Featured

Arista Networks: the unglamorous AI winner

If you’ve been following “AI infrastructure,” you’ve probably heard the same loop all year: GPUs, memory stacks, power, more GPUs.

And sure, that’s where the headlines are. But the part people skip is simpler: all that compute is useless if the data can’t move fast enough between servers. That’s where Arista Networks (ANET) keeps showing up.

One line to anchor this: Arista isn’t trying to be exciting. It’s trying to be mandatory.

What changed recently (with the actual numbers)

In Q1 2026, Arista reported revenue of about $2.71B, up 35.1% year over year. That’s not “nice growth.” That’s hyperscaler-budget growth.

Management also pointed to a full-year 2026 revenue target of roughly $11.5B (about 27.7% growth) and lifted its 2026 AI revenue target to $3.5B.

Meanwhile, FY2025 came in at $9.006B in revenue (up 28.6%). So this isn’t a tiny base effect anymore.

What’s interesting is how “AI revenue” is becoming a real slice of the pie, not just a buzzword stuffed into the slide deck.

Here’s the thing Wall Street keeps relearning

At first glance, networking looks like a commodity. Boxes, ports, speeds, feeds. But AI clusters turn networking into something closer to logistics: when the system gets big, small inefficiencies become expensive. Very expensive.

So the expectation game shifts. Investors aren’t paying for “switches.” They’re paying for capacity delivered on time plus software that keeps the mess manageable when you’re wiring together tens of thousands of GPUs.

And yes – that’s why the multiple is high. You don’t have to like it. You just have to recognize what the market is rewarding right now.

A quick detour: XPO isn’t a product launch, it’s a direction

Slight tangent, but it matters. Arista announced a multi-source agreement around XPO, described as a 12.8 Tbps liquid-cooled pluggable optics module with a 4x front-panel density improvement versus 1600G-OSFP optics. Microsoft and Dell’Oro both weighed in publicly.

The practical angle: if you’re a giant buyer trying to pack more compute into a fixed building footprint, density is the whole game. Arista’s materials around XPO talk about needing 75% fewer racks and saving about 44% of floor space in a representative design.

I’m not saying every claim lands perfectly in real deployments. But the direction is clear: the industry wants denser interconnect, and Arista wants to be the standard “plug-in” path instead of a science project.

How Arista actually makes money (the plain-English version)

Arista sells high-performance switching and routing gear, plus the software layer that makes large networks easier to operate. The “moat” argument is mostly about consistency: one operating system across product families, one control plane, less operational chaos for customers once they standardize.

That matters because big customers don’t wake up excited to swap networking vendors. They wake up terrified of downtime. If Arista becomes the default, displacement gets harder over time.

Enterprise expansion: VeloCloud is a real chess move

Arista also bought the VeloCloud SD-WAN business from Broadcom, with the deal effective around July 1, 2025. That’s a direct push into branch/WAN connectivity – basically, “not just the big cloud data center.”

Does SD-WAN magically transform the company overnight? No. But it helps Arista show up in more budget buckets, with more reasons to sell more gear and software to the same customer. That’s the point.

And candidly, it also lowers the risk of being “one market, one customer type.” Not eliminated. Lowered.

The parts I’m not hand-waving away

Two risks that matter more than the average investor wants to admit:

• Supply tightness can bite: Arista has called out constraints across key components (including optics) and the market is watching margins closely because rush orders and premiums are real costs.
• Concentration still exists: even with diversification efforts, a lot of dollars ultimately flow from a small set of huge buyers whose spending can change quickly.

Also, the valuation is not forgiving. When a stock is priced for clean execution, “pretty good” quarters can still disappoint. That’s not a moral statement. It’s just how markets work.

So what do you do with ANET from here (conservative style)

Here’s where I’m at.

If you’re already in: I’d treat it like a hold with standards. Meaning you keep it as long as the growth + margin profile stays intact and the AI revenue target keeps moving in the right direction (not necessarily up every quarter, but not quietly walked back).

If you’re not in: I’d rather start small than chase. A simple approach is 3 buys spaced out (now, after the next major update, and on any broad market selloff). Not fancy. Just honest about the price you’re paying.

One more small note: if you only like “cheap” stocks, ANET will annoy you. But sometimes the bargain is paying up for a business that keeps compounding while competitors scramble.

My tracking list for the next 90 days

• Does the company reiterate the $11.5B 2026 revenue target as the year progresses?
• Any signal that the $3.5B AI revenue target is getting easier – or harder – to hit?
• Margins: are they stabilizing, or still getting squeezed by component costs?
• Customer mix: do we get more evidence of broader demand beyond the usual giants?
• XPO momentum: more ecosystem support, clearer product timelines, or early adoption signs.
• Any meaningful integration updates on VeloCloud.

I’m not trying to make this sound like a sure thing. It’s not. But if AI spending stays elevated, networking is one of those “you can’t skip it” categories, and Arista is positioned like a vendor that gets called first.

Worth a look if you’re building a watchlist for the next earnings cycle – especially if we get a market-wide wobble that gives you a less painful entry.