Arista Reports $3.66 Billion Quarter, Targets 10 GW of AI Networking

Arista Networks posted $3.66 billion in third-quarter revenue, up 58.5 percent year on year, and set a target of supplying networking for 10 gigawatts of AI cluster capacity by the end of 2027.

The September quarter showed the clearest inflection yet from generative AI spending. Product sales reached $3.19 billion, up 66.9 percent year on year and 22.4 percent sequentially. Services revenue grew 17.9 percent to $467.6 million. Sequential growth of 20.5 percent overall was two to three times the company's normal pace and the highest in its history.

Profitability set records as well. Operating income rose 67.9 percent to $1.64 billion. Net income climbed 82.9 percent to $1.56 billion, or 42.7 percent of revenue — the highest margin the company has ever recorded. Arista exited the quarter with $14.36 billion in cash and equivalents, $11.4 billion in customer purchase commitments and $6.84 billion in deferred revenue.

What did CEO Jayshree Ullal say?

On the Wall Street earnings call, chief executive Jayshree Ullal said Arista is on track to sell about $3.5 billion in AI front-end and back-end networking gear and services in 2026. She expects that figure to double to roughly $7 billion in 2027. Campus switching will hit $1.25 billion this year and grow to $2 billion next year, an adjacency that takes share from Cisco and others.

At the midpoint of those targets, total revenue grows 52 percent in 2026 to $13.3 billion, then another 39.1 percent in 2027 to $18.5 billion.

Ullal described the shift in the customer base. In 2024 and 2025, Arista spoke of four or five customers running AI clusters of 50,000 to 100,000 Nvidia GPUs, with substantial InfiniBand still shipping. In 2026, more than a hundred customers use its AI-centric Etherlink switches. By the end of 2027 she forecasts linking ten or more different kinds of accelerators.

Microsoft and Meta Platforms, the two customers that each account for 10 percent or more of annual revenue, are working with Arista on ESUN scale-up networks for their respective Maia and MTIA XPUs, Ullal said. Both remain on track as major spenders through 2026 and 2027. Arista may add one or two more large customers in the coming year.

When does scale-up Ethernet arrive?

Not immediately. Arista has worked with AMD, Arm, Meta, Microsoft and Qualcomm on rack-scale switching based on its Etherlink switching and routing lines. Ullal said ESUN-based scale-up trials should start in 2027, initially with rack-scale systems at a radix of 128 or 144 accelerators per rack and roughly 1,000 across a row. Production rollouts are not expected until 2028.

Arista has not committed to UALink switching, the more direct competitor to Nvidia's NVSwitch stack, though third-party UALink ASICs are available from Astera Labs, Marvell, Upscale AI and possibly Panmnesia. The company also has no play for scale-in networking on accelerator packages themselves, a domain left to AMD, Nvidia, Broadcom, Marvell, Alchip and perhaps Intel and Qualcomm.

The commercial logic behind the roadmap is cost and openness. No party except Nvidia wants proprietary networks anywhere in the AI cluster interconnect hierarchy. Nvidia earned its first-mover advantage with tightly coupled, shared-memory GPU clusters built on NVLink and NVSwitch, but these face Ethernet-based alternatives running ESUN or UALink over Ethernet, plus native UALink for gluing GPU memories together and UCI-Express within accelerator packages.

Where is the money today?

Scale-up revenue remains future business. Arista currently earns from Ethernet switches for conventional cloud infrastructure, AI front-end networks that link clusters to storage and the outside world, and its R-series routing switches that let AI clusters span multiple datacenters and regions — a necessity where local power limits cap facility size.

In the meantime, the non-AI portions of the business tell a mixed story. Scale-across routing gear is growing well in 2026, while core cloud and datacenter switching and the edge segment grow only modestly — a dichotomy the source observes in servers too, where traditional workloads are expanding faster than in switching.

With ESUN trials starting in 2027 and production in 2028, Arista's near-term growth rests on front-end, back-end and scale-across Ethernet, with the 10-gigawatt and $7 billion AI targets anchoring the 2027 outlook.