Arista Networks recently announced its new 7060XE7 Series, a portfolio of 1.6 Terabit (1.6T) Ethernet switching platforms designed for rack-scale AI infrastructure. The announcement extends Arista’s Etherlink architecture from 800G to 1.6T and introduces three hardware configurations for air-cooled, liquid-cooled, and high-density 128-port deployments.

The 7060XE7 Series runs on Broadcom‘s Tomahawk 6 switch silicon and supports both Arista’s Extensible Operating System (EOS) and open network operating systems, positioning these platforms for hyperscale, cloud titan, and emerging AI cluster markets.

With validated deployments confirmed by Meta, Microsoft, and Oracle, Arista enters the 1.6T market with established commitments from hyperscalers. Arista already claims the number one market share in high-speed switching above 10 Gigabit Ethernet. The 7060XE7 targets the next growth inflection, with Arista projecting production-scale 1.6T deployments to ramp through 2027.

Details

The 7060XE7 Series delivers a maximum switching capacity of 100 Terabits per second with 1.6T per-port throughput, built on Broadcom Tomahawk 6 silicon.

The portfolio spans three distinct hardware configurations, each tailored to different thermal and deployment requirements. Arista engineers the 7060XE7 as a rack-scale radix, designed to integrate tightly with liquid-cooled XPU servers.

The core technical specifications:

• 7060XE7-64PS and 7060XE7-64PRS: Air-cooled rack switches in a 4RU form factor with 64 ports at 1.6T each. These systems support both Integrated Heat Sink (IHS) and Riding Heat Sink (RHS) optics configurations, providing deployment flexibility across data center environments with varying thermal management needs. Availability is targeted for Q4 2026.
• 7060XE7-64PRS-RV3-L: A 2OU liquid-cooled platform for high-density AI clusters, using 224G SerDes technology. This configuration draws DC power from ORv3 racks and has no internal fans, enabling direct integration with liquid-cooled XPU server infrastructure. Arista targets Q1 2027 for availability of this platform.
• 7060XE7-128PE: A 128-port, 800G air-cooled platform in a 4RU form factor using 100G SerDes. This configuration provides backward compatibility for environments transitioning from 800G to 1.6T infrastructure and is targeted for Q1 2027 availability.
• Linear Pluggable Optics (LPO) support: LPO integration reduces interconnect power consumption by approximately 60% compared with conventional optical modules, which Arista asserts lowers the overall total cost of ownership across high-density AI fabric deployments. This figure is a vendor claim and has not been independently validated.
• 224G SerDes: The liquid-cooled platform uses 224G SerDes technology to deliver 1.6T per-port bandwidth, enabling rack-scale signal integrity without external retimers or DSP-based modules.

The software stack supporting the 7060XE7 includes a suite of AI-specific EOS capabilities designed for the congestion patterns and collective communication workloads common in large-scale AI training:

• Dynamic Load Balancing (DLB) and Cluster Load Balancing (CLB) distribute traffic across the fabric and manage AI job completion times.
• Multipath Reliable Connection (MRC) support, which Arista describes as addressing the failure-amplifier problem, where a single link failure in a training job can stall the entire cluster. MRC provides redundant path management to prevent that outcome.
• Link Layer Retry (LLR) maintains performance stability at the physical layer, complementing MRC.
• PFC-aware Dynamic Load Balancing, PFC-aware Explicit Congestion Notification (ECN), and real-time telemetry to prevent head-of-line blocking.
• Congestion Signaling (CSIG) and Fast Congestion Notification Packets (CNP) for hardware-level congestion feedback.
• NetDI (Network Diagnostics Infrastructure) integration for uniform telemetry access across diverse operating environments.
• PDI (Platform-Dependent Infrastructure) abstraction for software portability across hardware architectures.

The 7060XE7 Series supports both Arista EOS and open Network Operating Systems, a deliberate design choice to serve cloud titan customers operating heterogeneous software stacks.

This dual-NOS support, combined with the Broadcom Tomahawk 6 silicon foundation, enables operators already familiar with Tomahawk-based platforms to extend their operational tooling.

