Arista Networks has crossed a major revenue threshold, posting its first-ever quarter above $3 billion as demand for artificial intelligence infrastructure continues to reshape the networking market. The company reported revenue of $3.036 billion for the second fiscal quarter, up 12.1% sequentially and 37.7% year over year.

“Customers see networking as the central nervous system for infrastructure from the client to campus to data and AI centers,” said Jayshree Ullal, Arista’s CEO, during the quarterly earnings call. “Our AI fabrics momentum with Etherlink switches now exceeds 100 cumulative customers from the initial four to five customers I spoke of in 2024.”

The $3 billion milestone is especially notable as a measuring stick for how rapid the AI buildout has become. Five years ago, Arista recorded $2.9 billion in revenue for all of 2021, Ullal noted. Today, the company is pulling in that amount in a single quarter.

Ullal said growth is occurring across nearly all parts of the portfolio, including back-end AI fabrics, front-end data center networks, campus switching and routing. The scale-across switching and routing market, which connects AI data centers and other large environments, is expected to be worth $15 billion to $20 billion by 2030, placing Arista in a strong position as carriers and cloud providers expand capacity.

Given the momentum, Arista now projects 40% annual growth. That represents an incremental $2.1 billion over the company’s 2025 Analyst Day goal of $10.5 billion and an incremental $1.1 billion over the projections Arista shared in May 2026, when it targeted $11.5 billion.

AI networking demand expands

The record result is not simply a reflection of a strong market. Arista’s AI-related product line has expanded rapidly since 2024, when Etherlink switches first appeared and began drawing customers evaluating high-bandwidth, low-latency networks for GPU and XPU clusters. Those initial designs have since moved from trials into production.

“I have never witnessed the combination of rapid innovation and scale deployment that we are seeing in AI networks,” said Kenneth Duda, Arista’s president and CTO. Duda and other executives described a competitive environment in which reliability, uptime and operational simplicity are becoming as important as raw data rates.

The shift toward AI workloads has forced Arista and its rivals to rethink network design. Traditional data center networks often relied on multipath designs that spread traffic across many flows. AI training jobs, however, create massive, synchronized communication patterns in which a single flow between two accelerators can be large enough to saturate a link. That behavior has driven the need for new approaches to congestion control, path selection and failure recovery.

Arista’s Etherlink product line is aimed squarely at that challenge. With more than 100 cumulative customers, the AI fabric business has moved from a proof-of-concept phase to a mainstream growth driver. Arista does not disclose a separate revenue line for AI fabrics, but executives have consistently said the category is one of the fastest-growing parts of the company.

Supply chain pressures beginning to ease

Arista faced a difficult supply environment just one quarter ago. Networking components such as memory, chips and wafers were in short supply, and rising component prices were pressuring margins. The situation has improved, although it has not disappeared.

“Arista has spent the last six months improving our supply chain to meet growing product demand, and we’re seeing significant improvements,” said Todd Nightingale, Arista’s president and COO.

Nightingale outlined several concrete steps the company has taken. Arista has signed multiyear agreements with leading vendors for strategic components, qualified alternative suppliers in critical areas and created supply chain capacity for next-generation AI technology. As a result, the company’s memory supply is now secured for 2026, with added visibility well into 2027 across DDR4, DDR5 and NAND memory.

“Relationships with our strategic silicon vendors continue to be strong, with really excellent collaboration in both supply chain and technical engagements,” Nightingale said.

The company has also improved its position for printed circuit boards and optics. Arista says it can now build capacity in a 12-month window, a meaningful improvement given the longer lead times that often accompany networking hardware. It strengthened commitments from key suppliers and improved inventory management for thousands of component SKUs.

“We’ve improved our lead times and inventory management of thousands of component SKUs, improving sub-component pipelining and multi-sourcing, and providing increased flexibility with reduced inventory risk,” Nightingale said.

