How flat is replacing fat in AWS data center networks
AWS just rewired its data centers. Here's why that matters for AI compute.
Why it matters
AWS is deploying novel network topologies (flat vs fat-tree) and passive optical components (ShuffleBoxes) to improve data center efficiency—a critical infrastructure optimization for AI workloads that demand massive inter-node bandwidth and low latency.
The key facts
10 to knowAWS deploying 'flat' network topologies to replace traditional 'fat-tree' architectures
New passive optical components called 'ShuffleBoxes' enabling efficiency gains
'Quasi-random' network topology approach reduces complexity while maintaining performance
Infrastructure optimization directly impacts AI training and inference cost-per-compute
Published on Amazon Science blog—peer-reviewed AWS research
AWS deploying 'quasi-random' network topologies in data centers
New passive optical components called 'ShuffleBoxes' enabling flat network efficiency
Flat networks now comparable to traditional fat-tree networks in practical deployment
Infrastructure optimization focused on data center networking efficiency
Published May 2026 on Amazon Science (credible source)
Go to the source
Amazon Scienceamazon.science
Publisher excerpt: “Quasi-random” network topologies and new passive optical components called ShuffleBoxes make more-efficient flat networks as practical as traditional “fat-tree” networks.