Exciting Announcement! In celebration of launching our AI Certification, we’re thrilled to offer a 50% discount exclusively. Seize this unique chance—don’t let it slip by!

Transcript

Alan Weckel :
Alright, thanks so much Vishal for spending time with me today .We’re gonna talk a little bit more about yourself, your company and kind of what we’re seeing in the SONiC community. So for folks that are watching, um, that aren’t familiar with you or Aviz Networks,can you give a little bit of background of yourself and the company

Vishal Shukla :
Yes.thank you Alan, for spending time with us, today. My name is Vishal Shukla. I’m, Co-founder and CEO at Aviz Networks. Essentially we are creating, you know, a network software stack,end to end, which can be explained in two different dimensions. The first one is creating networks for AI. The second one, second dimension is AI for networks. So let me just quickly explain, what does networks for AI mean from our point of view. So essentially what we look at is creating vendor agnostic networks, wherein the AI workloads do not matter. They’re run on a data center or they’re running on the Edge the networks for that can be created in a vendor agnostic way. When I say vendor agnostic, may, way means, you can actually put any hardware, and, you know, any kind of a NOS running to it, right So today, you know, we support SONiC, we support Spectrum X, and customers can, you know, essentially, standardize their operating layer. While they can choose the hardware, and the infrastructure, underneath the network. So essentially that’s the first dimension. we have, perfected the recipe for, you know, building the networks in a vendor agnostic way with SONiC. We just announced Nvidia Spectrum X, you know, integration with this operating layer, which we call ONES. So that’s the first part. you know, networks for AI. then there’s a second one, which is AI for networks. So, you know, essentially, in every vertical you see, AI is disrupting the way, you know, the, everyday works happen and networking is, is no difference there, right So what we are looking at is essentially. bringing, AI as a AI network assistant. We call it network copilot. and essentially helping, the network engineers, to, to do their, you know, mundane and routine job, properly, right. So, essentially, you know, introducing AI as an assistant and essentially helping them to, you know, get efficiency, using AI on top of it. So we have products aligned with the second dimension..

Alan Weckel :
Yeah, great. But we’re gonna dive into AI for networking and networking for AI in a little bit. I think it’s important how you frame that. ’cause they really are very, very different, in terms of that’s what they’re doing, Right.

Vishal Shukla :
Yeah. Yeah, yeah, absolutely.

Alan Weckel :
you’ve been really busy. We’ve been talking for a while and I’ve known you for a very long time. but it seems like in 2025 you’ve come out with a ton of new products. Can we kind of dive into what’s been kind of in the kitchen and is now coming out?

Vishal Shukla :
Yeah, so essentially if you, if you really look at it, the vision and the mission always has been to create the, you know, the network software stack.And we started with SONiCs, of course, in 2023 and 2024. We were busy creating the recipe for SONiC, which works perfectly. And is a perfect vendor agnostic, in a way. So 2023 and 24, actually we perfected that recipe. while the mission was always that, you know, we’ll have the end-to-end stack. So 2024, towards the end of it, we released network visibility, portfolio. so that means that we can now see inside the wires the application identification and everything like that. and towards the end of 2024, we essentially launched network copilot, which essentially is using the data, which is coming from, this network operating, you know, layer, which is ONES our network telemetry product. And the network visibility product, you put them both together and now you have essentially enough data which can train the AI for everyday networking. Right. So that’s on the technology side. Of course, a lot of different products have to be made when we talk about end-to-end network software stack, but then also on the business side, right.So the business side, you know, one of the ways to, you know, is essentially put. All these products in perspective go vertically, right So we have seen the adoption of our products set into, you know, to start with e-commerce, retail, healthcare, telco, um, and, and, and, and, and, you know, multiple such verticals.
We kind of look at the market in 15 different verticals. so we’ve been busy basically deploying these products into each different vertical and then essentially replicating it. Right. I mean, if you really, if you really look at it, SONiC is kind of proven, itself, already and now is the point for how you can replicate, that proof point into various different verticals.So there are enough examples, for, you know, the customers to see that yes, it works, right So we’ve been busy not only making our products, but also, proving it into different verticals. So, yes, I mean, it’s been an exciting three months so far, in 2025.

