Chips & Compute

Arista ships four encrypted routing switches for distributed AI inferencing

Arista unveiled four Etherlink R4 routing switches with line-rate TunnelSec encryption on every port, targeting distributed AI inferencing across data centers, regional hubs and the enterprise edge.

Arista expands its routing options to support secure AI inferencing
Arista expands its routing options to support secure AI inferencingBrecht Corbeel / unsplash
By Rebecca Stone6 min read

Why it matters

  • Arista announced four new Etherlink R4 routing switches with line-rate TunnelSec encryption (MACsec, IPsec, VXLANsec, hardware-based SRv6) on every port.
  • The 7282R4 and 7280R4 routers deliver up to 25.6 Tbps of non-blocking bandwidth with 100GbE, 400GbE and 800GbE support; the 2RU 7282R4 adds dual Supervisor Engines.
  • The hardened 1RU 48-port 7020HR4 and 7020R4 compact platforms support 1G through 400G SFP/QSFP or 1G/10G RJ45 ports.
  • The 7020HR4 ships in early 2027; the other new R4 platforms are available now.
  • Executives Brendan Gibbs and Tyson Lamoreaux said EOS security updates can often be applied without a reboot, eliminating network downtime.

Arista has released four new Etherlink R4 high-performance routing switches built to secure and improve the performance and resiliency of connections serving AI compute and storage infrastructure, with line-rate encryption integrated into every port through its TunnelSec feature. The announcement signals where the networking market now believes the next phase of AI infrastructure spending sits: not in massive centralized training clusters, but in distributed inferencing spread across data centers, regional hubs and enterprise edges.

The four new switches target large-scale distributed data center, AI back-end, and service provider environments where operators need high route scale, deep packet buffering and 400G connectivity, according to a blog post co-authored by Brendan Gibbs, Arista vice president of product line management, and Tyson Lamoreaux, senior vice president, Cloud/AI. TunnelSec covers MACsec, IPsec, VXLANsec and hardware-based Segment Routing over IPv6. All of the new boxes run Arista's Extensible Operating System, known as EOS.

Why is Arista refocusing on inferencing?

The two executives framed the product launch as a response to a structural shift in how AI workloads are deployed. They wrote that the industry has moved through distinct phases of buildout, each of which pushed the network in a different direction.

"The initial phase of artificial intelligence (AI) buildouts concentrated compute nodes in massive, centralized clusters, but the need for greater scale has led operators to span training clusters across multiple data centers with AI scale-across solutions. Now, the operational reality of centralized inferencing has emerged as the new frontier requiring a distributed solution," Gibbs and Lamoreaux wrote.

They drew a sharp geographic distinction between the two workload types: "Training can tolerate being tucked away near remote power grids, while inferencing cannot. As AI transitions from static prompts to real-time interaction, intelligence must be distributed across front-end networks spanning distinct geographies, regional hubs, and enterprise perimeters."

That distinction carries commercial weight for Arista and its rivals. Training clusters were served by back-end fabrics connecting GPUs inside a single site. Real-time inference moves the traffic problem to the wide area, where latency, uptime and encryption across organizational boundaries become the binding constraints — a segment where routing, not just switching, decides who wins the socket.

What is in the new lineup?

The release centers on two hardened compact routing platforms and two high-availability routers:

  • 7020HR4 and 7020R4: hardened 1RU, 48-port compact routing platforms aimed at high-performance edge applications, distributed AI inference, secure data center leaf, service provider access routing and high-end desktop applications.
  • 7282R4 and 7280R4: high-availability routers offering up to 25.6 Tbps of non-blocking bandwidth, with support for high-density 100GbE, 400GbE and 800GbE networking.

The 7020R4 Series platforms share a common forwarding architecture within the R4 family, built around dynamic deep buffers that Arista says deliver deterministic, lossless forwarding performance and support scalable resources. Buyers can configure both compact platforms with a choice of 1G/10G/25G/50G/100G/400G SFP/QSFP ports or 1G/10G RJ45 ports.

The 2RU 7282R4 adds support for dual Supervisor Engines, which provide redundant control planes — the feature Arista positions as the box's answer to the availability demands of inference traffic that cannot tolerate a failed brain in the chassis.

Why does one routing stack matter for security?

Availability is only half of Arista's argument. The company is also pitching operational simplicity as a security control. As inference workloads spread across data centers and edge locations, a network interruption at a single critical aggregation or peering point can affect the entire application chain, Gibbs and Lamoreaux wrote. Distributed inference therefore raises the cost of every bug, vulnerability and maintenance window in the fabric.

Arista's answer is software homogeneity. Because every box in the R4 family runs EOS, customers can patch one codebase across an entire footprint rather than requalifying software variants box by box.

"With Arista, one routing stack spans your entire footprint. Vulnerability mitigation is centralized, consistent, and validated at scale. Because EOS is built on a single, clean state-streaming software architecture, security updates are streamlined, predictable, and rapidly deployable," the executives wrote. "Customers can maintain machine-speed defense without having to requalify software variants across their AI fabrics. Moreover, these updates can often be applied without the need for a reboot, thus eliminating any network downtime. That secure software foundation then becomes the springboard for distributed, large-scale inference."

The no-reboot patching claim matters for inference operators specifically: an AI training job can checkpoint and resume, but a real-time inference service serving live users cannot pause while a router reloads its software.

How do the new boxes fit the existing R4 family?

The R4 portfolio was already populated before this announcement, and the four additions extend it at both the edge and the aggregation layer. The existing family includes:

  • The 7020, a fixed 10G to 25G leaf edge router with 100G uplinks.
  • The 7280, a fixed 100G leaf or spine aggregation box with 800G uplinks.
  • The 7800, a high-end modular 800G spine router featuring what Arista calls 3.2Tbps HyperPort Ethernet capacity.

Arista positions the 7800R as the heart of very large networks — the central spine box in large-scale enterprise, cloud hyperscaler and service provider deployments, in AI Ethernet fabrics, in cloud data centers, and in campus and metro aggregation sites. The new 7282R4 and 7280R4 slot in beneath it as aggregation and high-availability tiers, while the 7020HR4 and 7020R4 extend the family outward to the edge where distributed inference actually meets traffic.

Gibbs has previously described the R4 portfolio as Arista's Swiss Army knife, noting that it offers a common set of silicon and software capabilities across a range of platform options. That commonality is the strategic point: a customer can deploy the same forwarding behavior, the same buffers and the same EOS image from a hardened edge box in a regional hub to a 3.2Tbps modular spine in a hyperscaler data center.

When can buyers get the gear?

Availability is split across the lineup. The 7020HR4 will not ship until early 2027, while the other new R4 platforms are available now, according to Arista. The gap gives competitors room to respond in the hardened edge segment before Arista's most inference-specific compact platform reaches the market.

The release lands as the industry's center of gravity shifts from building training clusters to operating inference at scale. Arista is betting that the winners in that phase will be the vendors whose routing stacks can encrypt every port at line rate, survive a control-plane failure without dropping the application chain, and patch without a reboot.

Source: networkworld.com

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

728 articles

Related articles

  1. Arista posts record 58.5% revenue growth on AI networking
  2. Atsign launches EdgeAI with identity-based kill switch
  3. Coalition plugs OpenAI, Anthropic, Gemini and Cursor into Wirespeed
  4. OpenAI open-sources MRC, a networking protocol for AI training clusters
  5. Enterprise AI Network Spending Surges Before Readiness Catches Up

« Previous articleNext article »