NVIDIA GTC 2026 | Sam Khosroshahi, Lambda

Clips
News
More from NVIDIA GTC 2026

Sam Khosroshahi

VP Strategy

Lambda

AI infrastructure is under pressure to keep pace with agentic computing

In today’s market, organizations that fail to invest in forward-looking AI infrastructure risk being left behind as the race to support more complex workflows accelerates.But as demand for real-time data processing and super-intelligence hits a fever pitch, the next generation of applied AI applications, from advanced robotics to large-scale autonomous agents, requires a total paradigm shift. That pressure is prompting organizations to rethink how they build and scale their underlying systems, according to Sam Khosroshahi (pictured), vice president of strategy at Lambda Inc., an Nvidia Corp.-powered AI infrastructure company

play_circle_outline GTC 2026: Jensen Huang’s AI Factories, Lambda’s VR200 at Scale, Liquid Cooling and Photonics for Dense GPU Infrastructure

play_circle_outline Optimizing TCO: MFUs, Performance Tuning, Vera Rubin Chips, CPU Resurgence, and Custom Inference ASICs alongside GPUs

play_circle_outline NVIDIA claim: 40 millionX more compute in ten years; massive scale-up prediction.

Info
Transcript

Sam Khosroshahi, Lambda

Sam Khosroshahi

VP Strategy Lambda

In this interview from the Nvidia GTC AI Conference and Expo, Sam Khosroshahi, vice president of business development and strategic pursuits at Lambda, joins theCUBE's John Furrier to discuss how Lambda is building gigawatt-scale AI factories as demand for purpose-built infrastructure accelerates. Khosroshahi explains why modern AI compute cannot simply be bolted onto existing systems — it requires fully engineered architectures spanning liquid cooling, co-packaged optics and network acceleration where every idle GPU represents lost value. Lambda's research t... Read more

explore Keep Exploring

What is your takeaway from Jensen Huang’s keynote about AI factories—including the point that you can’t just bolt systems on, that performance and energy efficiency matter, and GPUs cannot be idle—and what are you showing here? add

Why does Lambda prioritize testing, benchmarking, and tuning its NVIDIA-powered infrastructure, and how has NVIDIA (and Jensen Huang) helped improve that performance? add

What computing and hardware trends do you see over the next 10 years, and how does Lambda plan to position itself? add

bolt Powered by CUBE AI

Sam Khosroshahi, Lambda

search

John Furrier

>> I'm John Furrier with theCUBE. We are here on the floor at GTC 2026, who has been here all week on the ground, the full team. We had our big events Sunday, day zero, big party after. I'm back with Sam. CUBE alumni, VP of Business Development and Strategic Pursuits. He's pursuing the future of AI infrastructure build out. It's the hottest trend right now. We still need more funding. It's building out an incredible pace. Jensen Huang on stage, talking about the economic value being created from AI. Sam, great to see you again here in the booth of Lambda.

>> Thank you, John. Great to see you as well.

John Furrier

>> So Jensen and the keynote really talked about AI factories and he laid down the gauntlet. This is an economic monetization opportunity for society. Energy is equal to currency. Tokens are everywhere. The future of work is all relying on. You guys are building the infrastructure for your customers. What's your takeaway? What are you showing here?

>> Well, first of all, thanks for having me. And it's a fantastic opportunity for the progression of AI across many different fields and practices. What we are most excited about is actually interfacing with folks like yourselves and our customers that are really building the next generation of applied applications such as robotics and large scale compute for agents and whatnot.

John Furrier

>> One of the things in this keynote that jumped out at me this year, and we always been talking about AI factory for two years now, but was the line of this is now you can't just bolt stuff on. You have to build it out. It's very dense. The performance matters. GPUs cannot be idle. That's unacceptable. And it's the dollars on the table, the energy involved. This is an engineered system. You guys are doing it. What's your reaction to that? And what did you like about it?

>> Well, I love the fact that there is an ambition that is being put out there and it's really raising the bar across every aspect of the compute platform. And so we're talking about taking co-packaged optical and photonics on the backend to accelerate the network and not have the dependency on copper. And you're starting to see the speed of light is coming into fruition. And the scale that Lambda is going to be deploying for VR200, we're very excited about. Of course, this is liquid cooling to chip, and it's accelerating the pace at which you can not only build, train, deploy models. And so you can do a variety of things all in one architecture, and that's very exciting.

John Furrier

>> Sam, one of the things that the NYSE Wired community has discovered, we interviewed a thousand people, AI infrastructure, AI for our biggest successful series. The number one thing that comes out is, "I want to go faster on the build out. We got to get to supply there." But the money making or the value being created is starting to be realized. Could you share your thoughts on some of the value that you're seeing being created while you're building out in real time?

>> Yeah. I think it's an extremely exciting time where we are building a gigawatt scale campuses and we are helping our customers deploy in three key areas. One is the scale, of course. Two is with the expertise to not only deploy chips, but actually have the expertise to support these things over the long term. So we talk a little bit about the final piece, which is like total cost of ownership. And Jensen has done phenomenal work in terms of putting together different aspects of this architecture. And then Lambda has the engineering team that is actually helping customers realize that and day in and day out.

