Amazon Thinks the Future of Data Centers Depends on a Technical Problem It Just Solved¶
A WIRED article by Lauren Goode reports that AWS has achieved a breakthrough with Resilient Network Graphs (RNG) — a "quasi-random" data center network architecture that has been deployed since late 2024.
The Problem¶
The legacy fat-tree architecture has been the standard since the 1980s. AWS currently has over 20 million kilometers of fiber deployed across its data centers using this design. The architecture was becoming a bottleneck for scale, cost, and power.
Prior Attempts¶
- Jellyfish (2012, UIUC research) — a random graph topology that showed theoretical promise but was too chaotic to wire in practice.
- Google's OCS — optical circuit switching approach.
Amazon's Solution: ShuffleBox¶
Amazon developed ShuffleBox, a custom optical device that makes chaotic cabling manageable. This enabled the practical deployment of quasi-random network topologies at scale.
Results vs. Fat-Tree¶
| Metric | Improvement |
|---|---|
| Routers/switches | 69% fewer |
| Throughput | 33% higher |
| Power consumption | 40% lower |
| Operating costs | 27% lower |
Deployment Timeline¶
- Late 2024: First deployed in Dublin, Ireland
- 2025: Expanded to Germany and Spain
- Present: Now the standard in most new AWS data centers
Important Caveat¶
The RNG architecture is not designed for AI training workloads — it is strictly for core cloud infrastructure. This is a general-purpose data center networking advance, not a specialized AI fabric.
Expert validation came from Brighten Godfrey (UIUC, co-author of the original 2012 Jellyfish paper), who called the achievement "remarkable." The academic paper detailing the work is titled "RNG: Flat Datacenter Networks at Scale" (arXiv, April 2025).
Source: WIRED — Amazon Thinks the Future of Data Centers Depends on a Technical Problem It Just Solved