Arista deep dive

Hi all,

My second ‘deep dive’, this time Arista Networks.

Although I’m a tech guy, I haven’t looked into networking much, so took a bit longer to (sort of) get my head around the concepts, so includes a bit of a networking primer if thats of interest.

I’ve also added in a quick “Porters 5 forces” section, and would be interested if people think other sorts of analyses could be useful here.

I always find the reverse DCF the most interesting part of the deep dive, ie, what the market is pricing in right now. Be interesting to see what happens next week!

Please let me know any thoughts, and ways I could improve these going forward.



That’s an absolutely awesome post. Unemotional… clear plain language. Really nice work!


1 Like

Great post and summary.

One trend that i think needs more exploring is SD-WAN.

Arista does not try to be a playet in this space and instead touts their tech/sw as a better approach.

A possible reality i see is that growth in the enterprise can be limited due to the existing/massive cisco footprint in almost every large enterprise customer out there. Entire networking teams with ccie credentials support these cisco investments. So it is not as simple as “roll out arista switches and start doing SDN”.

Cisco does a great job protecting its monopoly by controlling the narrative.
When SDN first rolled out, Cisco saw the threat and pushed hard to their proprietary ACI tech. They leveraged client roadshows with partners like redhat openstavk and netapp and touted benefits of ACI.

Now they have jumped on the SDN bandwagon via sd-wan and their acquisition of Viptela.
VMware similarly bought a major player in Velocloud.

This is a great hedge for them…and how they control the narrative:…

I know Arista is crushing it with cloud titans…will be curious to see how growth plays out in Enterprise moving forward.

I dont like cisco as a stock investment…just saying others should be cautious before declaring an old gorilla as dead. Hard to compare cisco revenues too, as they are heavy in security and ucs servers, etc… not apples to apples with Arista.

I would love to see Arista merge with nutanix.



good post.

  1. The explanation under “What is a router?” is really explaining ARP (address resolution protocol) - not really the main purpose of a router. A router allows you to connect a LAN (local area network like your home network) to a WAN (wide area network like the Internet.) A switch can’t. As you explain later in “The point is…”

  2. Cisco has IOS. Arista has EOS. My uneducated guess is that both do the same thing. Can you add a section to explain why EOS is (presumably) going to eat IOS’s lunch? e.g. maybe EOS supports more HW while IOS supports only Cisco’s HW (I assume) - would like to get some education in Arista’s edge here.



EOS is not operated per node, instead it is entirely software base and does basically what Nutanix and VMWare does for servers. It creates and abstracted virtual environment that allows you to manage all switches in your network. There is no one point of failure, and you can update or adjust the software straight from the virtual environment with no downtime, and minimal labor and maximum automation.

In addition Eos is interoperable with almost any third party software you want to pair if with.

Cisco operates in the opposite manner, in a proprietary manner, with software loaded per switch. Making for more difficult maintenance, more difficult upgrades, less automation, more labor, more downtime, and fewer third party interoperability options.

I do not know if this is true or not but Arista has states that Cisco could give it stuff away for free and Arista would still have lower total cost of ownership.

The larger data centers become the more necessary an EOS type structure is necessary as you simply cannot manage switches as individuals. There is a reason why Cisco has declared Arista their mortal enemy and has lost so much marketshare, systematically to Arista and white boxes, that Cisco no longer reports its largest product category (switches) as a separate product category.

Does not mean that it cannot improve itself going forward (which I am sure it is attempting to do) but yeah there is a big difference between IOS and EOS

Here is Arista’s take on EOS. EOS enables Arista to produce router functionality in the switch for a per port cost, from a slide put out last year by Arista at 33x less per port than Cisco charges for its router for the same (or even less functionality). Arista actually has a larger routing table than what Cisco or Juniper offer. That is 1/33 ($3,000 vs. $100,000).

What this all means in practicality, so far, is Arista has moved up to 15% marketshare, more or less from zero just 5 or so years ago against a monopoly and Cisco marketshare has gone down by nearly the same. Arista’s marketshare in 100GB is more than 25%. Thus the puzzlement of 25% growth rate for the rest of the year…

we can only judge the narrative from the results. And forward prognostication seems to be less aggressive than the narrative would indicate (I believe the market thinks Arista was unduly conservative in regard - and could produce another huge drop if this is not the case, like we had last quarter - although some of that has been recovered), but backward results have systematically shown the advantage.



