: An argument that as AI infrastructure spreads across multiple sites, the network (not the compute) becomes the gating factor, and that high-performance DCI is what turns scattered GPU clusters, storage, and compute into a "unified AI fabric" (from abstract only).
- How: Optical transport delivering high bandwidth, deterministic low latency, and security for moving large datasets and synchronising AI workloads across sites — i.e. the classic coherent-optics DCI playbook applied to training/inference traffic patterns.
- Where applicable: Multi-site AI clusters, campus-scale and metro/regional DCI. Caveat: the speaker is a VP of Sales at a DCI vendor (PacketLight), so expect a product-adjacent framing rather than an operator case study; no scale, latency, or cost figures are published. Treat as a vendor's view of the problem, useful for the requirements it articulates (deterministic latency, bulk dataset movement) rather than for evidence.