Not Nvidia. Not Cisco. This Is the AI Networking Stock Quietly Wiring Every Data Center

Arista Networks is presented as a beneficiary of AI infrastructure spending, with its Ethernet networking products helping data centers connect large groups of accelerators; its Etherlink SU-144 architecture supports up to 144 accelerators in a single-hop network and scales to 1,024. Q2 revenue rose 37.7% year over year to $3.03 billion, while net income increased to $1.21 billion from $888 million and diluted EPS rose to $0.95 from $0.70. Arista shares had gained 54% so far this year, and the company guided for Q3 revenue of $3.3 billion and diluted net income per share of $1.06–$1.08; competition and potential in-house equipment development by cloud customers remain risks.

Analysis

The key underwriting question is whether Ethernet becomes a durable share gainer in large AI clusters—not whether AI networking demand grows. If hyperscalers adopt open Ethernet at scale, Arista can benefit from higher network content per accelerator and software/control-plane attachment; the same shift could improve the appeal of AMD-based systems by reducing dependence on tightly integrated proprietary stacks. Broadcom is both a potential ecosystem enabler and a competitor for custom switching silicon, while NVIDIA’s proprietary interconnect remains the main substitution risk. Meta and Microsoft could gain operating flexibility, but also have bargaining power and incentives to internalize or dual-source networking.

Near term, the upcoming Q3 report is the catalyst: the stated revenue outlook implies a demanding execution test, but the article provides no valuation, order visibility, customer concentration, or evidence that the new architecture has reached meaningful deployment. The product’s claimed power savings are potentially valuable where power availability constrains rack deployments, yet should be treated as a vendor claim until independently reflected in customer adoption or economics. Arista’s strong share performance raises the bar for incremental upside; a good quarter may not suffice if forward guidance or AI orders disappoint.

Over 6–18 months, watch Ethernet adoption across accelerator vendors, hyperscaler capex, and whether proprietary fabrics retain performance advantages at scale. The thesis weakens on customer-built alternatives, delayed deployments, or evidence that AI networking revenue is concentrated and less recurring than expected. No valuation or consensus data are supplied, so an outright valuation call is premature.

Key Decisions for Investors

• Keep ANET on a catalyst watch rather than chase the reported outperformance ahead of earnings. Consider initiating or adding only if Q3 results and forward commentary confirm sustained AI-networking demand; size against the possibility that strong reported growth is already reflected in the share price.
• For a relative-value expression, consider long ANET versus short CSCO only after confirming valuation, estimate revisions, and business-mix differences. The thesis is faster AI-networking exposure at Arista, not that Cisco is structurally impaired; abandon it if Cisco demonstrates comparable AI order momentum or ANET’s relative estimates roll over.
• Track earnings commentary for AI-related orders, customer concentration, backlog/conversion, gross-margin direction, and deployment evidence for the new architecture. These missing data points determine whether product announcements translate into revenue and returns rather than merely broaden the narrative.
• Falsification triggers: ANET cuts forward revenue expectations, AI demand converts more slowly than implied by its outlook, or major deployments favor proprietary fabrics. Conversely, confirmed multi-customer Ethernet adoption and sustained guidance increases would support a more constructive 6–18 month view.