Arista Networks

Full briefing

Arista Networks, Inc. (NYSE: ANET) is headquartered in Santa Clara, California, and had approximately $7 billion in annual revenue in FY2024. Founded in 2004 by Andy Bechtolsheim, Ken Duda, and David Cheriton, Arista disrupted the incumbent Cisco-led data center networking market by building a software-centric operating system (EOS — Extensible Operating System) on top of merchant silicon from Broadcom, rather than developing proprietary ASICs. Arista's switches are the backbone of AI training cluster networking at major hyperscalers. Google, Meta, Microsoft, and AWS all use Arista's 7050X, 7060X, and 7800-series chassis switches — built on Broadcom's Tomahawk 4/5 switching ASICs — to provide the high-bandwidth, low-latency Ethernet fabric that connects thousands of GPUs in a single AI training job. xAI's Colossus cluster, one of the world's largest GPU deployments, uses Arista switches as part of its network fabric alongside NVIDIA InfiniBand. The battle between InfiniBand and Ethernet for AI cluster interconnects is a defining strategic question for Arista. NVIDIA's acquisition of Mellanox gave it control of the InfiniBand ecosystem, which dominated early large-scale AI training due to its lower latency and RDMA capabilities. However, Ultra Ethernet Consortium (UEC) — a coalition including AMD, AWS, Google, Meta, and Arista — is developing next-generation Ethernet specifications that close the latency gap and enable RoCE (RDMA over Converged Ethernet) at scale. If UEC Ethernet displaces InfiniBand as the dominant AI fabric, Arista is the primary beneficiary. Arista's CloudVision software platform provides network-wide telemetry, streaming analytics, and automated configuration management that allows data center operators to manage tens of thousands of switch ports as a single logical fabric. This software layer creates high customer stickiness and recurring revenue through subscription licenses. For 2025 and beyond, Arista is targeting the 800G and emerging 1.6T Ethernet switch market, where its 7060X5 platforms using Broadcom's Tomahawk 5 (51.2 Tbps capacity) are positioned to serve the next generation of AI clusters. The company has also expanded into AI networking for campus environments with its Cognitive Unified Edge architecture.