Arista Networks: the unglamorous AI winner
Everyone talks about AI like it is a chip story.

GPUs, memory, power, more GPUs. You know the loop.

Here’s the thing: you can buy all the compute you want, but if the data can’t move cleanly between machines, you are paying for a traffic jam. That is why Arista Networks (ANET) keeps ending up in the same conversations as the biggest AI projects, even though nobody is excited to talk about networking at a dinner party.

And honestly, that is a feature, not a bug.

Quick time anchor so we are not hand waving: in Q1 2026, Arista did about $2.71B in revenue, up 35.1% year over year.

They also pointed to a 2026 revenue target around $11.5B (roughly 27.7% growth) and an AI revenue target of $3.5B for 2026.

For extra context, full-year 2025 revenue was $9.006B, up 28.6%. This is not coming off a tiny base anymore, which is why people are paying attention now.

What matters is the psychology shift.

Networking used to get treated like a commodity. Boxes, ports, speeds, and then everyone argues about price. AI clusters change that. When you connect thousands of machines together, reliability and throughput stop being “nice” and start being the difference between a project that ships and a project that sits there burning money.

So yeah, the valuation is not cheap. It is not trying to be. The market is paying for consistency and for the ability to deliver at scale.

Slight tangent, but it matters.

The XPO announcement is not just a shiny spec sheet. Arista described XPO as a 12.8 Tbps liquid-cooled pluggable optics module with a 4x front-panel density improvement versus the prior optic form factor it referenced. They also floated the kind of space and rack reduction numbers hyperscalers care about when buildings are already full.

I’m not going to pretend every deployment hits the marketing math perfectly. Real life is messy. But the direction is obvious: buyers want denser connectivity without turning the data center into a science fair project.

Another underappreciated angle is that Arista is pushing further into enterprise.

They acquired the VeloCloud SD-WAN portfolio with the transaction effective around July 1, 2025. That matters because it gives them more ways to show up in branch and WAN budgets, not only in giant cloud builds.

Will that instantly rewrite the story? No. But it can make the business feel less like “a few giant customers run the whole show.” Less, not zero.

Now the part I do not want to gloss over, because this is where people get hurt.

• Supply tightness: if key components stay constrained, it can pressure margins and timing.
• Customer concentration: even with broadening demand, a small number of huge buyers can change spending quickly.
• Expectations are high: when a stock is priced for clean execution, a “good” quarter can still disappoint if it is not good enough.

That last one is the sneaky one. You can be right about the business and still have a rough ride if you overpay.

So what do you do with ANET?

Here’s where I’m at. If you already own it, I’d hold it only as long as the growth profile stays intact and the AI revenue target does not quietly drift lower. If you do not own it, I prefer starting small instead of chasing. Three smaller buys spread out is boring, but boring is fine.

And if you only buy “cheap” stocks, this one might never fit. That is ok. Not every good business is a bargain on the day you notice it.

I’ll be watching the next couple updates for one simple thing: does Arista keep making the AI portion of the business feel real and repeatable, or does it start to look lumpy. The answer usually shows up in plain language before it shows up in the headline numbers.