Analysis

The 7060XE7 advances Arista’s strategy of transitioning from a switching vendor to a full-stack AI infrastructure platform provider. The 7060XE7 directly supports that objective by delivering the bandwidth generation that large AI clusters are designed to achieve.

Several elements of this announcement carry weight:

• Hyperscaler validation at launch: The simultaneous announcement of confirmed partnerships with Meta, Microsoft, and Oracle removes speculative risk from the 7060XE7 roadmap.
• Three-topology AI strategy: Arista’s AI networking portfolio addresses scale-out (horizontal leaf-spine), scale-up (vertical rack-scale), and scale-across (distributed multi-datacenter) deployments. The 7060XE7 strengthens scale-out and scale-up coverage, while scale-across relies on Arista’s 7800 spine and partnerships with coherent optics vendors.
• Rack-scale system framing: Positioning the 7060XE7 as a rack-scale system rather than a discrete switch is a deliberate differentiation from commodity switching alternatives. This requires Arista to deliver on its co-design and integration commitments, including thermal, power, and software co-optimization with XPU vendors.

Impact for Practitioners

The 7060XE7 Series is directly relevant to infrastructure engineers and network architects building or scaling AI training and inference clusters in hyperscale and large enterprise environments.

The Q4 2026 availability of the air-cooled configurations and the Q1 2027 availability of the liquid-cooled and 128-port variants mean that procurement teams evaluating 2027 AI cluster buildouts need to include these platforms in their planning cycles now, given the 38-to-52-week component lead times Arista has cited in recent earnings guidance.

Practitioners evaluating the 7060XE7 Series should consider the following operational factors:

• Thermal and facility readiness: The liquid-cooled 7060XE7-64PRS-RV3-L requires ORv3 DC power infrastructure and integration with liquid-cooled XPU server racks. Data centers without existing liquid-cooling infrastructure must make additional facility investments before this platform can be deployed.
• LPO compatibility: Linear Pluggable Optics require XPU-side NIC support for LPO termination.
• EOS operational consistency: Organizations already running Arista EOS across their data center fabric will find that the 7060XE7 integrates with existing CloudVision automation and telemetry pipelines.
• Open NOS flexibility: Hyperscalers running SONiC or custom network operating systems can deploy the 7060XE7 without adopting EOS. This option preserves operator-controlled software stacks while leveraging the Tomahawk 6 hardware platform.
• Scale-up vs. scale-out timing: Scale-out is the current deployment reality, while scale-up is a 2027-and-beyond opportunity tied to ESUN specification maturation. Organizations planning tightly coupled scale-up AI fabrics today should monitor Arista’s scale-up engineering engagements and expect production-ready scale-up solutions to follow the 7060XE7 launch.

Final Thoughts

Arista Networks’ 7060XE7 Series is a well-timed, technically credible response to the bandwidth demands driving AI cluster architecture decisions. The combination of 100 Tbps aggregate switching capacity, 224G SerDes, LPO support, and liquid-cooled form factors addresses the density and power-efficiency requirements that define next-generation AI data center infrastructure.

Some questions remain about execution at scale:

• LPO adoption depends on XPU NIC ecosystem readiness that Arista does not directly control.
• The liquid-cooled platform’s ORv3 power requirement creates a facility barrier that limits near-term deployment to customers with existing liquid cooling infrastructure.
• And the scale-up networking opportunity, which Arista itself identifies as requiring ESUN specification maturity and co-packaged copper evolution, remains a 2027-and-beyond opportunity, meaning 2026 AI cluster infrastructure builds will be evaluated primarily on scale-out Ethernet performance (an area where Cisco and NVIDIA are both executing).

Overall, Arista’s 7060XE7 launch consolidates Arista’s position as the reference vendor for large-scale Ethernet-based AI fabrics. Arista’s $3.5 billion AI 2026 revenue target, its leading market share in high-speed switching, and the depth of its EOS software stack across the world’s largest AI infrastructure operators give Arista a durable advantage heading into the 1.6T transition.

The critical test for 2027 will be whether Arista can extend that advantage into the scale-up topology, where architectural choices and silicon dependencies differ substantially from those in the scale-out switching business that has driven its current growth. They’re off to a strong start.