Arista continues to defy expectations, build enterprise momentum

When Arista Networks was founded in 2004, I was skeptical the company would have long-term success. Arista came to market solving the low-latency network problem, and it did so by building performant products that leverage merchant silicon but through good hardware and software engineering could outperform competitors.

My skepticism came from having seen other vendors with similar approaches come up short. Prior to Arista, there were several vendors, such as Foundry Networks and Force10 Networks, that had similar go-to-market plans to build a better mousetrap that could compete with Cisco. While they had some success, data center revenue stalled, and the companies derailed their missions by going after other markets. Neither Foundry nor Force10 ever got close to $1 billion in revenue. I thought that would happen to Arista, but it did not. Arista has continued its mission of building good quality products that perform well and has pushed its way to steady growth.

Last week, Arista held its annual investor event at its offices in Santa Clara, Calif. The event was good forum to understand how the company will continue to grow and meet the high expectations of its investors. During the event, CEO Jayshree Ullal share guidance that the company is aiming to cross the $10 billion revenue mark in fiscal year 2026. That pace requires about 20% year-over-year revenue growth, which is hard to accomplish at this level.

Below are my key takeaways from the investor event.

The AI runway is massive and long

High performance is what Arista does best, and it’s exactly what’s needed for AI in both the front end and back end of AI networks. The battle between InfiniBand and Ethernet will rage on in the back-end networks, and Arista will get its fair share of revenue when Ethernet is the choice.

I’m not calling for the end of InfiniBand, but growth for that technology is capped, and Ethernet is picking up much of the new spend. Ethernet will likely dominate front-end networks as it is lower cost, easier to use and, in the front end, performance is on par with InfiniBand. Arista’s Etherlink portfolio is ideally suited for the rigors of AI, and I’m expecting at least five years of growth, likely longer, in AI networking. 

During her keynote, Ullal noted Arista is not only selling high-speed switches for AI data centers but also leveraging its own technology to create a new category of “AI centers” that simplify network management and operations, with a goal of 60% to 80% growth in the AI market.

Arista has its sights set on enterprise expansion

Arista hired Todd Nightingale as its new president a couple of month ago, and the reason should be obvious to industry watchers: to grow the enterprise business. Nightingale recently served as CEO of Fastly, but he is best known for his tenure as Cisco. He joined when Cisco acquired Meraki, where he was the CEO.

Ullal indicated the campus and WAN business would grow from the current $750 million to $800 million run rate to $1.25 billion, which is a whopping 60% growth. Some of this will come from VeloCloud being added to Arista’s numbers, but not all of it. Arista’s opportunity in campus and WAN is in bringing its high performance, resilient networking to this audience.

In a survey I conducted last year, 93% of respondents stated the network is more important to business operations than it was two years ago. During his presentation, Nightingale talked about this shift when he said: “There is no longer such a thing as a network that is not mission critical. We think of mission critical networks for military sites and tier one hospitals, but every hotel and retailer who has their Wi-Fi go down and can’t transact business will say the network is critical.”

Also, with AI, inferencing traffic is expected to put a steady load on the network, and any kind of performance hiccup will have negative business ramifications. Historically, Arista’s value proposition for companies outside the Fortune 2000 was a bit of a solution looking for a problem. But, as Nightingale stated, all networks are mission critical and that’s what Arista does best. The unified EOS architecture is central and provides consistent management and automation across the network. Arista is also addressing specific campus challenges, like traditional stacking limitations, with new solutions like Switch Aggregation Groups (SWAG) that offer a single management point for multiple switches using standard Ethernet cabling.

AI as an assistant, not a replacement of network engineers

Arista president and CTO Ken Duda provided a deep dive into AVA, Arista’s autonomous virtual assistant and what the company believes makes it unique. Duda was crystal clear that AVA is designed to a co-pilot and work with the engineers as opposed to replacing them. The product is designed to be an autonomous agent versus a reactive chatbot with the difference being the latter stays idle and then answers questions when called up, while the former is always running, proactively monitors the network, watches for events, and finds issues before they become critical.

This is consistent with the autonomous agents from other network vendors. What gives AVA its competitive edge is the way Arista manages the network data used to power it. The agent is built on Arista’s network data lake (NetDL), which collects real-time, high-fidelity streaming telemetry from every device in the network, creating a single, historically rich source of truth. AVA leverages this massive dataset to not only answer questions but also to apply advanced pattern matching, predictive analytics, and prescriptive solutions. In the previously mentioned survey, another question I asked was how likely a customer would be to change vendors if the AI was demonstrably better, and 92% stated they would.

The Blue Box opportunity

White box networking has been around for some time and has strong appeal with the hyperscalers. These are much cheaper switches. But you don’t get what you don’t pay for. So, while white-box acquisition costs may be low, the operational costs tend to be high.

Arista’s response to this is its own Blue Box, which doesn’t run Arista’s operating system but does have Arista hardware engineering. The switches have a network diagnostic layer called Netdi, which is an abstraction layer that sits above the basic drivers and can track several switch functions such as cable and optic health, the status of PHYs and other factors that help cloud companies manage these massive scale networks. This is currently aimed at the cloud titans but should have appeal among neocloud providers as well as a handful of large enterprises and telcos.

There’s work to do with the channel

There should be no question about Arista’s ability to grow above market rates with the cloud providers and in AI builds. Enterprise is another question. The company has great products, but its channel strategy is still relatively immature. Arista is a good fit for the Global 2000, which tend to buy direct. But the other two million businesses buy through channel partners. VeloCloud brings a managed service business, and that’s helpful, but building a traditional SI and VAR channel needs to be a focus.

As Arista embarks down the channel path, it will also need to engage in other lead-gen types of activities, such as webinars, customer events, trade shows and other activities that drive awareness, and build a pipeline that can fuel channel activities. Arista does do several webinars and digital advertising for its Campus Builders customer and channel events, but the volume of activity as well as the diversity of content would need to increase.

Also, channel programs tend to be built on discounts and rebates that help the partner make money, and this often comes with sacrificing margins, of which Arista currently has industry leading numbers. Arista has been an execution machine, and as it builds its enterprise momentum, this is an area to watch as an investment here will pay big dividends down the road.

Final thoughts

AI is going to create unprecedented opportunities in networking. In fact, the demands placed on the network are so great that I’m expecting many businesses to accelerate their typical replacement cycles to ensure the performance, security and resiliency required for AI are in place as organizations look to scale their initiatives. Arista is well positioned at the high end and has the products to go after businesses that fall below the biggest of the big.

Todd Nightingale brings a high level of enterprise expertise to Arista, and I’m expecting this to be the starting point for the next wave of Arista growth.