AWS has a cost-cutting networking upgrade. It's barely talking about it.

Amazon Web Services has rebuilt the plumbing inside its data centers, and the change should make cloud computing cheaper to run. The company is not telling anyone. That gap between a real cost-cutting upgrade and the decision not to promote it is the most useful thing to know about AWS this month.

The technology is called Resilient Network Graphs, or RNG, and it is now the default fabric in most new AWS data-center builds globally, according to the company's own Amazon Science blog post announcing the rollout. The post is short, technical, and buried under AWS's usual parade of product launches. Most customers will never see a clearer explanation than this one.

The numbers AWS is willing to claim are unusually large for a networking refresh. The company says RNG uses 69% fewer routers than the fat-tree designs it replaces, delivers up to 33% more throughput, and is projected to cut network-equipment electricity consumption by 40%. Tom's Hardware confirmed the rollout and the headline figures. AWS positions RNG, documented in the preprint paper arXiv 2604.15261, as the first flat network to scale to a full data-center size.

The mechanism is what makes those numbers plausible. A fat-tree design, the standard for the last decade, builds the network in rigid layers, so losing a router in one layer can take down a slice of the region. RNG spreads connections quasi-randomly across the fabric and uses a passive optical device called a ShuffleBox to make the cabling physically practical. A routing algorithm called Spraypoint then opens up nearly twice as many independent paths between any two servers. The result, in AWS's own measurement, is that losing 1% of the routers costs about 1% of capacity rather than a regional outage.

The interesting question is not whether the fabric works. It is why AWS is not pricing it in. Cloud networking is not a market where competitors can match a 40% power saving overnight, and customers cannot observe which topology is carrying their traffic. That removes the two pressures that usually turn an infrastructure efficiency into a price cut: competitive pressure to undercut rivals, and customer pressure to audit the bill. AWS's cloud rivals are not shipping comparable flat fabrics. Most customers do not have a tool that tells them which routers their traffic is crossing. The savings, if they exist at scale, become margin by default.

The Register's coverage of the rollout flagged the silence explicitly, calling out that AWS is treating a 40% power claim as a footnote rather than a sales argument. The outlet's earlier piece on the internal "Penrose" codename referenced billions of dollars in hardware savings and CO2 reductions, then walked past the implications. AWS is not the only hyperscaler with a quiet networking experiment in the wings, but it is the one that has already moved the new fabric into the default build path.

For a buyer the only useful move is to ask the question out loud. The honest test for any AWS account team is whether Resilient Network Graphs is already running in the regions a workload touches, and whether the savings show up anywhere in the bill. If the answer is no, the rollout is just margin dressed up as engineering. If the answer is yes, the question becomes why no one is allowed to know.

Watch the next AWS re:Invent keynote. A 40% power claim buried in a blog post is not a roadmap item. A stage announcement would be. The rollout now lives in a paragraph at the bottom of an Amazon Science post, with no roadmap, no customer-facing name, and no price line.