AWS Replaces Fat-Tree Networks

The main update is a modest reframing of AWS’s network rollout: the architecture is now described as already default in most new data centers, and the narrative adds that the reported performance gains are still largely AWS-sourced. A new executive attribution also appears, with Matt Rehder identified as the AWS leader behind the description.

AWS’s RNG story has moved from a new networking design in production to a broader rollout milestone: it is now reportedly the default architecture for most new AWS data centers by April 2026. The latest version also adds specific performance and efficiency figures, making the deployment look more established and operationally significant.

• RNG became the default architecture for most new AWS data centers by April 2026.
• AWS reports 69% fewer routers and switches than conventional designs.
• AWS reports up to 33% higher throughput and 40% lower network electricity use.
• First production deployment began near Dublin in late 2024.
• Broader rollout was reported by April 2026.

The main update is a clearer production-status picture: AWS’s RNG design is now described as already deployed and broadly standardized across most new data centers by 2026. The framing also tightens around operational networking benefits, emphasizing throughput, power, and equipment reduction rather than any AI-specific use case.

AWS is rolling out a new data center networking architecture called resilient network graphs, or RNG, that replaces conventional fat-tree designs with a quasi-random layout. The system combines the Spraypoint routing algorithm and ShuffleBox optical hardware to reduce cabling complexity, improve resilience, and lower power use. Amazon presents the design as a production infrastructure advance rather than an AI-specific system, with reported deployment across multiple regions and default use in many new AWS builds.