Alan Weckel :
Yeah, absolutely. Yeah. You mentioned the verticals. I’ve seen retail kind of take off recently. Do you see, have you seen the same thing?

Vishal Shukla :
Oh, yes, yes, definitely. I mean, retail has two portions in it. One is the data center, of course, that kind of aligns with, the, all the protocols and usage, which SONiC provides. but then you have Edge. The retail store as such.Right. And, and if you might have noticed, I mean, we launched a group called Pence, you know, it’s a POE Edge Networks with SONiC in the Linux Foundation community last year. And the whole reason for that was to essentially bring in the features for Edge, which SONiC, you know, primarily didn’t have because it was designed for data centers.Right. So essentially now we have the features, which are, uh. You know, required for the Edge. So of course, you know, that’s, you know, some of the reasons on how this is, you know, kind of aligning with the Edge use cases as well.

Alan Weckel :
No, that’s a great example. I kind of see the ecosystem playing out there. Let’s talk about trends, kind of what’s popping up when you’re engaging with customers as it relates to SONiC pain points, etc

Vishal Shukla :
So, um, you know, I mean, we look at six 50 group reports. I mean, that’s, you know, customers also look at it. so, the projections are pretty bullish on, on SONiC.As, as you might know. I mean, you, you have, published that report. But what we see in the market is that we don’t have to explain to customers what SONiC is. In 80% of the cases, customers know what it is. and, and essentially they understand, the, you know, the goodness, which is to provide, the, oftentimes the question, is that how I’m gonna use it, right. How, how can it be enabled, who will support it? how can I integrate in my NetOps layer, right and essentially, you know, it’s open source. So if, let’s say a security bug happens, how it’s gonna be taken care right now, in past two years, we actually have answered these questions and, and created enough examples, so we can actually show it to customers.So all the discussions which happens when we talk SONiC with customers is essentially, around these questions, which I just mentioned. And there is enough data, which we have now to kind of, give examples onto, you know, how to take care of them.

Alan Weckel :
Yeah, absolutely. Alright, um, let’s talk about maturity here.How is SONiC more mature now, or the ecosystem more mature now than say, two or three years ago?

Vishal Shukla :
yeah, so look, I mean, I’ll, I’ll talk that into, you know, three different dimensions. The first one is the protocols, which essentially, uh. The gap of the protocols have,becoming narrower. You know, over the period of time, IP Clause was the first use case, which was running a couple of years back. but then EVP and VXLAN , we have, you know, multiple vendors now supporting that. there’s still, you know, and the Edge, which we have added as Avis Network. Last, you know, last year with the Pence group, there are still some protocols which are not there in SONiC.But then, I like to say that it has, enough, protocols, which can take care of 80% of the customer use cases. So yes, it does not have MPLS and SRV6, but that’s purely Telco, right it does not have in, and when I say it does not have, I’m talking about the community, I. you know, open source version of it.Um, so it does not have like a media and entertainment kind of features, right but that’s fine. I mean, right. I mean you, data center Edge and AI fabric, the features required for that are there. And in the past two, three years we have seen that, not only the protocols came in, but also getting deployed with customers.So that’s one thing. The second one is the security. You know, I mean, there has been this question used to come to us. What happens if A CVE comes in, at least, you know, from the OEM point of view, they will provide you, the update for that PCERT, and everything. So that has improved a lot.I mean, I’ll give you an example of this open SSL issue, which happened last year in July. And we did some research on, you know, when the issue was reported, when it was acknowledged, when it was fixed, and, you know, when the update was provided. So, you know, one of the OEMI was tracking, there is that, you know, they, they, they basically, you know, July 1st when it came in, the fix to the customer was given somewhere around September, October timeframe. So around three, four months later. Right. And then I was looking at SONiC on when that open SSL issue was fixed. So it was July 1st. It was,you know, it was issued, uh. On July 26th, the fix was available and the, the, the customer, you know, is ready for customers to make a build out of it, from July 26th.So you see, I mean, there are two reasons, right for this, fastness, on the security side is that we have multiple community folks who are working on it. and then it is a microservices based system, so you don’t have to kind of do the regression for your code base and everything, and then give this one security fix, right you just touch the piece where this fix is required. So that kind of a, um, you know, a fastness in the security, the enterprise C readiness of SONiC has changed in the past two, three years. So that’s the second thing. you know, the first one was protocol. The second one is this Enterprise readiness, right from the community itself.The third one is the deployment.You know, two, three years back, there were not enough deployment, in various different verticals. and, and the traction. was, was not, that great as compared to what it is today after two, three years, right So we see deployment in, in all different, different verticals, with or without Aviz by the way. There’s enough data out there, that yes, it works. and essentially, any customer who is looking to deploy SO can actually look at some of the examples in their own vertical. And see that, how they have, you know, deployed it. So things have, you know, definitely become, you know, I would say 10 x to 20 x, better in each different dimension in SONiC.