John Furrier

>> Sam, one of the big things that Jensen said on the keynote that was really compelling, at least from my standpoint, was as we entered the Vera Rubin era, the other stuff is just as relevant. And you started to see price performance, the Pareto lines. They all have different use cases. You can have high performance stuff, go to the higher end Vera Rubin, you can move other workloads over here. This is starting to get into policy and tiering. That's a good sign that we're starting to see price performance. Performance continually push the envelope. But alternatives, that's going to change the economics, that's going to change the design. What's your reaction to that?

>> Well, look, fundamentally, Lambda obsesses over this because we have a research organization that works to test out and benchmark a variety of things running on our infrastructure that's powered by NVIDIA. And the second piece is we believe in the performance. We just recently wrote a paper around MFUs that allows customers to go to Lambda and download it and check things out because it's not just about deploying the infrastructure. Like I said, it's about tuning it so that it's driving the performance. Now, Jensen has made it extremely easy for us in three key areas. One, he's bringing CPUs back into fashion now. The Vera chip is something that we're very excited about. We're very excited about the Vera Rubin. And then we're also excited about custom ASICs for things like Inference. And it's all living all together and it's driving that performance around the use cases that you can drive on it.

John Furrier

>> Yeah. You guys are doing a great job performance. One of the things you're showing in the booth here besides all the cool technology is that you have a slogan called Reserve Your AI Factory. You've got region, data center regions all around North America. What does that mean? What does that mean for customers? Take us through that value proposition and how do people engage? How does someone reserve their AI factory?

>> Look, I think first of all, they should call us and they should go to our website. But we are looking for customers that want to experience expertise. Lambda has been in existence for a very long time and it was started by researchers and academics and folks that are really obsessing over scale, but also the nature of the work that exists on top of that infrastructure. So if you're looking back here, it's like you see all of the networking that goes into it, all of the design, and really we're building something for the long term. And so when customers work with us, they know that they're going to get the expertise, the support, and the lowest cost at scale.

John Furrier

>> All right. I got to ask you, since this is one of the busiest shows of the year, it's kind of a bellwether for the industry. You got supercomputing and open compute, other good shows all kind of come together, but there's tons of conversations. I've been walking around in the different lobbies, doing the lobby con. What are some of the conversations you've been having with prospects, customers, and industry? What are the top stack ranked, the top conversations?

>> Look, I think there's a lot of excitement around the photonics area, co-packaged optical. There's a lot of excitement around CPUs that's coming back into fashion, like I mentioned earlier. And one of the things that like Jensen mentioned in his keynote was this 40 million more compute power in 10 years, like 40 million X. And that I think was something that was quite remarkable. And it just shows you that in the next 10 years, nobody knows where we will be, but Lambda wants to be at the center of all that convergence happening.

John Furrier

>> You know, one of the things I've been hearing too on that note is the speed of agentic.

John Furrier

>> Yes.

John Furrier

>> Okay. You're seeing that the data can't wait. Speed into the systems are critical. That was one of those key points on stage. You just can wait a second. AI can't wait. AI needs the data superfast. That's going to power the agentic era. What's your thoughts on that?

>> Look, I think the convergence of this technology software and the workforce is all happening in real time, and we are very excited about it. And we think these investments will pay off over the long term for organizations that really lean in. We focus on deploying the best infrastructure and supporting that infrastructure and aiding customers through our applied AI teams, our machine learning engineers that are here and speaking with customers live, to get them to accelerate their path to Agentic or a variety of other thoughts.

John Furrier

>> Sam, great to share with you. Final question, since you're the Vice President of Business Development and Strategic Pursuits, what is your latest pursuit?

John Furrier

>> So we continue to work with frontier developers, AI labs, as well as the largest organizations in the world. We have a really exciting announcement that will happen if not today, tomorrow, depending on when you're watching this. But yeah, more to come on that. And we're going to continue to expand the types of engagement. For example, we were one of the first to deploy the co-packaged optical, one of the first to deploy Vera Rubin. So these are major announcements and we're at the forefront of what NVIDIA is building end to end.

John Furrier

>> The super intelligence mission that you guys have, we talked a year ago about this on stage that you're starting to see where the rubber meets the road.

>> Yes.

John Furrier

>> We're two years into AI factories. Yeah. Okay. We were going crazy two years ago. This is going to happen. This is so happening and the agentic headroom, if you're not on board with AI, you're going to be out of business. This is the mandate.

>> Look, the current market is one that is constrained across expertise, but it's also constraint around key components of the supply chain, memory, hard drives, all sorts of things, even the chips themselves. And power, of course, how could I forget? So Lambda has made these strategic investments long before this current market is seeing the shortages, and we're very grateful for that. So I think the lesson here is that organizations need to be forward-thinking and need to be able to invest into the future if not .

John Furrier

>> And what should I look at? The integrity of the supply chain, the infrastructure?

John Furrier

>> Yeah. I think everybody is aware right now that it's power. So power is one dimension. And then I think we talked a little bit about the aspects within the supply chain, which is memory, hard drives, but we can't really tell definitively what's going to happen in a few years. So what we do know is the prices of even H100s or Hoppers or previous generation is starting to increase as well. This is just demonstrating that the demand for AI is increasing. And so the sooner you're able to make those decisions and make those investments, the better it is for you long term.

John Furrier

>> Well, great to see you. Thanks for coming on. Do a little hit here. This is theCUBE and the NYSE Wired program, a CUBE Original, the NYSE Wired community, a team of experts, open network, bringing in the future. We're doing our part here on the floor. AI is memory, storage, real-time agents all hitting the scene this year. We'll see a lot more of this. Thanks for watching.