惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Blog — PlanetScale
Blog — PlanetScale
J
Java Code Geeks
月光博客
月光博客
Engineering at Meta
Engineering at Meta
WordPress大学
WordPress大学
Jina AI
Jina AI
小众软件
小众软件
U
Unit 42
云风的 BLOG
云风的 BLOG
Stack Overflow Blog
Stack Overflow Blog
雷峰网
雷峰网
博客园 - Franky
Microsoft Security Blog
Microsoft Security Blog
罗磊的独立博客
宝玉的分享
宝玉的分享
B
Blog
C
Check Point Blog
爱范儿
爱范儿
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
量子位
阮一峰的网络日志
阮一峰的网络日志
Vercel News
Vercel News
酷 壳 – CoolShell
酷 壳 – CoolShell

The Next Platform: In-depth coverage of high end computing

Uncle Sam Awards $2 Billion-Plus To Quantum Companies, But Wants A Cut Oak Ridge Starts Weaving Together A Quantum, Classical HPC, And AI System Stack Dell Bulks Up Hardware As AI Infrastructure Shifts To On-Premises Cisco Wins Over AI Customers With Merchant Silicon And Optics With Its IPO Done, Cerebras Can Get Back To Pushing The AI Envelope HPE Throws VM Users A Lifeline, Unifying Containers And VM Management In Cloud Stack OpenAI, Microsoft And Friends Build A Better, More Scalable Ethernet Compute And Memory Price Hikes Drive IT Spending Way Higher Sometimes, Air Is The Only Way For AI Systems To Keep Their Cool If You Can Make A Compute Engine, You Can Sell A Compute Engine Cleveland Clinic Simulates Large Proteins With Quantum-Centric Supercomputing Broadcom Helps CPU And XPU Makers Go Vertical With Compute Microsoft Committed To Doubling AI Infrastructure In Two Years Google Is A Full Stack AI Player, And Is Playing Well AWS Will Be An OEM, Just Like Google And Maybe Microsoft New Google Networks Tuned Up For GenAI Inference And Training Microsoft And OpenAI Remain Friends, Are Looking To Hook Up With Others AI-Driven CPU Shortage Saves Intel’s Financial Cookies The GenAI Battle Shifts From Frontier Models To Agentic Platforms With TPU 8, Google Makes GenAI Systems Much Better, Not Just Bigger Cisco Scales Out Quantum Systems With A Quantum Network Switch The Second Time Will Be The IPO Charm For Cerebras Imagine An Army Of AI Minions Handling Incident Response AI Will Soon Drive A Third Of TSMC’s Business Bechtolsheim & Friends Breathe Life Into Pluggable Optics One Last Time How HPC And AI Digital Twins Accelerate Quantum Error Correction The Embrace Of AI In Design Transforms Cadence And Its Customers Nvidia Brings The Power Of Open Source AI Models To Quantum Computing Building The Imperfect Beast For Enterprises, GPUs Need Virtualization As Much As CPUs Ever Did
Arista Rides AI Scale Out Networks, Moves Into Scale Acro...
Timothy Prickett Morgan · 2026-05-08 · via The Next Platform: In-depth coverage of high end computing

Like so many other IT suppliers to the tech titans, it will not be long before AI networking will be the largest part of the business at Arista Networks. In fact, if the company can get more supplies of switch ASICs and other components, it could happen this year. If not, the cross over with the core datacenter networking business that has defined the company in its first decade and a half will be eclipsed by AI networking for scale up, scale out, and scale across fabrics in 2027 for sure.

Given the shortages in the IT industry for all kinds of components, Arista has visibility into its hyperscaler and cloud builder customers, and not because these customers are generous with their capacity planning, but because if they aren’t honest about what they need a year early, Arista can’t get the components scheduled and the gear built.

We don’t know what the supply constraints are for the Arista products, but we do think it is telling that Arista raised its guidance for 2026 from $11.25 billion to $11.5 billion and its for AI-related networking from $3.25 billion to $3.5 billion when talking about its first quarter results ended in March. That is another way of saying that the rest of the business is stuck in the mud. Which is why Wall Street was not happy with Arista when the numbers came out.

To get a better picture of where Arista is going and where it has been, we put together this summary by product category a few quarters ago, and we have updated it with the most recent guidance:

And here is the raw data behind that chart:

Here’s the thing, though. Arista has actually turned in good numbers, and it is positioned to benefit mightily from the GenAI boom, particularly as it moves from supplying scale out Ethernet fabrics to supporting the ESUN scale up protocol on switches beginning in 2027 as well as scale across Ethernet fabrics that glue datacenters together for very large scale AI systems, which the company can sell now.

Jayshree Ullal, Arista’s chief executive officer, explained that opportunity on the call with Wall Street analysts thus:

“The advent of ESUN, Ethernet for Scale-Up Networking, specifications allows for increasing or decreasing computing power in a flexible manner with Ethernet to automatically adapt to workload demands. Scale up will be a new entry for Arista in 2027 and beyond, where we will be working closely with our customers to build AI racks with very fast interconnects for co-packaged copper, CPC, or open co-packaged optics, CPO, as well as supporting collectives and memory acceleration.”

“Scale-out or horizontal scaling involves adding more machines to a leaf/spine fabric, moving workloads across multiple servers or nodes or even connecting other elements like storage or CPUs. As you scale out massive datasets, bottlenecks can be resolved with collective and protocol acceleration at L2, L3, and cluster load balancing, all at wire rate. The system must deliver consistent performance without degradation as more nodes participate. Arista is a shining example here with greater than a hundred cumulative customers to date in 800 Gb/sec Ethernet deployments, and we expect the addition of 1.6 Tb/sec in 2027 at production scale.”

As for scale up and scale across network sales, Ullal did not make any specific predictions, but she did say that in the mid-term, scale across would drive at least a third of the company’s AI sales to two thirds for normal scale across Ethernet networking. She offered no guidance on ESUN scale up switch sales on top of that, and said that was really going to not be material until 2027 and into 2028. Ullal did say that the majority of customers getting ready for early trials are waiting for 1.6 Tb/sec ports, but a few customers are experimenting with ESUN atop 800 Gb/sec gear.

As for the supply chain issue that is governing Arista’s sales to a large extent, Ullal said that it started out with a DRAM problem that might be a few quarters, and now it is a wafer and package fabrication problem that is affecting the whole chip industry and it may take one or two years to normalize. Hence the rise in purchase commitments at Arista, which we presume customers are prepaying with deferred revenues, also double compared to this time last year, to help cover those costs and lock in prices now as best can be done.

In the March quarter, Arista had $2.31 billion in product revenues, up 36.6 percent year on year and up 10.3 percent sequentially. That is a pretty good first quarter for any IT supplier in terms of growth. Services revenues rose by 27.3 percent to $397.7 million, up a smidgen sequentially.

Add them up and Arista had $2.71 billion in revenues, up 35.1 percent year on year and up 8.9 percent from Q4 2025. Operating income was $1.16 billion (up 34.8 percent) and net income was $1.02 billion (up 25.7 percent, and representing 37.8 percent of revenues.

Arista ended the March quarter with $12.35 billion in cash and equivalents, up by 51.6 percent since last year and giving it huge latitude in what it does and does not do. Including prepaying for components to make sure it has a fair share of supply.

My model suggests that Arista had $2.52 billion in datacenter revenues in Q1 2026, up 32.3 percent, with operating income of $1.08 billion, up 32 percent.

It is hard to find a company that is expanding into new markets and yet keeping an even keel and does it as well as Arista does.