Nvidia GTC Highlights—Top AI & GPU Announcements Explained

This episode is sponsored by METER, the company building better networks from the ground up.

METER delivers a full stack networking solution, wire, wireless, and cellular, all-in-one integrated system.

From design and deployment to management and support, METER takes care of it all so your business can stay fast, secure, and connected.

Thanks to METER for sponsoring. Go to METER.com-slash-fears-network to book a demo now. That's METER.com-slash-fears-network.

Is NVIDIA just a chip company? What is open-claw? And how big is it going to be?

I'm Diana Gouverts, and this is The Five Nine.

NVIDIA held its GTC event this week in San Jose, so we got together with three of our favorite analysts to get their top takeaways from the show

and learn a little bit more about where the tech giant is headed, and no, they are not just a chip giant anymore.

It seems NVIDIA is trending towards being a systems company. To unpack what exactly that means and why it matters.

We are here with Anchal Sag, Leonard Lee, and Jack Gold. Thank you all for being here today.

Let's jump right in with the first question, which is a round table of your top takeaways. So guys, why don't you give them to me straight?

I would say there are a few. I think because NVIDIA has become such a big company, covering so many different aspects of the IT industry,

I think one, they're very much on the whole open-claw trend. I think they very much are invested there,

and wanting to enable agentic experiences, because I think they realize that if they can get people to individually experiment with agentic experiences, it will grow for them,

and enable bigger and more of that, and inherently agentic applications, or whatever you want to call agents, they're more resource intensive.

So I think that's always a good thing for a company like NVIDIA that wants to sell more GPUs.

But I think the broader takeaway, I would say, is that they're no longer just a GPU company.

You know, it's not just about GPU hardware and software. It's now about, you know, they're selling full racks with CPUs.

You know, they're now selling GROC LPUs, and they're now selling all those things together, plus networking. You know, they're doing CPO.

So they're, you know, they're in the networking business, their networking business is going to probably be bigger than most networking companies revenue, just because of how big their footprint is on a global scale.

And I think that's one thing a lot of people are missing. It's like NVIDIA's scale is so large that, you know, they're talking about a trillion dollars before the end of 2027, right?

So that is a absolutely gargantuan level of scale that every single segment of their business, you know, there's going to be a point where their CPU business will be bigger than some CPU companies.

So I think some people are really not appreciating the scale of what they're doing, but if you look at it, you know, they aren't really selling people individual GPUs.

They're not selling people blades of GPUs. They're not even selling people racks of GPUs. They're selling them entire pods, which are like, you know, half a data center or even the whole data center, right?

So I think that's the thing to understand is that like NVIDIA's value really is most appreciated at scale.

And that's kind of what Jensen's been trying to convince people of, you know, the whole buy more sell, save more mantra that he's been telling.

I'm surprised he hasn't said it this year, or at least at this GTC. He's got other messages to, you know, communicate, but, you know, there were a lot of key takeaways, but I would say that's the biggest one.

One of the overarching impressions that I got out of the event this year was that, and I posted about this, is that NVIDIA is no longer just a GPU company.

And, you know, this is important because that is how NVIDIA has been characterized. Its identity is based on GPU, but we saw throughout the course of the event GTC 2026, is that there's a lot more to what NVIDIA has to do in order to react to some of these shifts that are happening in the AI.

The AI space that Jensen's what GPU can do. And so what we're seeing is NVIDIA become much more of a heterogeneous computing company.

And so if you look at what I thought was one of the more interesting announcements coming out of the shares event, which was the Vera Rack, we're talking about a full CPU rack.

And the way that they were positioning this rack was for the supportive agentic AI, right.

And so as you start to look at what the future of computing looks like, it's probably far less of this, you know, diametric or this, let's say, complete takeover, accelerate, quote unquote accelerated computing.

You're looking at something that's a lot more hybrid and driven off of what will likely be, you know, diverse requirements for compute and a lot of it may not be AI.

Jack, what about you? I know you had a note out on this during the conference and pretty interesting takeaway. I think it related to inference. Talk to me about that for a second.

One of the things that I think NVIDIA showed in this year at GTC. And by the way, it was interesting that they, if you look at the overall amount of content they had of small piece of that was around their chips.

So it really is about becoming a systems company to Leonard's point.

But also it is indicative of the fact that the AI marketplace is moving, it's moving rapidly towards an inference based model and not a model training model, a training environment.

And so NVIDIA made its name on these monster GPUs that you know a kilowatt each or more now inference can't handle it. You can't work without an inference inference is going to be distributed.

It's going to be at the edge. It's going to be in the data center. It's going to be on my device that I carry with me.