Alan Weckel :
Definitely agree there. We’ve seen that ourselves and the data . more customers You mentioned more verticals. Yes. Kind of. It’s expanding on every domain out there. To me it looks like if we look at additional ecosystems, on top of SONiC, right. It seems like more is forming there. Are you seeing the same thing?

Vishal Shukla :
Yes. I mean, so one of the things with the, with the open source, uh. you know, commercialization and productization. If you look at, you know, Red Hat, kind of started this whole thing in 2022, 2 years back, but back then, I mean, open source was not that, you know, people didn’t know about open source, right.So how they sold it, they created RHEL. Which is a proprietary version of open source and that is how they basically commercialized the whole thing. But in 2024 and 25, open source is essentially a way to go. Any new company which comes in, they first of all look at what all, what all open source is available.They start there. And then they go forward from that, right. So, making a proprietary version of open source is actually an old way of doing things. but then, how do you actually productize the open source while keeping it really open source, right So that’s where this, there’s a layer and an ecosystem around that comes in because customers, they will still look for the experience, which they’re getting from proprietary solutions, right And when I say experiences, it’s not just about support, it’s also about tying it. It goes into their NetOps, tying it into their security system, tying it into their tooling, and operating it, like a clockwork, right. So this, ecosystem and, and in your report, Alan, you say it as a controller, if, you know, in 650 group reports. And what I see is that this controller ecosystem has to be built around, if that has to go into the enterprise, because at the end of the day, customers are looking for a solution. While keeping the network operating system source code open, the solution has to be provided. So the whole operating layer of orchestration, observability, alerts, monitoring network, SLA, 24/7 support as a software, then tying into network visibility, right Tying into security systems. So all of that ecosystem is coming up and we are leading that formation of that ecosystem. from, from a SONiC specific, point of view. And I think it’ll grow, that’s where it is heading towards.

Alan Weckel :
I definitely agree. Um, let’s dive into AI. We can’t kind of get this part in an interview without talking about AI. Let’s start with, um, networking for AI. Yeah. And kind of what’s going on there.

