Connectivity Revolution or Evolution Inside Data Centers?

· Source: Big Data & AI News - EE Times · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cloud Computing & IT Infrastructure, Emerging Technologies & Innovation · Depth: Advanced, medium

Summary

AI workloads are fundamentally transforming intra-data center connectivity, making networks as critical as compute. This shift blends revolutionary demands with optical evolution. Copper cables are being replaced by optical fiber for high-performance connections beyond 5 meters, with speeds advancing from 100 Gbps to 400 Gbps, 800 Gbps, and emerging 1.6 Tbps. This supports massive scale-out AI clusters, requiring 3-6 optical transceivers per GPU, totaling over a million short-reach units for hundreds of thousands of GPUs. LightCounting forecasts a doubling of Ethernet optical transceiver and co-packaged optics sales in five years. Current optical module designs (FRO, LRO, LPO) balance performance, power, and latency. Future innovations like Extra-dense Pluggable Optics (XPO), introduced in early 2026 for 12.8 Tbps, Near-Packaged Optics (NPO), and Co-Packaged Optics (CPO) aim to integrate optics closer to compute, enhancing efficiency and density.

Key takeaway

For AI Architects designing next-generation data centers, your network strategy must prioritize optical fiber and scale-out architectures. You should evaluate current options like LRO and LPO for efficiency in short-reach links, while actively planning for emerging technologies such as XPO, NPO, and CPO. Integrating optics closer to compute will be crucial for achieving the required density, power efficiency, and ultra-low latency demanded by future AI workloads.

Key insights

AI workloads necessitate a revolutionary shift to optical, scale-out intra-data center networks, integrating optics closer to compute.

Principles

In practice

Topics

Best for: Investor, CTO, VP of Engineering/Data, AI Architect, AI Hardware Engineer, MLOps Engineer

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Big Data & AI News - EE Times.