It's going to be power constrained. It's going to also be cost constrained. You know, NVIDIA chips aren't cheap.

They cost a pretty penny. And so what NVIDIA is trying to do is is trying to say there's an interesting graph. I can't remember exactly what Jensen titled it, but basically it was by our monster systems because the cost.

Per token generated with our big systems is going to be better than the cost cost for token of a smaller system.

And that's just his way of saying, hey, you know, we want to play in the inference space as well. So it's going to be a very interesting world going forward for NVIDIA.

It's going to be fine, but there's going to be a lot of competition from arm based world from even from Intel to some extent, but it's going to be a really distributed AI world that they may or may not have as much play in.

And it seems like the point that everybody noticed across all of you guys is that NVIDIA isn't just a chip company anymore. They're really kind of positioning themselves as a systems company.

Let's dig into one of those new kind of segments that they're talking. And, you know, we heard announcements at the show around OpenClaw, NemoClaw, NemoTron, it all kind of sounds a little bit sci-fi-esque, but what are those things? And why do they matter?

OpenClaw is obviously something that's not exclusive to NVIDIA. It's kind of an architecture for allowing people to run agents on their own computer and to do it in a way that fits their needs.

Once it's configured to run OpenClaw, you can basically tell it to do whatever you want. And in some cases, it does more than what you want, because you haven't built enough guardrails to control what it's trying to do, which is why NVIDIA, you know, with their NemoClaw stuff has put some protections in place to prevent your OpenClaw implementation from getting over its skis a little bit.

And the whole Nemo naming is really just an NVIDIA branded thing for their own internal brand of open models. So, yeah, you know, they've really pivoted heavily towards open models, as well as kind of deploying their own architecture for open source models.

You know, they're really proud of how many different models they have. You know, they have a world model. They've got the NemoClaw models. They've got models for robots. You know, they have all kinds of models specifically for making it easier to implement things and to do it in an open source way.

This is interesting, because we have to understand how quickly things are moving, and that's not necessarily a good thing. OpenClaw came out, this whole notebook thing, really became a thing just a little over a month ago.

And now, what we see with Nemo, or Neo, is that NemoClaw is what NVIDIA claims is an enterprise-grade version of OpenClaw, right?

And what OpenClaw really is, is an agentic framework, and the purpose of it is really to be deployed on device as a, let's call it a personal AI.

So, there's a lot of companies that have been trying to do this, you know, Apple, Lenovo, Honor, Microsoft, you know, Google, you name them, to create these safe, you know, let's say consumer-grade, enterprise-grade, AI's, right?

And so, now we have OpenClaw coming in as an open framework that was the positive and get hub for anybody to basically download and start messing around with their own agent, now kind of take off, right?

The question is, who's using this stuff, right? Because it's also becoming a huge concern for the cybersecurity community that are very concerned about its use and its ability basically to augment what a threat actor can do, not only within, you know, someone else's system, but enabling them to scale out and do very high precision attacks.

And so, this is an interesting development, I kind of think it's pretty dangerous stuff, there's a big question mark as to whether or not the stuff has actually been hardened and secured in a way that is safe for consumers or even enterprises.

But one of the things that's really interesting about NemoClaw is they introduced this whole idea of, or this concept of open shell, it's still difficult to see how well-formed that is and how well, you know, how consistently it's been instituted, but the thing going back to this is like really moving really fast.

A lot of people still don't even know how to use OpenClaw safely, it's difficult to, I guess, fathom that you've arrived at Enterprise Grade in this time scale.

So it's going to be really interesting going forward given all the excitement around OpenClaw and NemoClaw to see whether this can be deployed safely in a corporate environment or, you know, within organizations or even for consumers.

I want to follow up on a point that Leonard made, which is, you know, has there been enough time, is there enough security built into NemoClaw, right?

Because OpenClaw is not that old, it only came out a couple months ago, has there been enough time to harden that, to mature it, to actually deliver on Enterprise Grade, which is a higher bar than I think most people realize in terms of security and reliability and all other sorts of KPIs.

Totally, I would say if there was any other company involved other than Nvidia, I would say probably not, but if you paid attention to what was happening, the creator of OpenClaw said he was working with Nvidia to fix some security issues.

So they've actually been working together for quite some time, and if there's a company to vet security at an Enterprise Grade, it is Nvidia.

So I would say I'm not as concerned if we're talking in a video-based solution. I think there will probably be vulnerabilities that are discovered. It's inevitable.

But I would also say that it may be time for some companies to experiment, but maybe not deploy.