Vishal Shukla :
Yeah. So, I mean, if you see, you know, there are, there are, you have, Nvidia Spectrum X you know, that’s, that’s the ethernet, you know, of all Nvidia solutions. And then there is a second way wherein you have, you know, the GPUs from Nvidia and then you have. You know, the networking vendors, providing the switches, you know, for making that particular fabric. Uh. I think that they both will coexist. there will, there will never be that everything is, is Spectrum X, a hundred percent Deployments in the world is a spectrum X. there’ll never be the, that a hundred percent of the networks are non spectrum X. They both will be there. What I believe is that if you are looking for a non spectrum X kind of a solution, then the standardization, because if you are putting a non spectrum X, let’s say you are doing that right. You are doing that because you want some control, right I mean, that’s why you are doing it, right So if you are doing, if you are doing that to keep the control, then SONiC is the right way to do it because SONiC provides you the right level of standardization and gives you the real control actually. because you can actually now, put any kind of a network under the hood, even any kind of EPO under the hood with, because of SONiC Dash and everything, right So either Spectrum X or, or you wanna put another network, then SONiC is the right way to go. At least that’s what we believe. the problem however, with customers can say that, Hey, how do I operate these two things together? Because, you know, in the networks for AI, there will be a backend network, then there will be a frontend network, right so you need to probably have a single operating layer. For the backend and the frontend networks it does not matter if you are using Spectrum X or you are using SONiC or you are using some other OEM solution. So that’s the problem we are solving, essentially having our operating layer, the product name is ONES integrating with Nvidia Spectrum X. So it can handle that, you know, either the backend or front end. Does not matter where you are using, those, those solutions or, you wanna extend your networks with a SONiC based. So your operating layer will be the same, right.So what I tell to the prospects when I talk to them is that in this world where the infrastructure is changing very fast, right you need to standardize your operating layer, because today you’re gonna use a 400 gig switch. Tomorrow it’ll be 800, then it’ll be 1.6 terabytes, right. And this thing is not gonna stop. I mean, look at Nvidia in a year. They are doing like, you know, the spinning GPUs at a crazy rate, right LLMs are coming at a crazy rate. So at the end of the day, the definition of compute will keep on changing, and because of that network will keep on changing. Right. So, how do you kind of embrace this, this continuous changing, infrastructure? The way to embrace it is essentially creating a single operating layer, which can, you know, embrace this comprehensive, you know, choice of vendor. And, include your own automation on top, right .So whatever ways you wanna do it, you have to standardize the operating layer. And, and under the hood, if you wanna standardize non spectrum X, then SONiC is a, we believe is the right way to go because that actually provides you the choices and the control, which, which customer wants.

Alan Weckel :
I think it’s important to note that both scale up and scale out are gonna be a 10 billion plus market in and of themselves for ethernet.

Vishal Shukla :
Yes.

Alan Weckel :
And it’s kind of cool if you think about it. It’s all greenfield.

Vishal Shukla :
Yes.

Alan Weckel :Right. Like while yes, we’re incorporating it into existing data centers. That’s the great part of having a SONiC conversation. It’s a natural insertion point. ’cause it’s Greenfield.

Vishal Shukla :
That’s, that’s right. I mean, SONiC provides you the standardization which allows you to kind of change. Your infrastructure, as, as it is needed, right. Um, and, and, and it, it basically, you know, provides you a way wherein the supply chain, issues can be taken care of because it’s not choices and control. It’s not just about prices, right. It’s also about that if you want to deploy today, you can deploy today. You are not, you know, dependent on a single vendor to do that, right So, so these are some of the levers, which SONiC provides you, if you are, if you are looking to, you know, get the control of the networks and, and you know, if you are not doing Spectrum X, then I think the SONiC is the right way of doing it.

Alan Weckel :
Great. So that was a great conversation kind of on networking for AI, the infrastructure. Yeah. Let’s flip over to AI for networking.

Vishal Shukla :
Yeah.

Alan Weckel :
How is AI helping networking kind of above the actual infrastructure?

