Arista Networks has announced the 7060XE7 Series, a new family of networking platforms that deliver 1.6T throughput per rack and are purpose-built for artificial intelligence infrastructure. The portfolio includes fixed switch configurations and modular rack-scale systems that can be configured for vertical or horizontal AI workflows, addressing the growing demand for high-bandwidth, low-latency networks in AI data centers.

The 7060XE7 family is built on Broadcom Tomahawk 6 silicon, a 51.2T switch chip that enables 64 ports of 800GbE or 128 ports of 400GbE in a single device. Arista is also collaborating with AMD on next-generation compute silicon and network interface cards (NICs) to support scale-out AI fabrics. The platforms run Arista's Extensible Operating System (EOS), which includes advanced features for low-latency forwarding, intelligent packet buffering, and congestion management to handle the intense microbursts common in AI training and inference workloads.

Strategically, the 7060XE7 Series marks a shift 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 air, liquid, and hybrid cooling technologies, allowing customers to deploy scale-up and scale-out AI fabrics in diverse data center environments.

Specific configurations include the 7060XE7-64PS and 7060XE7-64PRS, both 4U air-cooled rack switches available in Q4 2026. These support pluggable integrated heat sink (IHS) and riding heat sink (RHS) optics, with IHS targeting current air-cooled data centers and RHS designed for future liquid-cooled AI fabrics. The 7060XE7-64PRS-RV3-L is a specialized 2OU liquid-cooled platform featuring 224G SerDes, DC power from the ORv3 rack, and no internal fans, integrating with liquid-cooled XPU servers for maximum power efficiency. It will be available in Q1 2027. The 7060XE7-128PE, also coming in Q1 2027, provides 128 800G ports in a 4RU air-cooled design using 100G SerDes, offering deployment flexibility and backward compatibility.

On the software side, EOS is the featured network operating system, but the family also supports open-source alternatives such as Software for Open Networking in the Cloud (SONiC) and OpenSwitch. Key software features include full support for the Open Compute Project's Multipath Reliable Connection (MRC), an RDMA-based transport protocol that enables a single reliable connection to simultaneously use many network paths over Ethernet. MRC allows endstation NICs to stripe traffic across multiple links and paths, handling out-of-order packets automatically and responding to congestion signals by shifting load to the best-performing paths while avoiding failed links. Arista notes that this approach achieves very high fabric utilization with good load balancing and interoperates seamlessly with scale-across and WAN networks using standard dynamic routing protocols.

Additional software capabilities include load balancing, congestion management, telemetry, and diagnostics, all of which are core to AI networking. The platforms are designed to support the extreme bandwidth densities required for large-scale AI training clusters, where thousands of GPUs or XPUs communicate via collective operations like all-reduce and all-to-all.

Arista's new family enters a competitive ecosystem of vendors tapping into the 1.6T Ethernet market, including Cisco, Nvidia, Celestica, and others. The 7060XE7 Series has garnered validation from major cloud providers such as Microsoft Azure, Oracle Cloud Infrastructure, and Meta, as well as silicon partners AMD and Broadcom. Industry analysts note that the portfolio signals where large-scale AI fabrics are headed: higher bandwidth, better power efficiency, and tighter integration between compute, optics, silicon, cooling, and network operating software.

Arista has a long history in data center networking, having been founded in 2004 by former Cisco executives. The company's EOS operating system, based on a single binary image across all platforms, has been widely adopted in cloud and enterprise data centers. With the 7060XE7 Series, Arista aims to leverage its experience in high-performance networking to capture a larger share of the AI infrastructure market, which is projected to grow significantly over the next several years.

AI workloads place unique demands on networks, including the need for low latency, high bandwidth, and the ability to handle congestion from synchronized communication patterns. Traditional Ethernet networks have struggled with these requirements, leading to the adoption of specialized technologies like RDMA over Converged Ethernet (RoCE) and InfiniBand. However, Arista's implementation of MRC and EOS features aims to make Ethernet a viable alternative for AI fabrics, potentially reducing costs and increasing flexibility for customers.

The 7060XE7 Series also supports advanced cooling options, which is crucial as AI clusters generate significant heat. Air-cooled platforms can be deployed in existing data centers, while liquid-cooled systems are designed for the highest-density clusters. The liquid-cooled 7060XE7-64PRS-RV3-L uses no internal fans, relying on facility-level cooling infrastructure to improve power and thermal efficiency. This aligns with industry trends toward liquid cooling for high-performance computing and AI.

In terms of optics, the platforms support pluggable IHS and RHS optics, allowing customers to choose between air-cooled and liquid-cooled transceivers as needed. Arista is working with optical module vendors to ensure availability of compatible optics for the 1.6T speeds. The 224G SerDes used in the liquid-cooled platform enable higher lane speeds, reducing the number of optical lanes required for a given bandwidth and improving overall system efficiency.

Broadcom's Tomahawk 6 silicon, which powers the 7060XE7 family, offers 51.2T of switching capacity with 800GbE ports. This chip enables Arista to deliver 64 800GbE ports or 128 400GbE ports in a single switch, providing the density needed for AI training clusters. Tomahawk 6 also includes advanced features for telemetry, congestion management, and load balancing, which are leveraged by Arista's EOS.

Arista's collaboration with AMD is focused on optimizing the integration between Arista switches and AMD's compute and networking silicon, including AMD's upcoming AI accelerators and Pensando NICs. This partnership aims to deliver end-to-end solutions for AI infrastructure, from the compute node to the network fabric. By working together, the two companies can ensure that their products interoperate seamlessly and provide the performance needed for large-scale AI workloads.

The 7060XE7 Series is expected to be available in phases, with initial air-cooled systems shipping in Q4 2026 and liquid-cooled and high-density systems following in Q1 2027. Arista is currently sampling the platforms with select customers and expects to begin volume shipments later this year. The company plans to demonstrate the 7060XE7 portfolio at upcoming industry events, including OCP Global Summit and OFC.

As AI continues to drive demand for faster, more efficient networks, Arista's 7060XE7 Series positions the company to compete in the rapidly evolving AI networking market. With its combination of high bandwidth, flexible cooling, advanced software features, and ecosystem support, the portfolio offers a comprehensive solution for organizations building next-generation AI infrastructure.

Source: Network World News