Amazon unveils RNG networking design, boosting data center efficiency by 33% and reducing energy use by 40%

Amazon Web Services has quietly done something the networking world has debated for years: it ripped out the standard playbook for connecting servers inside data centers and replaced it with something built on quasi-random math. The new architecture, called Randomized Network Graph, is now the default infrastructure handling the majority of AWS workloads.

The result, according to a research paper co-authored by AWS engineers, is cost savings between 9% and 45% compared to traditional fat-tree designs, with equal or better performance across different traffic patterns.

How RNG actually works

RNG uses quasi-random graphs to create a “flat” network where data can take many different paths between any two points, replacing the traditional hierarchical fat-tree topology.

The architecture relies on two key innovations. First, a custom distributed routing protocol called Spraypoint that figures out how to spread traffic efficiently across this randomized mesh. Second, passive optical devices called ShuffleBoxes that handle the physical connections without the power-hungry electronics that traditional switches require.

The research paper detailing the deployment appeared on arXiv in April 2026, marking what the authors describe as the first large-scale production deployment of expander-based network fabrics.

The energy math matters

AWS reported a global Power Usage Effectiveness of 1.15 across its 2024 data centers. PUE is the industry’s standard efficiency metric: a score of 1.0 would mean every watt goes directly to computing, while higher numbers reflect energy lost to cooling, lighting, and other overhead.

AWS announced at its re:Invent 2024 conference that mechanical energy use, primarily cooling systems, could drop by up to 46% during peak cooling periods. The company also highlighted 12% enhancements in compute power alongside support for high-density AI racks using flexible cooling solutions.

These improvements are designed to increase compute density without raising water usage per megawatt.

What this means for crypto infrastructure

A significant share of blockchain infrastructure runs on cloud providers. Validator nodes, RPC endpoints, indexing services, DeFi backends, NFT metadata storage: all of it sits on servers that AWS, Google Cloud, and Microsoft Azure compete to host.

The 9% to 45% cost reduction range is broad, but even the low end is significant at scale. No specific cryptocurrency tokens or protocols were associated with the RNG or efficiency upgrade announcements.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.