Vishal Shukla :
Yes. So look, I mean, AI gonna be, you know, is actually already disrupting, a lot of different verticals on how the everyday work happens there, right. and, and typically what we have seen. The AI insertion points start with something which is very, you know, very routine and, and mundane. That’s where the AI insertion points happen. I mean, think of it, when Tesla came, with their first car, their first AI function was, you know, the lane assist and, you know, the, the, you know, the cruise control. Now cars have had that functionality for the past 10 years, right radar based, cruise control and everything was there, but they started, proving the technology and getting customer trust, by doing the same thing, which is already being done. but using AI right. So, the insertion point of the AI would be something, where, you know, it can be trusted, by taking care of this routine and mundane job. not intrusive, but something which sits on the periphery of operations. So, that’s what we are trying to solve. you know, by putting AI in networks, things which are required, takes a lot of time. and nobody wants to do that. So things like, you know, checking how much percentage of your networks are in compliance, and it could be security compliance, it could be configuration compliance, it could be, operation compliance, right. and continuously checking that. audit reports. The audit reports, for the security purposes, right. It’s like a hundred page document, PDF document, which kind of has to be made in a certain format, and, and has to have certain evidence attached to it and everything. Right now, generative AI is, is, is great at that, you know, generating documents and everything. As long as you, you, you put the context right, right. Um, so I think this AI, gonna come in networking as well. It just has to start at the right point. The second thing which, which this AI has to be good at is, is essentially being vendor agnostic. because if you really see that, think of a human. Network engineer, right. that network engineer has to learn pretty much all the infrastructure, right it could be Cisco, it could be Arista, it could be Juniper, it could be Palo Alto firewall, Zscaler, load balancers. It has to learn everything, right And then it has to learn all the tools. So AI is no different, right. And that’s why the vendor agnostic is very important. The AI has to understand that this is Cisco, this is Arista, and this is Juniper, and this is Palo Alto. And here are my tools, ServiceNow, JIRA, Zendex, and everything, because this is where the data is. Right. So it has to kind of mimic the network engineer.
And that’s why, like a human network engineer who is vendor agnostic, the AI also has to be vendor agnostic, right. So that’s what we are seeing. A lot of use cases which we are making are from that point of view, then keeping the data, on-prem. That’s another important, you know, part which basically, you have to keep the data. So no internet connection required is one of the things which we are looking at. So you train the model, keep it on-prem, because today your data is kept on-prem, right. So these are a few of the elements which we are looking at, and, and building this AI for the networks for various different use cases.

Alan Weckel :
Absolutely. It’s kind of like tearing down the silos between NetOps, SecOps and DevOps. Yeah.

Vishal Shukla :
I mean, it’s the end of the day. You see, whatever, you, the decision you are making or whatever analysis you are doing, it comes from the data. Right, which essentially the best place to get the data is networks, because that’s where all the data, you know, comes from and, and goes through and, to and fro, use that data to essentially train the AI and then you skin the cat, the, the way you wanna skin it. Right. architects may ask architect kind of questions. The NetOps guys can ask firefighting questions, right Procurement guys can ask for that. Okay. How the network has behaved, because now I’m gonna negotiate with the vendors. So I should know that last time when we bought with some promises. is it, is the promise, did the promises. Hold or not. right So those kinds of things are there. So AI is, you know, you skin the cat the way you wanna skin it, but the data is what it’s required to train.

Alan Weckel :
Great. You gave a lot of great examples there. Let’s kind of ask the question a little bit differently and talk about copilot, right Mm-hmm. There’s a lot of talk in the industry about copilots helping humans. How does that relate here in this conversation?

Vishal Shukla :
Yeah, so, so network copilot, this is essentially our product name. but what it is, is essentially ai, assistant and. And essentially we believe that, there will be, you know, from, you know, a couple of years from now, this will become like a normal, for, for pretty much every network engineer, to have some sort of a copilot, helping doing their everyday job. Right. Now what exactly, you know, the, the, the use cases, it’ll have, how tightly it’ll be bound. will it be cloud-based or will it be, you know, on-prem? Those kinds of things are what we are, you know, discovering, working with, you know, our customers right now. but the idea of what, what, what we think where it’ll live is that it’ll not just be a black box product. It’ll, it’ll not be that. It has to be a product, which can not only tap into the existing. Data sources and everything, which of course, that’s where the AI will get the data, but also can tap into the existing automation, right. So, you don’t have to kind of say that, okay, here is my ai, now ai will do everything. No, that’s not how it’s gonna work. It’ll essentially be that, okay, here’s my AI, and it is, it is trained enough to know where to get the data and what workflows to trigger. In what condition, and that’s where the Agentic AI comes into the picture because, you know, you take the data from one point, you, you do the analysis, what you have to do, and then you trigger different kinds of agents, right an agent could be an existing automation agent, could be a workflow agent, could be, something as simple as crawl the website, get the data, that, okay, what all upgrade parts available for me. Right. So Network copilot is essentially designed for that kind of a purpose, to provide the ai, network assistant with all the elements of, you know, keeping the data private, tapping into various data sources, keeping the community LLM. So, so essentially tomorrow if a new LLM comes in, the performance of that can be passed on to the customers, right and then essentially being vendor agnostic. so that it can work with any different infrastructure. So these are some of the things which we are doing with network copilot.

