Observing Arista Networks rise by almost 6% in a single Tuesday session while the majority of investors continue to focus on the typical AI suspects is particularly ironic. The business doesn’t produce chips. Language models are not trained by it.
It doesn’t even sell to customers directly. However, Arista’s switches are doing something perhaps more fundamental inside the data centers that power Meta’s social platforms and Microsoft’s cloud: ensuring that all those pricey AI chips can actually communicate with one another.
| Category | Details |
|---|---|
| Company Name | Arista Networks, Inc. |
| Stock Symbol | ANET (NYSE) |
| Headquarters | Santa Clara, California |
| Founded | 2004 (as Arastra) |
| CEO | Jayshree Ullal (since 2008) |
| Market Cap | $168 Billion |
| IPO Date | June 2014 |
| 52-Week Range | $66.59 – $164.94 |
| Primary Products | Multilayer network switches, EOS operating system |
| Key Markets | Data centers, cloud computing, AI infrastructure, high-frequency trading |
| Official Website | investors.arista.com |
You wouldn’t think you were looking at a $168 billion company that has grown to be essential infrastructure for the AI revolution if you were standing outside Arista’s Santa Clara headquarters. There is no ostentatious campus like the expansive Google complex across the street or Apple’s spaceship. Somehow, that subdued presence feels right. While others took credit for the fountains, Arista has always worked somewhat under the radar, constructing the plumbing.
Rosenblatt analyst Mike Genovese, who upgraded Arista to a buy rating with a $180 price target, was responsible for the company’s recent stock increase. His logic was based on a concept known as XPO, or extra-dense pluggable optics, which may sound complicated until you see what it does.
Arista has essentially discovered a way to transfer more data through smaller spaces more quickly than rivals had anticipated. Such efficiency is not merely desirable for AI hyperscalers constructing enormous training facilities. It distinguishes projects that pencil out from those that don’t.
According to Genovese, this technological advantage could propel revenue growth of about 40% per year for the next two years, which is significantly more than what Arista’s own management has been willing to publicly pledge.
This type of bullish call raises the question of whether management is simply acting conservatively or if the analyst has knowledge that management does not. The smart money favors the latter because of Jayshree Ullal’s performance since becoming CEO in 2008.
After fifteen years at Cisco, Ullal joined Arista, and her journey has a poetic quality. She quit the networking company to take the helm of a fledgling startup, which eventually challenged Cisco so successfully that the latter filed numerous lawsuits alleging intellectual property theft.
Arista paid $400 million to resolve those legal disputes, which lasted from 2014 to 2018, while maintaining its reputation and technology. Success isn’t always the best form of retaliation. It’s creating something that your former employer feels sufficiently threatened to file a lawsuit.
Essentially, Arista created speed. Arista switches handle transactions where microseconds count, whether you walk past the trading desks at RBC Capital Markets or into the Chicago Board Options Exchange. The port-to-port latency of the company’s 7150S series is less than 380 nanoseconds, which is so quick that light itself hardly has time to cover the distance of a football field. Early on, high-frequency traders became aware of this, and by 2009, major Wall Street firms with bets measured in millionths of a second accounted for a third of Arista’s clientele.
However, Arista has been drawn in a different direction by the AI boom: hyperscalers, which must link thousands of GPUs into cohesive training clusters. Moving massive amounts of data without creating bottlenecks is more important than cutting nanoseconds. This problem is precisely addressed by the XPO technology that Genovese highlighted, which fits more optical connectivity into typical rack spaces.
Because Arista switches are the least expensive option, Microsoft and Meta are not purchasing them. They are purchasing because the network cannot be the weak point when billions are being spent on AI infrastructure.
But beneath Tuesday’s stock movement lies a more general question. How sustainable is growth that is so closely linked to investments in AI infrastructure? The technology industry has demonstrated a propensity to fluctuate, sometimes within the same quarter, between exuberant enthusiasm and abrupt doubt.
Whether businesses like Microsoft continue to invest in AI at their current rate or choose to take a break and review what they have already created will have an impact on Arista’s revenue trajectory. The stock is currently intriguing because the answer is unknown.
We can see that Arista has successfully managed earlier technological shifts. The company’s Extensible Operating System was innovative when it was first released and has held up well over time. It is based on a Linux kernel with modular agents handling various networking functions.
Arista’s switches are easier to automate and integrate into software-defined networks because they can incorporate standard Linux tools and programming interfaces, unlike competitors who are restricted to proprietary architectures. As data centers evolve from collections of fixed hardware to increasingly programmable environments, that flexibility becomes increasingly important.
Where management sees opportunities can be inferred from the company’s acquisition strategy. Ultra-low-latency Layer 1 switching with port-to-port delays as low as four nanoseconds was made possible by the 2018 acquisition of Metamako. Big Switch Networks was acquired in 2020, adding expertise in software-defined networking. Arista entered campus and branch networking last year when it acquired the VeloCloud SD-WAN portfolio from Broadcom. These are not haphazard wagers. These are deliberate forays into nearby markets where Arista’s fundamental skills are applicable.
Competition, however, never stops. Despite losing patent battles, Cisco is still a formidable force. HPE is now acquiring Juniper Networks, which offers various advantages. Additionally, networking could be brought in-house by one of Arista’s own clients, such as Google or Amazon, which manufacture custom silicon. The fact that Arista has kept in touch with these hyperscalers despite their vertical integration elsewhere indicates that the company offers something that is challenging to duplicate internally. That’s comforting, but it won’t last.
Based on the numbers, Arista’s market capitalization is predicated on sustained robust growth. The current price at $133 is well above the bottom but below the peak, and the 52-week range from $66.59 to $164.94 illustrates how drastically sentiment can change. The accuracy of Genovese’s $180 target is likely to depend more on whether AI infrastructure spending sustains momentum over the next twelve to eighteen months than on Arista’s technology, which appears to be sound.
But it’s not just the AI opportunity that makes Arista appealing. The company was positioned for this moment without necessarily anticipating it thanks to a series of design choices made over the course of two decades. the decision to use Linux for development. the emphasis on programmability. the focus on minimal latency. When Andy Bechtolsheim and David Cheriton started the business in 2004, these were not AI tactics. They were sound networking concepts that also happened to hold up very well.
Observing Arista’s development serves as a reminder that infrastructure companies, while quietly becoming indispensable, frequently lack the narrative excitement of consumer-facing businesses. Modern life depends on water pumps and electrical transformers operating covertly, but no one writes breathless profiles about their manufacturers. Similar to this, Arista is a crucial but unglamorous company that functions best when you don’t think about it.
More information about whether management agrees with Genovese’s optimism regarding growth rates will be available during the May 5th earnings call. Until then, investors must balance what we know—solid technology, solid customer relationships, growing addressable markets—against what we don’t—the sustainability of AI spending, competitive reactions, and macro circumstances that might limit data center budgets. When weighed against a reasonable set of unknowns, it’s not a bad set of knowns.
Arista’s position is satisfying in some way. Not ostentatious, but essential. The infrastructure that makes those stories possible, not the story that everyone tells. The businesses creating dependable, high-performing plumbing may prove to be better investments than those promising revolutions as AI continues its occasionally chaotic expansion. The movement of the stock on Tuesday indicates that at least some investors are beginning to believe that. The only thing that matters is whether they arrive early or late.