Amazon Web Services (AWS) has expanded its custom hardware stack with its own DWDM transponder specifically designed to scale for both metro and long-haul connections.

The hardware leverages the concept of dense wavelength division multiplexing (DWDM), which combines multiple encrypted signals on a single fiber. Instead of buying off-the-shelf networking gear that's built for general purposes, AWS’s custom DWDM transponders give the hyperscaler far greater control of its entire network, providing it with major bandwidth boosts.

The hyperscaler claims its custom transponders offer 73% more bandwidth than compared to its initial attempt at a transponder, while reducing power consumption by nearly 35%.

Matt Rehder, AWS’s VP for core networking, said that for customers, the hyperscaler’s use of its own souped-up transponders means “faster data transfer speeds, improved application performance, and more sustainable operations without increased costs.”

“Unlike traditional vendor solutions, our tailored approach allows for precise power management across the network,” Rehder wrote in a blog post. “The unified architecture eliminates redundant systems, while granular control over data rates ensures optimal efficiency for each connection. These advancements not only enhance network performance but also contribute significantly to AWS’s environmental commitments.”

AWS has a storied history of building its own data center hardware, including its Arm-based Graviton processors, the GPU-level Trainium line, custom NICs like the K2v6, and chip-to-chip interconnects such as the NeuronLink-v2.

The hyperscaler’s focus on doing its own thing extends to its wider networking stack, what with its own in-house developed multipathing fabric protocol: Elastic Fabric Adapter or EFA.

All of these disparate components come together to power AWS’ vast data center infrastructure – with its efforts on full display following the recent unveiling of Project Rainier, a colossal compute cluster being used by Anthropic, for which SDxCentral was given an exclusive look at its underlying network.

Its latest innovation then sees AWS push beyond the data center itself into the fiber optic backbone connecting its global infrastructure.

According to Rehder's blog post, development of the enhanced DWDM transponder began in January 2024, with systems fully deployed for metro networks by February 2025. A few months later, in July, the hyperscaler had its first long-haul connection spanning more than 932 miles (1,500 kilometers) operational.

“This represented a significant shift in how data transmission systems are built,” explained Omid Momtahan, senior network development manager of AWS network product development, who led the project. “After we successfully demonstrated and deployed the technology for metro connections, we proved that a simplified, customized approach could deliver superior performance, higher availability, and agility in operations and deployment.”

The finalized iteration is much smaller than more traditional transponders. Despite its size however, AWS engineers were able to at least double channel capacity while cutting power consumption. That enabled it to maintain the same total bandwidth with half the number of channels, some 32 instead of 64.

The system also offers flexible data transmission rates adjustable in 100-gigabit increments, allowing the hyperscaler to fine-tune performance for each network's unique characteristics rather than being locked into preset configurations.

Arguably most integral for AWS was the fact that it creates what Rehder described as a “unified operating environment across data centers, metro connections, and long-haul networks,” given that it uses the same hardware, operating system, and tools.

“This integration enhances reliability and enables granular security controls, real-time threat detection, and rapid updates without third-party risks,” Rehder wrote. “AWS support teams use a single toolset across all networking equipment, enabling faster issue resolution and more stable connections.

“This unified approach allows proactive problem identification before customer impact, resulting in more reliable service, reduced downtime, stronger security, and consistent connectivity.”

The hyperscaler’s post revealed that AWS has already extended deployment of the custom transponders to Europe, with eventual plans to use the technology for all new long-haul connections throughout its global infrastructure.

AWS’s dedication to its own hardware saw its engineers recently develop JBOK (just a bunch of K2v6 network interface cards) – an entire cabinet dedicated just for its own custom NICs.

The concept, named in a nod to JBOD (just a bunch of disks), came about as Nvidia’s NVL72 rack-scale system uses shorter 1U server trays, which meant the hyperscaler couldn’t then add its own network cards into the trays.

JBOK, then, is a completely separate cabinet just off to the side that’s full of Amazon’s custom K2v6 NICs. While it might sound inherently complex, taking up additional white space, AWS claims it works far better than its prior attempt at a workaround, in which it added extra server trays to fit its network cards, which resulted in more bugs than in simply using Nvidia’s native NVL72 design.