Alan Weckel :
Yeah. Copilot has a ton of promise too ’cause we have a huge one staffing shortage. Yeah. and then two Right. You know, people don’t wanna do those mundane tasks, so it allows your employees to kind of move to a more fulfilling role. Yeah. And kind of do that. So it’s like a win-win for organizations.

Vishal Shukla :
That’s right. And you know, one of the customers, I will just give you one example. One of the customers was telling us that, look, we have around 20, 30 network engineers in, in our team, and. And everyone is writing their own, Ansible playbook, with a, so there is essentially no standard. So one of the things which AI we want is to just, you know, have our engineers follow the standard, right And, basically just train it. So when we say a sample, Ansible playbook, it should come out the same. Of course the variables, IP addresses and everything will be different, but the way you design the Python scripts and everything, they wanna follow the standard, right So this is one of the use cases, which has, by the way, nothing to do with the, you know, the, running your networks and everything. They’re just looking, the standardization for. For the automation they’re developing. Right. So, and I was surprised by that look, I mean, they’re actually looking at these kinds of use cases as well, which is very interesting, on what all use cases can come out of, you know, ai. We are still learning that.

Alan Weckel :
Yeah, absolutely. It’s kind of exciting. And that’s a great segue to the last question here. We’re rapidly going through 2025. What kind of excites you for the rest of the year?

Vishal Shukla :
So I’ll, I’ll, you know, answer it again in two different dimensions because we look at our execution in these two dimensions. One is networks for AI and another one is AI for networks. So for networks for AI, we’d like to replicate SONiC, into various different verticals. Like I said, we look at our prospects in 15 different verticals. So the idea is to basically have enough proof points in each different vertical, right. again, in a vendor agnostic way, again, in an open source community way. right. So that’s, that’s, that’s one.Then essentially have, you know, the Spectrum X, you know, deployment and media spec, spectrum X deployment. And, and with that, the front end network, with SONiC, right. So that’s essentially on the networks for AI. On the side, we do have recipe. Now is the point to replicate it better, right from the AI for networks point of view, we are looking to essentially make sure that we have. Enough, deployments, again, in various different verticals, which kind of solidify the use cases for what a, you know, AI assistant for networks looks like. right and, and yes, we do have customers that are deploying it, but you need to have enough repeatability, that actually, you know, tells you how AI will come, and, and get deployed, into the networking world. So we are, we are on the quest of that. While we have deployed at multiple places, we need to find out, you know, the repeatability in at least 10 to 20 different use cases. So essentially that becomes a way on how you deploy AI and what, what it is good for, right. So in 2025, of course, SONiC and multiple verticals and, and basically, you know, find the repeatability with 10, 20 use cases across multiple customers. For, for some good use cases, which can actually define this market.

Alan Weckel :
So yeah, absolutely. Yeah. I completely agree. Well, Vishal, always a pleasure talking to you, so thank you so much for spending the time with me today, and we’ll talk soon.

Vishal Shukla :
Thank you. Thank you, Ellen. Thank you very much for your time.

Alan Weckel :
Okay.

Contact Us

AI for Networks, Networks for AI — Vishal Shukla & Alan Weckel on SONiC & NetOps

Alan Weckel :Alright, thanks so much Vishal for spending time with me today .We’re gonna talk a little bit more about yourself, your company and kind of what we’re seeing in the SONiC community. So for folks that are watching, um, that aren’t familiar with you or Aviz Networks,can you give a little bit of […]