Amazon said it has solved one of the most important problems standing between today's data centers and the next generation of cloud computing: how to move massive amounts of data more efficiently.
According to a new Wired report, Amazon has been quietly deploying a new data center networking architecture called Resilient Network Graphs, or RNG, across its cloud infrastructure since late 2024.
The company says the design has dramatically improved the movement of information within Amazon Web Services data centers, a critical challenge as artificial intelligence and cloud storage place growing pressure on the physical systems that underpin the internet.
Brighten Godfrey, a computer science professor at the University of Illinois Urbana-Champaign and an expert in networking, told the outlet the innovation is "remarkable." Traditional data centers often rely on a "fat tree" network design, a hierarchical structure that routes data through layers of switches and routers.
Amazon's new approach flattens that structure using quasi-random connections between routers. In practice, that means data can travel through more possible paths instead of being forced through predictable chokepoints. Amazon says the result is substantial. The company claims RNG uses 69% fewer routers, delivers up to 33% better throughput, and could reduce electricity consumption from network equipment by 40%.
Matt Rehder, vice president of AWS Network Engineering, told Wired that "By essentially flattening the network, we eliminated the bottlenecks that come with traditional networking designs."
AWS also said the architecture became the default for most new data center builds globally by April 2026. The key to making the system work is a device Amazon calls the ShuffleBox, a passive optical component that helps organize complex fiber connections. Random networks have long interested researchers because they can be efficient and resilient, but they are notoriously difficult to cable and operate at real-world scale.
Amazon says ShuffleBox makes the quasi-random design manageable for technicians, allowing new racks to be connected without rewiring entire sections of a data center. The company also developed a routing system called Spraypoint to move traffic across the new network.
That matters because a flatter, less predictable layout requires smarter routing. In a traditional hierarchy, the structure itself helps determine where packets should go. In a quasi-random network, software has to do more of the thinking.
Amazon's announcement comes as Big Tech faces mounting scrutiny over the rapid expansion of data centers, especially their demand for land, water, and electricity. Amazon has already committed billions of dollars to new data center projects in the United States, including major investments in Pennsylvania and Ohio, while also pursuing nuclear energy partnerships to help meet future power needs.
The timing is not accidental: AI has turned infrastructure into a competitive battleground. Companies that can move data faster and train models more efficiently may reduce costs and improve reliability, while also easing some pressure on energy grids. Amazon's pitch is that RNG does not just make data centers faster, it makes them more sustainable and cheaper to build.
Still, the claims will be watched closely as Amazon's numbers come from its own testing and public research, and the full industry impact will depend on how the architecture performs as it spreads across more regions and heavier workloads.
