Why Lightning AI, the third-largest neocloud, standardized on Netris for network automation and multi-tenancy.

Lightning AI Fireside Chat: Frank Basso, VP of Infrastructure, Lightning AI & Alex Saroyan, CEO & Co-Founder, Netris Recorded at NVIDIA GTC 2026

---

INTRODUCTIONS

Alex:

Hi. I’m Alex, the CEO and co-founder of Netris. Welcome to Fireside Chat with Netris, where we talk about GPUs, networking, and business. I’m excited to introduce our friend and partner and customer, Frank, who’s a VP of Infrastructure. Shall I introduce you as Voltage Park or Lightning AI?

Frank:

Formerly known as Voltage Park, yes. Lightning AI going forward. I’m Frank Basso. I’m the Head of Infrastructure at Lightning AI. We’re currently the third-largest neocloud on the planet. We’re a premier-level NVIDIA partner in the NCP world. And also we’re a partner on the software side as well, which is unique. So we cover all the bases. We are a full-stack cloud provider, and you could say the first of the hyperscale neoclouds. I’ve been with the company a little over a year. A network engineer and infrastructure guy through and through for many, many years. Happy to be here today and have this conversation. It’s going to be fun.

---

NETWORK AUTOMATION PROBLEM & PATH TO NETRIS

Frank:

When I joined, I went, “This is a perfect opportunity.” What are we using for network automation? Because I’m a recovering network engineer. I spent 30 years as a network engineer, starting in government, and then moved through, always doing network engineering and security. So I’m like, what are we using for automation? Right? And everyone’s like, “I don’t know. We have a bunch of Ansible scripts that we do things and we manipulate the network with Ansible.” I’m like, right, right. But what are we using? Where are the repos? Where are the backups? Where are the things like that?

Frank:

And then one of our engineers said, “Have you ever heard of Netris?” And I said, no, let me take a look at Netris. Oh, this looks pretty cool. And the part I appreciated about you the most was, it’s okay. Doesn’t matter. When you’re ready, give me a shout. No pressure at all. No, no, like, “Hey, when are we going to do a sale?” “When are we going to do this?” And I’m like, okay, these guys are pretty cool.

Frank:

We’re at the highest level of NVIDIA NCP. And the NVIDIA team flew from Israel to come see us and visit us in San Francisco, and they were pitching us on Spectrum and Spectrum-X as a platform. They ran through this two-hour-long thing with all the performance and use, with RoCE and without RoCE, all these things. And they’re like, do you know Netris? Yeah, I talk to Alex all the time. How do you use Netris? Oh, we use it to reconfigure and program all our labs and do all our benchmarking and testing. I’m like, okay. We ended up going with Spectrum-X. So we’re starting with Spectrum-X, we’re doing a small 4K cluster build.

Frank:

And then we signed a deal with one of NVIDIA’s portfolio companies to build 10,000 GB300s with them. And we went, “yes, please.” And then we realized — oh, has anyone on the team ever used GB300s? Has anyone dealt with that InfiniBand fabric before, at that level? By the way, we were a SONiC shop, and now we’re going Cumulus. How many people know Cumulus? One junior network engineer. I’m like, okay.

Alex:

And then NVLink.

Frank:

Yeah, and NVLink. I need to de-risk this deployment. Let me call Alex. NVIDIA spoke highly of you, and that was a good thing. And we checked up on you a little bit. Some people had used your products. We’re like, okay, so we’re going to use this for this deployment, and that deployment is underway right now in Chicago, it’ll be online in another 30 to 40 days. It’ll be online first half of this year.

---

SOFTGATE: SCALE & PARTNERSHIP

Frank:

And then we started looking at SoftGate. And we’re like, we have a customer that keeps breaking hardware, and we don’t want to keep buying bigger and bigger firewalls and routers. How can we handle this? And you led us through the horizontal scale, right? And that customer pushes half a terabyte of traffic as their base. They are also going to burst to more like 880 to 900. You walked us through it on a whiteboard, and we all got really comfortable in that moment. The great part is that you decided to work with us on it, right? You’re like, hey, great, if you’re willing to try this out, and we invest in hardware for it to run on, and it’s a true partnership.

Alex:

For cloud providers, Netris is designed to solve this specific problem.

Frank:

100%. We’ll go API-driven, because you’re API-first, and we’re API-first. We have our own node lifecycle management platform in-house, plus all our other projects that plug in. And it’s just a call for us. We’re really excited about this. And the first — the next act that we’ll do after this one, is that we want to then refit all of our data centers with that at the head end. What if we just wanted to put SoftGate in front? And we walk through it on the whiteboard, you’re like, sure, we could totally do that, and this is how it works.

