Arista Networks has taken the wraps off its 7060XE7 Series, a new portfolio of 1.6T networking platforms purpose-built to serve as the foundation for rack-scale AI infrastructure. The announcement marks a significant step in the evolution of data center networking, as AI workloads demand unprecedented bandwidth, low latency, and efficient power and thermal management.

The 7060XE7 family includes both fixed switch platforms and configurable rack-scale systems, targeting racks for vertical and horizontal AI workflows. All models will run Arista's Extensible Operating System (EOS), which has been enhanced with low-latency and intelligent packet buffering to handle the intense microbursts typical of AI communication and collective patterns. The company stated that these capabilities are critical for maintaining throughput and minimizing tail latency in large-scale distributed training environments.

Silicon Foundation and Ecosystem Collaboration

At the heart of the 7060XE7 Series is Broadcom's Tomahawk 6 silicon, which delivers 1.6T of switching capacity per chip. This marks a generational leap from previous 800G and 400G platforms, enabling higher port densities and reduced power per bit. Arista is also working closely with AMD on next-generation compute silicon and NICs to enable scale-out AI fabrics that can seamlessly integrate with the new switches. The collaboration aims to optimize end-to-end performance by aligning network and compute capabilities.

Strategically, the 7060XE7 Series signifies a transition for Arista from offering standalone high-performance switches to providing integrated rack-scale systems that address the extreme density, power, and thermal efficiency requirements of AI. The platforms support scale-up and scale-out AI fabrics using air, liquid, and hybrid-cooled technology, allowing customers to choose the cooling method that best fits their existing data center infrastructure.

Detailed Platform Configurations

The 7060XE7 family includes several distinct models tailored to different deployment scenarios:

• 7060XE7-64PS and 7060XE7-64PRS (4U Rack Switches): Available in Q4 of this year, these air-cooled systems support pluggable Integrated Heat Sink (IHS) and Riding Heat Sink (RHS) optics. IHS is designed for current air-cooled data centers, while RHS targets future liquid-cooled AI fabrics and extreme port density environments. These switches provide 64 ports of 1.6T, enabling high-bandwidth spine or leaf roles.
• 7060XE7-64PRS-RV3-L (2OU Liquid-Cooled Platform): This specialized model is optimized for high-density clusters and features 224G SerDes for increased signal integrity. It uses DC power from an ORv3 rack and contains no internal fans, integrating directly with liquid-cooled XPU servers to maximize power efficiency. Availability is slated for Q1 2027.
• 7060XE7-128PE (4U Air-Cooled Switch): Also arriving in Q1 2027, this device provides 128 ports of 800G in a 4RU air-cooled design, utilizing 100G SerDes. It offers deployment flexibility and backward compatibility for environments that need to mix different speeds.

All models run EOS as the primary network operating system, but the family also supports open-source alternatives such as Software for Open Networking in the Cloud (SONiC) and OpenSwitch, giving customers flexibility in software choice.

Multipath Reliable Connection (MRC) Protocol

One of the portfolio's most notable features is full support for the Open Compute Project's Multipath Reliable Connection (MRC). MRC is an RDMA-based transport protocol that allows a single reliable connection to simultaneously use many network paths over Ethernet. Arista's CTO Kenneth Duda and distinguished engineer Alan Judge explain that MRC enables endstation NICs to stripe traffic across multiple links and paths to the receiver, with out-of-order packets handled automatically. This approach responds to network congestion signals (ECN and packet trimming) by shifting load to the best-performing paths, avoiding links and paths that cannot reach the destination.

MRC continually monitors each path, steering around congestion, avoiding paths with link errors, and bypassing failed links entirely. According to Duda and Judge, this methodology has been proven in production to achieve very high fabric utilization with good load balancing, while interoperating seamlessly with scale-across and WAN networks that use standard dynamic routing protocols. The protocol is a critical enabler for AI workloads because it reduces the need for complex flow scheduling and improves overall network efficiency.

Software Enhancements for AI Networking

Beyond MRC, Arista has integrated a suite of software capabilities into EOS that are fundamental to AI networking. These include advanced load-balancing algorithms that can distribute traffic across multiple paths based on real-time congestion, fine-grained congestion management to prevent bufferbloat, comprehensive telemetry for visibility into fabric health, and diagnostics tools for rapid troubleshooting. The operating system also supports programmable automation via APIs, allowing customers to integrate the switches into existing orchestration frameworks.

By combining these software features with the hardware innovations of the 7060XE7 Series, Arista aims to deliver a unified platform that can handle the unique traffic patterns of AI training and inference. These patterns often involve all-to-all communication with synchronized barriers, which place intense demands on packet delivery and latency.

Market Context and Industry Validation

The new Arista family joins a growing ecosystem of vendors pushing into the 1.6T Ethernet space, including Cisco, Nvidia, and Celestica. The move reflects a broader industry consensus that Ethernet will play a dominant role in AI infrastructure, particularly as an alternative to proprietary interconnects like InfiniBand. Arista's early collaboration with cloud giants such as Microsoft Azure, Oracle Cloud Infrastructure, and Meta provides strong customer validation for the platform.

Sameh Boujelbene, vice president of data center switching and AI networks market research at Dell'Oro Group, highlighted several aspects of the 7060XE7 Series that stand out: sustainable bandwidth growth, improved power efficiency, and tighter integration between compute, optics, silicon, cooling, and network operating software. She also noted the strength of the ecosystem partnerships with AMD and Broadcom as a key differentiator.

The 1.6T transition is not just about speed—it represents a fundamental rethinking of how switches are architected for the AI era. With support for pluggable heat sinks, liquid cooling, and high-density SerDes, the 7060XE7 Series is positioned to meet the needs of both current and future data centers. As AI workloads continue to scale, the ability to efficiently manage power and thermal loads will become as important as raw bandwidth.

Arista's focus on rack-scale systems rather than standalone switches also reflects a growing trend toward converged infrastructure in which compute, storage, and networking are tightly integrated. By offering configurable racks that include power distribution, cooling, and networking in a single chassis, Arista is helping customers reduce deployment complexity and accelerate time-to-value for AI projects.

Source: Network World News