Arista Networks (NYSE:ANET) executives said demand for AI networking remains broad-based across cloud providers, AI labs, neoclouds, enterprises and campus customers, while supply availability remains a constraint on the company’s ability to meet that demand.
Speaking at the Rosenblatt Age of AI technology conference, Chief Financial Officer Chantelle Breithaupt said the company’s Etherlink AI networking portfolio has expanded from four or five customers in 2024 to more than 100 customers. The portfolio initially focused on 800G scale-out networking and has since expanded to address scale-across and, eventually, scale-up AI networking opportunities.
Demand Seen as Project-Driven, Not Supply Pull-Forward
Breithaupt said Arista does not view current demand as primarily driven by customers placing orders early because of component shortages. Instead, she said customer conversations and project timelines point to underlying demand for AI infrastructure and faster deployment.
“We do not see pull forward at all,” Breithaupt said, adding that any acceleration would be more likely tied to customers seeking to reach AI outcomes sooner than to supply scarcity. She noted that long lead times make it unlikely customers would make commitments without confidence in their projects.
Still, supply remains a limiting factor. Breithaupt said Arista and its suppliers have made progress supporting the company’s 40% revenue guide, but supply is not unconstrained heading into next year. “If we get a bit more supply, we could even go a little higher than the guide,” she said.
Software and Scale-Across Expand AI Opportunity
Brendan Gibbs, Arista’s vice president of product line management, said the company’s EOS network operating system remains central to its AI networking proposition. He characterized the software’s core value as providing “speed with quality,” helping customers deploy and operate AI clusters more quickly while providing analytics and visibility into cluster performance and GPU utilization.
Gibbs said software becomes increasingly important as customers extend AI deployments across geographically distributed data centers. In such scale-across architectures, customers must manage workload placement, congestion, packet loss, routing and encryption over longer distances.
Arista has identified scale-across as roughly one-third of its business under its most recent guidance, Gibbs said. He described the category as an expansion opportunity involving more ports, systems and software intelligence, rather than simply a need for a single routing or optical transport device.
The company said its existing position in large Ethernet-based scale-out AI clusters provides an advantage in pursuing scale-across deployments. Gibbs said scale-across is a natural extension for vendors that have already won the underlying AI cluster deployments.
Scale-Up Investments Target Longer-Term Market
Arista is also investing in Ethernet-based scale-up networking, a market Gibbs said outside industry analysts have estimated could reach as much as $20 billion by 2030. Scale-up networking would connect processing units within a rack and support coherent memory across different XPUs, or accelerators.
Gibbs said Arista is helping define the Ethernet Scale-Up Networking, or ESAN, standard through the Open Compute Project. Breithaupt cautioned that scale-up will not be material to Arista’s 2027 financial results because standards work, pilots and trials remain ahead. The company expects the opportunity could become a new segment beginning in 2028.
The executives said they expect the scale-up market to include proprietary NVLink implementations, lower-feature white-box-equivalent offerings and branded, value-added Ethernet solutions. Arista does not intend to enter the white-box business, Breithaupt said, but sees an opportunity for customers that value a consistent EOS-based operating environment across scale-up, scale-out and scale-across deployments.
1.6T Plans, Margins and Campus Growth
Gibbs said Arista is currently conducting 1.6T product trials with large customers and expects volume production next year. The company has announced both air-cooled and liquid-cooled options. Arista also introduced open co-packaged optics, or open CPO, and XPO, a liquid-cooled, pluggable 1.6T optical technology. Gibbs said XPO is generating customer discussions and product-planning activity but is not yet contributing to revenue or deferred revenue.
On profitability, Breithaupt said product mix remains the primary driver of gross margin. She said investors should assume a 62% to 64% gross-margin range unless the company provides different parameters. Price increases intended to offset higher component costs are designed to preserve margins rather than expand them, she said.
Breithaupt also said Arista expects deferred revenue to end the fourth quarter above its year-earlier level, although quarterly balances may fluctuate. She described deferred revenue as a complementary demand indicator that has been moving through the business within expected timelines.
Outside AI data center networking, Breithaupt highlighted momentum in campus networking. She said Arista is targeting a 5% market-share gain and expects campus revenue to rise from $800 million last year to $1.25 billion under its current guide, driven in part by greenfield customer wins.
About Arista Networks (NYSE:ANET)
Arista Networks, Inc is a technology company that designs and sells cloud networking solutions for large-scale data centers and enterprise environments. The company is best known for its high-performance switching and routing platforms, which are used to build scalable, low-latency networks for cloud service providers, internet companies, financial services, telecommunications, and enterprise IT. Arista’s offerings emphasize programmability, automation and telemetry to support modern, software-driven network architectures.
Central to Arista’s product portfolio is its Extensible Operating System (EOS), a modular network operating system that provides consistent programmability, stateful control and advanced visibility across the company’s hardware platforms.