---

FABRIC SCALE & INFINIBAND DISCOVERY

Frank:

There’s tens of thousands of connections and hundreds and hundreds of devices in every data center. Our smallest data center is ten megawatts. The fact that we’re an InfiniBand shop, and we didn’t even know you did InfiniBand, when we went to talk to you — we were talking North-South, and you’re like, oh yeah, what are you doing to manage all your InfiniBand? And when he said UFM, I think you giggled a little bit.

Alex:

You should not define your tenants in multiple places, right? North-South, and then InfiniBand, and then now NVLink is coming in. These are three fabrics. You want one place to define your tenancy. And you want this one place to be able to talk to all these different fabrics.

Frank:

Yes. And the team is very excited about this, because instead of manual efforts or maintaining different pools of automation that then have to be called from another overlay system, which reaches out to trigger these things — it’s one-stop shopping.

---

CUSTOMER TYPES & COMPLIANCE

Frank:

Because the DPU equation is really attractive. We have a whole range of customers. So we have bare metal: Don’t touch my box. Here’s your cluster. Full isolation. You can’t log into it. But they have full access to the BMC, as well. We do for monitoring management from the outside. That’s a high-end direct training model customer. Then we have multi-tenant within the same cluster. Splitting it up, carving up too and scaling out, you know, rail-optimized rentals. Still on bare metal, where we provide operating system, Kubernetes, and Slurm. So all the way to PaaS layer. Then we have VMs with all those things. And then we have straight-up K8s on VMs, as well.

Frank:

It boils down to: are you sure all the security was applied? Are you sure there’s isolation? And we maintain, we have a SOC 1, a SOC 2, a SOC Type 2. We have ISO 27001. We have HIPAA BAA. We have ISO 42001 — the new AI one. And now we’re going for FedRAMP, and grabbing evidence out of Netris for that, that it was done within the audit capability of it, is going to make those certification renewals a lot easier for us.

Alex:

You got the API logs, you see everything, every action done is logged.

Frank:

Who did it and why? That’s all the evidence they want.

---

LIGHTNING AI MERGER & THUNDERCAT INTEGRATION

Frank:

You know, we merged with Lightning AI. So we’re formerly Voltage Park, and all the farms and that brand remain Voltage Park. But Lightning AI, we’re super excited about this, as well. That development team, about 30 engineers, was asking, like, how do you automate your network? We’re like, well, what you’re going to want to know about is this platform called Netris. And so we’ve actually already added a field within that says “Managed By,” you know, how it’s provisioned, how it’s managed, and the network elements “Managed By,” and now there’s a Netris flag.

Alex:

Oh nice.

Frank:

“Managed by Netris” means Thundercat will call Netris to provision those nodes as they’re placed into inventory. So yeah, we’re already developing against it. So that’s cool.

---

PARTNERSHIP PHILOSOPHY & BUILD VS. BUY

Alex:

You mentioned the collaboration part, the support part. And the feedback part. This is how we operate, and we think that this is the right way of running this kind of business. Myself, my co-founders, we are coming from the operations world. We worked with vendors. There were great examples, not so great examples. We want to be this great example. Being partners with your infrastructure customers, not only is it an amazing way to deliver lots of value to your customer, but it is also a way to learn from working with your customers, also a way to learn and improve your product. Some of these things that you have seen in Netris and you like a lot, are coming from years of us working with other customers and learning with them and improving. There will be some things that we will learn from you, and maybe other generations of customers will benefit from that. But first you will benefit. So this is us bringing this kind of collective experience. When you’re building in-house, you compete with everyone, and you cannot access this collective experience.

Frank:

I would rather partner with a company who is willing to work with me. True partnership, non-transactional. True partnership. We suffer through the hard times together, and we’re both better for it on the other side. And that’s super important to me. And knowing that when the problems happen, when things go south, that I know that I have your cell phone number, and I’ll call you.

Alex:

Of course, and you should.

Frank:

And you’ll answer, even if you’re in the middle of nowhere Asia. That’s true partnership. Every data center I deploy after, yeah, Netris is on the build line. Like that, cost is factored in. What does it take to establish a new data center? Well, Netris nodes and Netris software. It’s in our COGS calculation for the data center. And at first, it wasn’t. It’s like what is this? Why does it cost this much? Why are we doing this? You explain it. And everyone goes, oh okay. And so that’s the way we’re doing it. We have incredibly smart engineers. We built a version of this before, but it’s now getting so sophisticated. And we have disparate things and a lot of drift. Why not use one partner who can handle it all.

Alex:

We’ll get to the rest at some point, but one problem at a time.

Frank:

Yes (wink, wink). Yes, you will.