Amazon has publicly unveiled its Resilient Network Graphs (RNG) data center networking architecture, which the company...

Amazon has publicly unveiled its Resilient Network Graphs (RNG) data center networking architecture, which the company claims delivers up to 33% higher throughput while reducing network power consumption by 40% and using 69% fewer networking devices than traditional fat-tree topologies. The design, based on random graph theory, has been secretly deployed since 2024 and is now the default network for most AWS workloads. Amazon says RNG can reduce infrastructure costs by up to 45%, potentially saving billions across its global cloud footprint. The architecture was first deployed in Dublin in 2024 and has since expanded to Germany and Spain, and is being rolled out in new data centers. This innovation directly improves AWS's cost structure and operational efficiency, strengthening its competitive position against Microsoft Azure and Google Cloud. For semiconductor suppliers, the reduced reliance on networking hardware could impact demand for switches and related chips from companies like Marvell and Broadcom, though the overall boost to AWS capacity may increase demand for compute and memory components. The announcement was reported by Tom's Hardware and corroborated by Amazon's own blog on LLM observability (though the blog focuses on SageMaker, not RNG). The company's IR page confirms no additional filings. The RNG launch is a clear positive for Amazon's AI infrastructure capabilities and cost leadership.

Sources:

• Tom's Hardware
• AWS Blog on SageMaker Observability (indirect corroboration of AWS's focus on infrastructure efficiency)
• CNBC article on Snowflake deal (additional context on AWS AI demand)