And also, it's okay to be conservative. I think there's a lot of companies that are conservative, but also if you want to be at the edge, there's a little bit of risk.

But I think part of an Nvidia NemoClaw deployment is the level of security and safety that they are inherently trying to promote for their customers.

Jack, I wanted to see what your take is on what they're doing around systems, not just around OpenClaw, but some of the other announcements that came out of GTC that had you saying, huh, this isn't just a chip company.

They're making a systems play here. What struck you there?

When we look at the market going forward, our prediction is over the next one to two years, 80 to 85% of AI workloads are not going to be training. They're going to be inference. They may be agentic, but agentic looks a lot more like inference than it does training.

And so it's very important that Nvidia has a major stake in the space.

They're doing that in a couple of different ways at the processor level. They're talking about not only having a GPU. They're building their own CPUs.

They're building their own LPUs. This is the rock thing. Rock is really important for them from an inference perspective, from an agentic perspective. They're still working with Intel on CPUs. So that's not going to go away. They're working with AMD as well.

They're going to be working with AMD. They're going to be working with everybody. You don't really care, but they would certainly like to sell their own.

And they're putting a whole lot of effort behind their back end system software. So for instance, on the verse, which is kind of their digital twin world, right.

They're putting together a specific version that actually helps you build out your data center.

They're working with partners that know about electricity and plumbing and building and construction, all that kind of stuff.

And so, and there's going to be other versions. There's one for healthcare. There's one for science. There's one for physics. There's one for all kinds of stuff.

So they're looking at saying, we want to be a full blown womb to tomb systems company. We want to be more like, let's say, an IBM, right, than an Intel, whether they can, whether they can fully get there, they're still going to have to rely on their partners that certainly can't go away.

But hardening, hardening, open cloud with nemaclaw is a good first step to get into the enterprise play. Look, 50% or more of their business is still in the hyperscaler world.

So they've got to put together a strong story for hyperscalers. And so what, what does a systems play mean for them from the hyperscaler perspective.

It's all the above. It's GPU CPU's, LP's, it's software. It's modeling. It's all of that kind of stuff. And so I think what we're going to see from Nvidia over the next two years is a continuation of not reducing their efforts around chips.

Certainly they're going to build massive chips and racks and systems.

But more focus on how do we put all of this together in a big bundle for you?

Wrap it up with a nice little bow. I just want to clarify for everybody on in the audience, it's Grock not that Grock that Elon Musk has. It's a Grock with a queue.

Grock is a company that was founded in 2016. They came out with their first LPU, which is like a language processing unit in 2019.

And it's funny you mentioned IBM Jack because before NVIDIA straight up brought Grock, IBM had actually signed a partnership with them because they recognized that the technology could potentially be a game changer for enterprise inferencing because it can be done at a low cost.

So one other thing that I wanted to touch on that came out of GTC kind of tangentially. It was news that kind of Cisco is working with NVIDIA, specifically NVIDIA's RTX Pro Blackwell series GPUs to kind of create something called the AI grid, which has nothing to do with power.

If you're thinking of the power grid, like I was nope, that kind of completely threw me for a loop. But what they're trying to do is basically create a grid of intelligence and do so using network operators existing infrastructure.

So putting those GPUs out on the edges of the network using Cisco's mobility platform and basically turning that into a web of inferencing.

So I'm curious what you guys think of that because to me, if operators, we already know that Comcast and AT&T seem to be working on stuff around this.

Like could that be the mythical revenue generating thing that they need to really tap into AI as a revenue generator and not just a cost cutter.

I'm kind of curious what you're thinking around that kind of an announcement. Jack, maybe you go first this time.

Let's talk about Cisco for a moment. Cisco, everyone knows Cisco is a networking company. Over the last three years, they've really switched.

They're still a networking company. Don't get me wrong. But where they're really going is they want to be the infrastructure conduit, if you will, for AI across all platforms.

They have a huge security effort as well around AI and they have their own chips and they have their own networking, et cetera.

They also have a monstrous install base. So Nvidia is smart to work with Cisco. Cisco needs Nvidia as well to try to put together a channel connectivity channel, if you will, perhaps.

And so there's benefit to both of those guys to work together. It's really important.

There's another issue that they didn't really speak to very much, which is very important from a networking perspective. And that is as we look at inference and agents, it's no longer about network speed.

It's about network latency.

If you've got an agent running, you can't wait three seconds for it to take an action, especially if it's running a machine tool, you know, physical AI environment.

If you're working in less than 20 or 30 milliseconds, it doesn't work. It is very important that Nvidia has a really strong networking partner to manage interconnectivity, to manage bandwidth latency, but also to manage network security.

