Test report DSG-7283 · Rev D · tested October 10, 2026

AI Datacenter InfrastructureDevice under test

Arista Raises 2026 Guidance by $1.1 Billion as AI Networks Scale

Arista raised 2026 guidance by $1.1 billion to $12.6 billion after Q2 revenues crossed $3 billion for the first time, with the AI networking forecast lifted to $3.6 billion.

Read
4 min
Words
740
Node
5nm
Operator
Grace Kim

Spec summary

  1. Arista raised 2026 revenue guidance by $1.1 billion to $12.6 billion, implying 44 percent growth over 2025
  2. Q2 revenues rose 37.7 percent to $3.04 billion — the first quarter above $3 billion
  3. AI back end/front end network forecast lifted by $100 million to $3.6 billion for 2026, about 2.1X growth
  4. Cash and equivalents reached $13.34 billion, up 50.9 percent; customer purchase commitments rose 2.7X to $9.7 billion
  5. CEO Jayshree Ullal sized 2030 scale-across switching at a $15 billion opportunity plus $5 billion in optics
As AI Networks Scale In Three Directions, So Does Arista
Fig. AAs AI Networks Scale In Three Directions, So Does Arista — AI-generated

Arista Networks raised its full-year 2026 revenue guidance by $1.1 billion to $12.6 billion after posting its first quarter above $3 billion in sales, with revenues up 37.7 percent to $3.04 billion in the June quarter.

The forecast implies 44 percent growth over 2025 revenues. Product revenues rose 38.8 percent to $2.61 billion, with growth across nearly all product lines and segment categories. Services revenues climbed 31.3 percent to $430.5 million, reflecting adoption of the EOS network operating system plus CloudVision, the LANZ latency analyzer, Path Tracer, DANZ Monitoring Fabric and Connectivity Monitor tools. Arista no longer breaks out that composite number separately.

Operating income rose 39.7 percent to $1.38 billion, and net income gained 36.5 percent to $1.21 billion — a 40 percent margin on revenues.

How strong is the balance sheet?

Cash and equivalents reached $13.34 billion at the end of the June quarter, up 50.9 percent year on year. Deferred revenues stood at $6.87 billion, up 69 percent. Customer purchase commitments hit $9.7 billion, up 2.7X from a year earlier.

That war chest matters in a tight supply chain. Arista builds revenue backlog, collects some payments up front, and then negotiates aggressively for wafers, packaging and memory — paying more when necessary to secure allocation. The company's own deferred revenue charts show an AI spike in the 2020s that dwarfs the cloud spike of a decade ago.

What does the new AI forecast include?

Arista lifted its AI back end and front end network forecast by $100 million to $3.6 billion for 2026, roughly 2.1X growth year on year. The campus segment is expected to generate about $1.25 billion this year, an estimated 53.4 percent growth.

The AI category now includes scale across networks — the links between datacenters and regions. CEO Jayshree Ullal broke down the split on the earnings call:

  • Scale across: about 30 percent of the AI category, or $1.08 billion in 2026
  • Scale up and scale out combined: the remaining 70 percent, or $2.52 billion

Scale up networks couple GPU and XPU memories inside a node or rackscale system into a coherent whole; scale out networks more loosely connect multiple nodes or rackscale systems to share work.

Ullal sized the 2026 scale across market at roughly $3 billion for switching plus $1 billion for optics — implying Arista holds a little more than a third of this emerging space, consistent with its overall datacenter networking share. Looking to 2030, she put scale across switching at around a $15 billion opportunity, with another $5 billion in related optics on top.

Where does scale up go from here?

Scale up remains a large opportunity, especially for switches built on Broadcom's Tomahawk 6 Ultra switch ASIC. But that market can only grow as fast as customers move away from Nvidia's NVLink and NVSwitch, either by adopting other XPUs or by pushing Nvidia toward a cheaper, standardized approach.

Ullal framed the timeline directly: "Where it's non-Nvidia, Arista will be excited and will be working more closely both in the scale up and scale out — in some cases to build custom racks with their custom processors so that we can better tune our network with the behavior for their inference or training engine. In the Nvidia cases, it's going to take a bit longer."

She drew a parallel with the earlier InfiniBand-to-Ethernet transition: "I feel a little bit like two or three years ago we were talking about InfiniBand and now we don't mention InfiniBand. But it took two or three years to move to Ethernet. So it will take time to go from a proprietary scale-up that's been around a long time with NVLink to these other alternatives, even if Ethernet is really good. But I expect us to do much better there."

Neoclouds, sovereign operators, AI model builders and large enterprises cannot build their own network software stacks and largely buy Arista's tools alongside the switches, since the alternative — assembling SONiC plus open source tooling — is, as the market demonstrates, a hard sell.

On the datacenter side specifically, Arista's datacenter business generated an estimated $2.82 billion in sales for the quarter, up 37.7 percent, with operating profit of $1.28 billion, up 39.7 percent — a 45.4 percent operating margin.

via The Next Platform (Source)

Filed under

  • arista-networks
  • ai-networking
  • datacenter-infrastructure
  • scale-up-networks
  • ethernet
Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering marketplaces and e-commerce at Die Signal.

250 articles

Same lot · LOT-C1C6

« Previous articleNext article »