Cisco has IOS. Arista has EOS. My uneducated guess is that both do the same thing. Can you add a section to explain why EOS is (presumably) going to eat IOS’s lunch?

Wouldn’t the fact that Arista has been eating Cisco’s lunch in this area for several years now be adequate evidence for that?


To pile onto Saul’s quote “wouldn’t the fact that Arista has been eating Cisco’s lunch…be adaquate evidence?”

The strategically critical data point is this:

Arista was started by former Cisco employees. It’s CEO was over Cisco’s entire Data Center and Switching team.

One must consider the distinct advantage that Arista holds as former insiders to a company that is 10x its market cap. There is some runway yet to be had for ANET.

I’ll be honest, the leadership team at ANET and their approach - especially that of the CEO - gives me a much higher conviction in the company compared to similar companies with equally impressive results.


Thanks Tinker, appreciate it.

My understanding is EOS is the operating system installed on every switch/router. I believe CloudVision is their centralised switch management system (which talks to EOS).

EOS gives you a unified interface into all of your switches, but at an individual switch level. IOS does this too.

However, EOS is all the things modern process-specific software should be, modular, extensible etc. You can install 3rd party processes on the switch that can talk to EOS, at the same level as an EOS process.

As far as I can tell, Arista went “Cisco sucks. Their switches are expensive, and their software is old and monolithic. Lets make the best switch and the best software, and go after them in the datacenter switch market”.

Cisco is now playing catchup with their IOS XE, IOS XR, and NX OS operating systems.

The very fact that they have these multiple OS’s is big plus for Arista. Supporting multiple operating systems would be an expensive pain in the ass for them.

I’ll add this to the document, quoting you if thats ok?

But your (and Saul’s) point about “we can only judge the narrative from the results” is important. Arista appear to have a significant software advantage over Cisco in the markets they operate in based on historic results.

I’ve altered the “beliefs required to own the stock” section in light of your comments, which gets me to 4.1/5 level.



Thanks Dreamer,

I must admit I got a bit bogged down in the SD-WAN stuff, but you’re right, it is one of the big… something or others. Trend? Maybe…

Jayshree is a bit down on it, from her blog:
“Although we have never promoted “SD-LAN” nor understood the “SD-WAN” hype, Arista has redefined software driven networking and pioneered the convergence between LANs and WANS.”

I did spend a day or so trying to understand it, but really didn’t get it. I don’t get what SD-WAN gives that Arista doesn’t. But I should try again. It may just be marketing spiel, where Cisco can say “Yeah, we do SD-WAN”, and Arista can’t even though capability is the same.


1 Like

A couple of things.

One. East West traffic versus North South traffic. In a virtual network the East West traffic never leaves the virtual enviroment. Or it might never leave the virtual environment.

Imagine an Amazon data center. You can have your server hitting my server. Both of us being customers of Amazon Cloud services. When you server asks for data from my server, because we are both hosted on Amazon Cloud, the request may not leave the virtual enviroment. It would never hit a physical switch or pass through a cable. This is East West traffic. Just because the
traffic never leaves the server center, in fact it might not even leave the box it is hosted in, does not mean it does not need to go through a router. In the case where your server and my server are hosted in the same box, the data will need to pass through a virtual router. I believe this is a strenth of Arista networks.

If I were a network adminstrator, jumping from VMware’s virtual IOS to Cisco’s IOS or anyone else’s would cause massive headaches. If Arista can seemlessly integrate with VMware and seemlessly handle both the North South traffic in addition to the East West traffic, they have a unique place. A place that neither VMware that lacks a robust North/South component. (I am not sure about this, but VMware came out of the server side of the house, not the transport side) Cisco has great North/South technology, but is weak in the East/West virtual traffic. (It seems, I have not been exposed to their technology except in training situations)

Two. White box capablity is absolutly critical. I know that the ability to
work on white boxes is critical. Anyone caught attempting to sell proprietary hardware in the telecommunications industry will not be tolerated. I suspect that Cisco is going to pay lip service to SDN. That will cause significant blow back.