Arista Networks reported third-quarter 2026 revenue of 3.66 billion dollars on November 3, up 58.5 percent from 2.31 billion dollars a year earlier and up 20.5 percent from the second quarter. The company described the result as its fastest year-on-year quarterly revenue growth since its initial public offering in 2014, and the figure cleared Arista's own guidance of about 3.3 billion dollars by more than 350 million dollars.
Profitability kept pace with the growth. Arista posted a GAAP operating margin of 44.9 percent, up from 42.4 percent a year earlier, while its non-GAAP operating margin reached 49.1 percent, up from 48.6 percent. GAAP diluted earnings per share rose to 1.22 dollars, compared with 0.67 dollars a year earlier, and non-GAAP diluted earnings per share reached 1.20 dollars. In an unusual twist, the GAAP figure came in slightly above the adjusted figure for the quarter.
The driver is the network itself. As cloud providers and enterprises pack more accelerators into AI data centers, the switches that move data between those processors have become as critical as the processors themselves. A training cluster is only as productive as the fabric connecting it, and Arista has built its business on high-performance Ethernet switching and the software that manages it. That position is now paying off at a scale few networking vendors have matched, with demand coming from both large AI customers and established enterprise accounts.
Management's outlook suggests the momentum is not fading. For the fourth quarter, Arista expects revenue between 3.9 billion and 4 billion dollars, a non-GAAP operating margin of 48 to 49 percent, and non-GAAP diluted earnings per share of 1.26 to 1.30 dollars. The company also raised its full-year 2026 revenue forecast for the fourth time, to about 13.3 billion dollars, and gave investors an early look at 2027, projecting revenue of about 18.5 billion dollars, growth of 38 to 40 percent.
Arista is also pushing an open Ethernet approach for AI systems, developing architectures meant to handle both scale-up connections inside tightly coupled systems and scale-out networks linking much larger clusters. The company has been collaborating with other chip and platform vendors on those open designs, betting that customers will favor the flexibility of Ethernet over proprietary interconnects as their AI installations grow denser and more complex.
There are reasons for caution alongside the record numbers. Arista reported about 3.21 billion dollars in inventory at the end of September, up from 2.25 billion dollars at the end of 2025, a buildup that supports deliveries when components are scarce but leaves the company exposed if customer schedules slip or designs change. Total deferred revenue stood at 6.84 billion dollars at quarter end, essentially flat against June, even as recognized revenue surged.
The larger question hanging over the whole sector is spending discipline. Cloud customers are committing enormous capital to AI infrastructure while investors scrutinize the returns, and a slowdown at a handful of large buyers would reach networking suppliers quickly. Arista's next test arrives with its fourth-quarter results: revenue approaching 4 billion dollars, margins near current levels, and evidence that orders extend beyond the first wave of AI buildouts. For now, the company's November report makes one point clearly. In the AI era, the network is no longer plumbing in the background. It is one of the fastest-growing businesses in the data center.