Arista Networks introduced its new 7060XE7 Series, a family of 1.6 Terabit Ethernet platforms designed to serve as the networking foundation for next-generation AI clusters. The portfolio expands Arista’s Etherlink architecture and targets both scale-out and scale-up AI deployments, supporting environments ranging from air-cooled systems to fully liquid-cooled racks. The company positions the new platforms as a transition from standalone switching infrastructure to integrated rack-scale AI systems capable of supporting hundreds of thousands of AI accelerators.
The 7060XE7 family includes multiple configurations built around 100 Tbps switching capacity and support for 224G SerDes technology. Arista said the systems address the increasing density, power, and thermal requirements of AI infrastructure while enabling deployment flexibility across diverse accelerator architectures. The portfolio includes air-cooled 64-port and 128-port systems as well as a liquid-cooled 2RU platform designed for integration into Open Rack v3 environments. Support for Linear Pluggable Optics (LPO) can reduce interconnect power consumption by approximately 60%, helping operators maximize compute density within constrained power budgets.
Arista highlighted collaboration with major cloud operators including Meta, Microsoft, and Oracle, along with ecosystem partners AMD and Broadcom. The systems incorporate AI-specific software capabilities within Arista EOS, including congestion management, dynamic load balancing, Multipath Reliable Connection (MRC) support, and advanced telemetry. The platforms will begin shipping in phases, with the 64-port air-cooled systems scheduled for availability in Q4 2026 and additional liquid-cooled and 128-port configurations following in early 2027.
- New 7060XE7 portfolio delivers 1.6T Ethernet connectivity for AI fabrics
- Up to 100 Tbps switching capacity per platform
- Supports both scale-up and scale-out AI architectures
- Available in air-cooled and liquid-cooled configurations
- LPO technology reduces interconnect power consumption by approximately 60%
- Built on Broadcom Tomahawk 6 switching silicon
- Supports both Arista EOS and open network operating systems
- Features MRC resiliency, advanced congestion management, and AI-specific load balancing
- Initial systems ship in Q4 2026
“The AI era requires a shift in how we think about the network—it is no longer a standalone layer of infrastructure, but a tightly integrated component of the AI supersystem,” said Tyson Lamoreaux, Senior Vice President of Cloud and AI Networking at Arista Networks.
🌐 Analysis
Arista’s announcement reflects the industry’s transition toward rack-scale AI architectures built around 224G SerDes and 1.6T Ethernet connectivity. As hyperscalers deploy increasingly dense AI clusters, networking has become a primary determinant of overall system performance, particularly for large-scale training workloads that require low-latency collective communications among tens of thousands of GPUs and XPUs.
The launch also reinforces Ethernet’s momentum in AI infrastructure. Arista joins Broadcom, AMD, Meta, Microsoft, Oracle, and other ecosystem partners in advancing Ethernet-based AI fabrics as an alternative to proprietary interconnect approaches. The inclusion of liquid-cooled switching platforms aligns with broader industry efforts to support AI racks approaching and exceeding 100 kW per rack, where cooling and power efficiency become critical deployment constraints.
| XPO (eXtended Performance Overlay) for AI Fabrics | |
|---|---|
 | |
| Initiative | XPO (eXtended Performance Overlay) |
| Announced | March 2026 |
| Purpose | Improve Ethernet performance, resiliency, and scalability for AI training and inference clusters. |
| Architecture | Overlay software layer operating above Ethernet and RDMA fabrics. |
| Key Functions | Multipath Reliable Connection (MRC), congestion management, adaptive load balancing, telemetry, failure recovery, and job completion optimization. |
| Target Scale | AI clusters containing tens of thousands to hundreds of thousands of GPUs, XPUs, and accelerators. |
| Relationship to UEC | Designed to complement Ultra Ethernet Consortium specifications for AI networking. |
| Founding Supporters | Arista, AMD, Broadcom, Cisco, Juniper Networks, Meta, Microsoft, NVIDIA, Oracle and other ecosystem partners. |
| Connection to 7060XE7 | The new Arista 7060XE7 platforms incorporate XPO-aligned capabilities including MRC resiliency, dynamic load balancing, congestion signaling, and AI fabric telemetry. |
| Industry Significance | Represents a major industry effort to establish Ethernet as the preferred interconnect architecture for hyperscale AI infrastructure. |