Agent security is something, and Leonard hit on this a little bit earlier agent security is something that people take for granted and they shouldn't.

And I mean to your point about Nvidia, having almost half of their customer base be hyperscalers, having Cisco on board helps them not only tap into enterprise, but also the telco marketplace, which we know is become an emerging focal point for the company, right?

Last fall at a different GTC, they announced a billion dollar investment in Nokia. They're working on AI ran. They're working with T mobile on physical AI implementations like telco has suddenly kind of come to the floor as perhaps the next place.

And Nvidia pedals it to GPUs as training becomes less of a focus.

What it means is I think they're self aware. I think they're self aware that, you know, to your point, they're not going to do this alone.

And truthfully, if you look at how Nvidia does things for the most part, they don't really like being the sole partner, sole player in any kind of new market.

They like to work with the companies who are trusted in that space and to scale with them.

I think a perfect example of that is what they've done in the data center, you know, they really have leaned heavily on Dell, HP, super micro, all the companies who are well known in the space to go out there and build these servers for their customers.

And also to service them because you know Nvidia really doesn't have the infrastructure to maintain and service their customers.

And that's why I think it's, you know, especially in telco. There's a need, you know, the GSIs are absolutely necessary to keep things going, you know, running.

So I think that's a very fair assessment of them knowing what they can and can't do.

And that they need to work with partners who are already trusted in the space.

You know, I was out, you know, we actually did a tour of the telco zone yesterday at GTC.

And you know, we talked to Softbank, we talked to Deutsche Telecom, we talked to NTT data, you know, we talked to Booz Allen, who's like their partner in the whole, you know, US thing.

So we have, they have a lot of efforts. I think, you know, they're smart in the sense that they're like trying to do different things with different players that cater to their strengths.

It is really truly a Nvidia plus Cisco effort here. And I did have a chance to chat with Mossom here about this.

And I think one of the things that makes this effort unique and I think it will be a good discovery is the collaboration that Nvidia has with Cisco in secure AI factory.

Right. And so the reason why I bring this up is one of the things that Cisco is focused on is security.

So they have AI defense. They have, you know, all these artifacts that they've, you know, developed over the course of the last year and a half to address security for AI, right.

And so I think this is going to be a good discovery. This will be a positive in that it will give Cisco and Nvidia, especially with AT&T, the opportunity to explore and figure out what are those security capabilities that need to be, you know, developed and implemented it in order to have safe.

Agentic, as well as just gender of AI at the edge. There's huge open questions about how these things will be deployed. Will they be, you know, you know, deployed as containers, right, like applications deployed as containers.

Is it going to be serverless? What do you do in order to isolate agents as well as maybe even functions? How do you deploy and protect data?

Because people just think in terms of logic and the GPU. They don't think in terms of memory. They don't think in terms of storage.

There are all these things that are required in order for you to actually implement generative AI. These are all things that actually have been missing from the talk track oddly.

But we'll be reconciled as Nvidia and others look at AI ran and they start to kick the tires on. What does it take to actually make this stuff safe?

And then what are the economics and the business of AI ran going to look like? And those are things that have not been thought out very well and are largely theoretical and hypothetical.

That's a perfect place to leave me into my last question, which is in 30 seconds or less, what is the one big question that you came away from GTC with that still needs to be answered?

It's not even a scale question because I clearly shown that they can do scale. I think in telecom, I think the challenge is going to be how much of the market will they take?

How much of 6G is going to be powered by Nvidia? How much of the AI ran alliance is going to actually be meaningful?

I think the transition from 5G to 6G is going to be very crucial. And how much of that transition are they going to actually be able to capture?

Now is the perfect time to start that process, but it's unclear how much they're going to end up actually capturing.

The big question for me is with Agentech in particular with Nemo, NemoClaw, which Jensen is characterized as the new chat GPT moment. How are you going to make that safe?

There are a couple of things that I worry about. Number one, and we just had this discussion a little while ago is security.

AI is great, but if you don't have security wrapped around it well enough, it can do some real damage, especially as we move to physical AI. So that's number one.

Number two is what does an AI system in the future look like? And can Nvidia really do that just by themselves? I think the answer to that is no.

So I worry about will they be making the sort of just about acquisitions. It's more about partnerships. It's more like not about Grock. It's more like what they're doing with Cisco.

I think there's going to be a lot more of that to be determined with whom and how soon and what it's going to look like.

Thank you guys so much for your time. I think that's a great place to leave it. And we'll see you again next time. And to all of our listeners and viewers, wherever you are listening,

or watching make sure you like and subscribe, and we will catch you again next time.