Amazon has implemented a new data center network architecture that accelerates data transmission and reduces energy consumption, according to Wired.

The solution is based on a "quasi-random" topology. AWS claims to have been working on it since 2023, with implementation in its infrastructure beginning in late 2025. For this project, the company developed its own optical device called ShuffleBox, which automatically shuffles cable connections between routers.

Matt Reder, Vice President of AWS Network Engineering, stated that the team has managed to "flatten the network" and eliminate bottlenecks typical of traditional architectures. He noted that Amazon is the first to scale this approach to a real infrastructure level.

In April, the company released a research paper titled RNG: Flat Datacenter Networks at Scale. RNG stands for resilient network graphs.

According to Amazon, compared to traditional networks, the RNG architecture:

  • requires 69% fewer routers and switches;
  • increases bandwidth by 33%;
  • reduces energy consumption by 40%;
  • lowers operational costs by 27%.

The first deployment took place in Dublin in 2024, followed by rollouts in Germany and Spain. Currently, most new AWS data centers are being built using RNG.

Since the mid-1980s, data centers have primarily used the fat-tree topology—a multi-level hierarchy of switches. While reliable, it is rigid and requires complex cabling infrastructure. Amazon's network connects approximately 20 million kilometers of fiber optics.

The project was inspired by the Jellyfish concept proposed by researchers at the University of Illinois in 2012, which suggested a random graph topology as an alternative to fat-tree but faced routing and cabling challenges.

One of the paper's authors, Giacomo Bernardi, explained that the team initially tested a more regular scheme inspired by Penrose tiling. Simulations showed weak resilience and modest efficiency gains, leading engineers to switch to a quasi-random model.

AWS emphasized that the architecture was not specifically designed for generative AI, as model training patterns are too centralized for a random graph. The focus is on optimizing the underlying network infrastructure.

It is worth noting that the largest regional electricity transmission operator in the U.S., PJM Interconnection, reported that the data center boom has led to an additional $23.1 billion in costs just within its jurisdiction.