Arista’s 7060XE7 Series: 1.6T Switching for the AI Era
As artificial intelligence workloads continue to demand unprecedented levels of bandwidth, low latency, and efficient power usage, networking vendors are racing to deliver infrastructure that can keep pace. Arista Networks has taken a major step forward with the launch of its 7060XE7 Series, a new portfolio of 1.6T networking platforms specifically engineered for rack-scale AI deployments. The series represents a strategic shift from standalone high-performance switches to integrated, configurable rack-scale systems designed to handle the extreme density and thermal requirements of modern AI training and inference.
Portfolio Details and Configurations
The 7060XE7 family comprises several models tailored to different data center environments and cooling methods. The 7060XE7-64PS and 7060XE7-64PRS are 4U rack switches available in Q4 2026, supporting air-cooled operation with pluggable Integrated Heat Sink (IHS) and Riding Heat Sink (RHS) optics. IHS is suited for current air-cooled facilities, while RHS targets future liquid-cooled AI fabrics and extreme port density scenarios. For high-density clusters, the 7060XE7-64PRS-RV3-L is a 2OU liquid-cooled platform featuring 224G SerDes, designed to integrate with liquid-cooled XPU servers and maximize power efficiency. This model uses DC power from the ORv3 rack and contains no internal fans, making it ideal for environments where heat management is critical. It will be available in Q1 2027. Additionally, the 7060XE7-128PE offers 128 800G ports in an air-cooled 4RU design, utilizing 100G SerDes for deployments requiring backward compatibility and flexible configurations.
Silicon and Software Foundation
At the heart of the 7060XE7 Series is Broadcom’s Tomahawk 6 silicon, which provides the 1.6T bandwidth necessary to support massive data flows between AI accelerators. Arista is also collaborating with AMD on next-generation compute silicon and NICs to enable scale-out AI fabrics, signaling a broader ecosystem play. The switches run on Arista’s Extensible Operating System (EOS), which includes low-latency and intelligent packet buffering to manage the intense microbursts characteristic of AI communication patterns. EOS also supports open-source alternatives such as SONiC (Software for Open Networking in the Cloud) and OpenSwitch, giving operators flexibility in network operating system choice.
Multipath Reliable Connection (MRC) and Advanced Protocols
A standout feature of the new series is full support for the Open Compute Project’s Multipath Reliable Connection (MRC) protocol. MRC is an RDMA-based transport protocol that allows a single reliable connection to simultaneously use many network paths over Ethernet. This is critical for AI workloads that require high fabric utilization and load balancing. Arista’s implementation of MRC enables endstation NICs to stripe traffic across multiple links and paths, with automatic handling of out-of-order packets. The protocol responds to network congestion signals such as ECN (Explicit Congestion Notification) and packet trimming, shifting load to the best-performing paths and avoiding problematic links. According to Arista’s technical leaders, MRC has been proven in production to achieve very high fabric utilization while interoperating seamlessly with scale-across and WAN networks using standard dynamic routing protocols.
Beyond MRC, EOS includes a suite of features crucial for AI networking: advanced load balancing algorithms, congestion management, telemetry, and diagnostics. These capabilities are essential for maintaining performance in large-scale GPU clusters where traffic patterns are both bursty and highly synchronized.
Market Context and Competitive Landscape
The 7060XE7 Series enters a rapidly expanding market for 1.6T Ethernet solutions. Competitors such as Cisco, Nvidia, and Celestica have also announced products in this space, but Arista’s focus on rack-scale systems—integrating compute, optics, cooling, and software—differentiates its offering. Industry analysts have noted strong customer and ecosystem validation from major cloud providers including Microsoft Azure, Oracle Cloud Infrastructure, and Meta, as well as silicon partners AMD and Broadcom. The portfolio addresses the growing need for both scale-up (within a rack) and scale-out (across racks) AI fabrics, which require different bandwidth and latency characteristics.
The shift from standalone switches to rack-scale systems is driven by the physical challenges of AI infrastructure. Modern AI accelerators generate enormous heat, and traditional air-cooled data centers struggle to maintain optimal operating temperatures for high-density clusters. Arista’s inclusion of air, liquid, and hybrid-cooled options in the 7060XE7 Series allows customers to choose the best cooling method for their specific environment, whether it’s a greenfield liquid-cooled facility or a brownfield air-cooled data center.
Importance of 1.6T Ethernet for AI
The jump to 1.6T Ethernet is not just about raw speed. For AI workloads, especially large language model training and real-time inference, network bandwidth must keep pace with the exponential growth in model parameters and data sets. A single GPU cluster may contain thousands of accelerators, and the interconnect bandwidth between them is a key determinant of training time. 1.6T switches provide enough capacity to connect multiple NICs at 400G or 800G without oversubscription, reducing bottlenecks and improving overall throughput. Additionally, advanced features like MRC and intelligent buffering ensure that the network can handle the unique traffic patterns of AI, such as all-reduce and all-gather collective communications.
Arista’s timing is strategic: the AI networking market is projected to grow at a compound annual rate exceeding 40% over the next five years, driven by hyperscaler investments and enterprise adoption. By offering a comprehensive portfolio that covers both hardware and software, Arista positions itself as a one-stop shop for AI fabric deployment.
Looking Ahead
The availability timeline of the 7060XE7 Series—most models shipping in Q4 2026 and Q1 2027—aligns with expected deployment cycles for next-generation AI clusters. As the industry moves toward 1.6T as the standard for high-performance AI, Arista’s early entry gives it a competitive advantage in securing design wins with hyperscale and large enterprise customers. The company’s long-standing partnership with Broadcom on silicon and its open approach to software standards like SONiC and MRC further strengthen its value proposition in an ecosystem that prizes interoperability and flexibility.
With the 7060XE7 Series, Arista is not just releasing faster switches; it is redefining how AI infrastructure is designed—integrating networking, compute, cooling, and software into cohesive, rack-scale building blocks. This holistic approach is likely to become the blueprint for future AI data centers as demand for computational power shows no signs of slowing.
Source: Network World News