One less visible but increasingly important area is liquid cooling. As AI clusters consume more power, more customers are adopting liquid-cooled infrastructure. Nightingale said Arista has established a liquid-cooling supply chain “capable of driving and delivering the next generation of AI infrastructure.” That includes vendors for cold plates, quick disconnects and tubing, with capacity agreements tied to next-generation AI systems.

Ullal cautioned that the supply chain is not entirely healed. “I don’t want you to believe that suddenly we waved a magic wand and all our problems have gone away,” she said. “The industry is going to have a two-year problem [with memory and other silicon availability challenges], and I don’t think we get out of it as an industry until 2028. But Arista is taking individually and specifically steps in the first half of this year that we believe will have results in the back half of this year.”

EOS innovations target AI reliability

The quarterly call also gave Arista’s technical leaders a chance to outline new capabilities in the company’s EOS operating system. Duda pointed to three technologies that he said are helping to differentiate Arista in AI networks.

The first, Smart System Upgrade, is designed for a world where software updates happen more often, often because of security vulnerabilities. Traditional switch upgrades usually require a full reboot, causing downtime and operational disruption. Arista says SSU can upgrade switch software without interrupting traffic.

“Frequent upgrades are a hard reality today, especially as AI both uncovers security vulnerabilities and creates tools to exploit them,” Duda said. “While many competing systems require a full reboot to address these issues, leading to expensive and disruptive downtime, Arista EOS handles these upgrades seamlessly.”

Second is Multipath Reliable Connection, or MRC, which aims to maximize utilization of accelerators. In first-generation AI networks, every packet in an XPU-to-XPU flow had to take the same path through the fabric. If two flows hashed to the same path, they could experience collisions and slow each other down. MRC enables a sender to spray a single flow across many paths, with the receiver reassembling out-of-order data. That removes the performance penalty associated with fabric cache collisions.

Third is Segment Routing IPv6, or SRv6. While SRv6 itself was not invented for AI, Duda said applying it to load balance AI fabrics is a game changer. The sender tags each packet with a stack of SRv6 segment IDs that dictate the exact path through the network. Arista couples that capability with real-time congestion signaling to shift packets away from hotspots.

“Because Arista EOS provides a single unified operating system, we support this SRv6 intelligence all the way from the scale-out fabric to the long-distance scale-across routing,” Duda said. “It gives our customers the combination of high-quality top performance and operational simplicity that Arista is known for.”

Optics strategy stays open

Arista also weighed in on the furious debate around optical interconnect technologies for AI. Some vendors are pushing proprietary co-packaged optics for performance, while others believe traditional pluggables have enough life left to remain the dominant choice.

Ullal said Arista expects pluggable optics and copper to carry the majority of AI networking traffic through 2028-2029. “I think there’s very much a philosophy there [of] copper if you can, optics if you must,” she said.

Within racks, two- or three-meter distances can often be handled by copper. Pluggable optics become necessary for longer distances. Some companies are now experimenting with proprietary versions of co-packaged optics, which place the optical engine close to the switch chip to reduce power and heat. But Arista is not committing to proprietary designs.

“Arista is not a fan of five different proprietary implementations,” Ullal said.

Instead, the company is pursuing an open CPO approach, using socketed optical engines and pigtail fibers that allow modules to be fully pretested before deployment. The goal is a true open interface that can work with multiple vendors and multiple switch configurations.

“We don’t think open CPO is going to happen overnight,” Ullal said. “But the idea here is to have a truly open interface that can operate with multiple vendors and multiple switch configurations.”

Arista says CPO and near-packaged optics are expected to enter trials in 2027, though they will represent a small part of the overall market in the near term.

The company also announced extended pluggable optics, or XPO, a form factor designed for high-speed optics. XPO has attracted more than 100 optics module suppliers as part of a multi-source agreement to build and support the specification. That broad support, Arista contends, should give customers options and prevent lock-in to any single optical vendor.

For now, Arista’s focus remains on executing against strong demand. The question is less about whether AI networking will continue to grow and more about how quickly Arista can obtain components, deliver products and maintain the high availability that operators expect. The company believes it is making progress on all three fronts.

Source: Network World News