Arista Networks just printed its first $3 billion revenue quarter, scaling the datacenter networking business from a high-teen grower into a $12-billion-plus run-rate platform in the span of nine quarters, and the print matters not because of the headline number but because of the gross-margin mix shift underneath it. Revenue of $3,035.7 million in the second quarter of 2026 grew 37.7% year over year, GAAP operating margin reached 45.4%, and non-GAAP diluted earnings per share of $1.02 represented 40% growth on the prior-year quarter, even as the company absorbed a deliberate gross margin compression to 62.9% GAAP (from 65.2% a year ago) in order to fund the AI-fabric buildout that is now visibly driving the top line. The single load-bearing observation, in our view, is that Arista is no longer a "datacenter switching for hyperscalers" story; it is an "Ethernet fabric for AI training and inference clusters" story, and the new 1.6 Tbps Etherlink platforms the company introduced this quarter are the productization of that pivot.
The mechanism behind the print is straightforward but important. Product revenue grew 38.8% to $2,605.2 million, service revenue grew 31.3% to $430.5 million, and Americas-region revenue grew 35.6% to $2,337.9 million, with EMEA accelerating to 55.6% growth as international Tier-1 cloud customers began matching the U.S. hyperscaler deployment cadence. The Q3 2026 guidance of approximately $3.3 billion in revenue, 48% to 49% non-GAAP operating margin, and $1.06 to $1.08 non-GAAP diluted EPS, all of which were issued with the print, imply that the company expects the AI-fabric mix shift to continue compressing gross margin in absolute terms while expanding operating margin through operating-leverage absorption, and that the second $3-billion-plus quarter is now the base, not the peak.
The single load-bearing risk is hyperscaler capex concentration. Arista does not disclose named customers in this filing, but the Americas concentration at 77.0% of Q2 revenue, the explicit risk-factor language about a "limited number of customers" representing a "substantial portion of revenue," and the $8.4 billion of remaining performance obligations anchored to customer contracts with acceptance terms together imply that a single quarter's worth of AI capex digestion by two or three Tier-1 buyers could swing reported revenue by 10% to 15% in either direction. The falsifiable clock is the Q3 print itself, expected in early November 2026, with management's $3.3 billion revenue target as the first hard test of whether the AI-fabric transition is durable capacity expansion or front-loaded buildout that ends in 2027.