Arista Networks has taken the wraps off its 7060XE7 Series, a new portfolio of 1.6T networking platforms designed to provide the foundation for rack-scale AI infrastructure. The 7060XE7 family features fixed switch platforms and configurable rack-scale systems, targeting racks for vertical and horizontal AI workflows. All will run Arista's Extensible Operating System (EOS), which includes low-latency and intelligent packet buffering to manage the intense microbursts typical of AI communication and collective patterns, Arista stated.
The 7060XE7 family is built on Broadcom Tomahawk 6 silicon. Arista is also working with AMD on next-generation compute silicon and NICs to enable scale-out AI fabrics, the company said. Strategically, the 7060XE7 Series signifies Arista's transition from offering standalone, high-performance switches to providing rack-scale systems that can handle the extreme density, power, and thermal efficiency AI requires, Arista stated. The platforms allow customers to build scale-up and scale-out AI fabrics using air, liquid and hybrid-cooled technology.
Specific Configurations
Specific configurations include: 7060XE7-64PS and 7060XE7-64PRS 4U Rack Switches: Available in Q4, these air-cooled systems offer support for pluggable Integrated heat sink (IHS) and Riding heat sink (RHS) optics. IHS is aimed at current air-cooled data centers, and RHS would be aimed at future liquid-cooled AI fabrics and extreme port density, Arista stated. 7060XE7-64PRS-RV3-L: This is a specialized 2OU liquid-cooled platform for high-density clusters, featuring 224G SerDes. This system uses DC power from the ORv3 rack and contains no internal fans, integrating with liquid-cooled XPU servers to maximize power efficiency. It will be available in Q1 2027. 7060XE7-128PE: Also coming in Q1 2027, these devices provide 128 800G ports in an air-cooled 4RU design, utilizing 100G SerDes, for environments requiring deployment flexibility and backward compatibility.
Software and Protocol Support
On the software side, EOS is the featured network operating system, but the family also supports open-source software such as Software for Open Networking in the Cloud (SONIC) and OpenSwitch. One of the portfolio's key features is the inclusion of full support for 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. MRC is an open protocol where endstation NICs stripe their traffic across multiple links and paths to the receiver, with out of order packets automatically handled, wrote Arista's Kenneth Duda, president and CTO, and Alan Judge, distinguished engineer, in a blog about the technology. MRC responds to network congestion signals (ECN and packet trimming), shifting load to the best-performing paths, and avoiding links and paths that can't actually reach the destination altogether. MRC monitors each path, steering around congestion, avoiding paths with link errors, and avoiding failed links, the authors stated. We've proven in production that this approach achieves very high fabric utilization with good load balancing, while interoperating seamlessly with scale-across and WAN networks utilizing standard dynamic routing protocols, Duda and Judge wrote.
The software also supports load balancing, congestion management, telemetry and diagnostics, and other technologies that will be core to AI networking, Arista stated. The new Arista family joins a growing ecosystem of vendors looking to tap into the 1.6T Ethernet world, which includes Cisco, Nvidia, Celestica and others. Arista Network's new 7060XE7 Series is a strong signal of where large-scale AI fabrics are heading: higher bandwidth, better power efficiency, and tighter integration between compute, optics, silicon, cooling, and network operating software, wrote Sameh Boujelbene, vice president, data center switch and AI networks market research for Dell Oro, in a LinkedIn post. Among the features that stand out to her are strong customer and ecosystem validation from Microsoft Azure, Oracle Cloud Infrastructure, Meta, AMD, and Broadcom.
Background and Market Context
The 1.6T Ethernet space is rapidly evolving as hyperscalers and enterprises seek to interconnect thousands of accelerators for AI training and inference. Arista's new platforms leverage the latest Tomahawk 6 chipset from Broadcom, which provides 51.2 Tbps of switching capacity in a single device, enabling 64 ports of 800G or 128 ports of 400G in a compact form factor. This silicon generation also introduces advanced features like in-network computing and programmable packet processing, which are critical for reducing tail latency and improving collective communication efficiency in AI workloads. The integration with AMD's next-generation compute silicon and NICs highlights a trend toward tighter collaboration between switch and server vendors, as AI fabrics require optimized endpoints to fully utilize the network's potential. AMD has been expanding its networking portfolio through acquisitions and partnerships, and this collaboration could lead to more efficient GPU-to-switch connectivity using RDMA over Converged Ethernet (RoCEv2) and other protocols.
The move to rack-scale systems reflects a broader industry shift from standalone switches to integrated solutions that include power, cooling, optics, and management within a single rack. Arista's liquid-cooled 7060XE7-64PRS-RV3-L, for example, is designed to work directly with ORv3 standard racks that supply DC power and liquid cooling loops, eliminating the need for internal fans and reducing power consumption by up to 30% compared to air-cooled alternatives. This is particularly important for AI clusters where hundreds of switches generate significant heat and power demands. The hybrid cooling approach allows customers to mix air- and liquid-cooled switches within the same rack, enabling phased migration to higher-density topologies.
The 7060XE7 Series also marks a significant step forward in Open Compute Project standards adoption. By fully supporting MRC, Arista is betting that open, multi-path transport will become the de facto standard for AI Ethernet fabrics, competing with technologies like NVIDIA's InfiniBand and other proprietary solutions. MRC's ability to stripe traffic across multiple paths and respond to congestion in real time could reduce the need for expensive, high-radix switching and instead allow simpler leaf-spine architectures to achieve similar performance. The protocol's built-in support for packet trimming and ECN ensures that it works seamlessly with standard Ethernet congestion control mechanisms, lowering the barrier to adoption for existing data center operators.
From a timeline perspective, the first devices in the 7060XE7 family will sample in late 2026 and become generally available in early 2027. By then, the AI networking market is expected to have grown to over $10 billion, with Ethernet share increasing as non-InfiniBand fabrics gain traction. Arista's long-standing relationship with major cloud providers such as Microsoft Azure and Meta provides a strong validation path, as these customers have been early adopters of previous Arista platforms for AI infrastructure. The company's focus on open standards and disaggregated software also aligns with the preferences of hyperscalers who want to avoid vendor lock-in.
In addition to the hardware, Arista is investing heavily in software automation and analytics for AI fabrics. EOS now includes AI-specific features like dynamic load balancing that monitors per-flow latency and reroutes traffic before congestion degrades performance. The system also supports streaming telemetry protocols such as gRPC and OpenConfig, allowing network operators to integrate with third-party AI orchestration tools. Arista's CloudVision management platform can be used to automate provisioning and monitoring of the 7060XE7 switches, reducing the operational overhead of managing large AI clusters. The combination of hardware advancements and software intelligence positions the 7060XE7 Series as a comprehensive solution for next-generation AI data centers.
Source: Network World News