Welcome From GTC And The Setup

Speaker 0:01

Welcome to Uplink, where we explore the world of digital infrastructure, uncovering the technology fueling AI and cloud innovation with the leaders making it happen. I'm so thrilled to be here. We're at GTC in San Jose. I flew in this morning. When did you I did too? Okay, we both got in this morning from New York.

Speaker 1 0:22

Yes.

Speaker 0:22

I've come in from Australia, so I've got you beaten slightly. Um so what we've got with us today, Renen, who is the founder and CEO of an incredible company called Vast. And it's uh it's a rare opportunity to have a conversation. I wanted to chat about a whole range of things, and Renen said he's happy to chat about anything, so we'll see where this conversation goes. Firstly, this is the overnight success story, 10 years in the making. So Vast has literally come out of the gates and has just become so famous in the last probably few years, but you've been at this for 10 years. And prior to that, you had an incredible journey at Extreme I.O. where I know you took a company from zero dollars uh I think in revenue to three billion in three years. So it's not your first uh have go at actual storage. And maybe tell us a little bit about the journey from Extreme I.O to VAST. And then I've got a whole range of questions I'd love to ask you about some of your philosophy around what you've built, and then lead into uh sort of where we're headed from an AI perspective, neo clouds, you name it. So maybe we'll start with the you're at Extreme IO and you decided to do something different, I guess.

Speaker 1 1:30

Yeah. Well, Extreme I.O. was acquired uh into EMC. And so that $3 billion, we had a lot of help from uh the at the time the best sales force in the world.

Speaker 1:40

Um so it was acquired before. So you built the product and then came into EMC and then EMC scaled that through the machine.

Speaker 1 1:48

We started to build the product. When we were acquired, it was two years into it. We didn't actually have a full product yet. It took us another year within EMC to go generally available.

Speaker 1:59

And then crushed it.

Speaker 1 2:00

And then, yes, that first year was more than it was Flash. That was the it was.

Speaker 2:04

And back then, that was pretty different or rare at the time.

Speaker 1 2:09

It was new. Yeah. Um hard drives were the predominant spinning discs. Yes. And Flash was new. It was a very small system. It was just for VMs and databases. It was a very niche type of product. Um, but it was the early days of of all flash systems.

Speaker 2:27

And so Flash was born, so to speak. And at the time, was it performance or was it just that Flash had just come into existence and here's this new thing?

Speaker 1 2:36

So Flash existed for many years, but not in the data center. USB. Exactly. It was a consumer um technology, which is weird. Usually you come from the top down. That's true. Flash went bottoms up, and uh, but suddenly it made sense uh because of performance reasons, you start using Flash in the data center, and Extreme IO was one of those first systems.

Speaker 2:59

And so then you this is the question that I was thinking. You obviously had this sort of contrarian idea that at the time must have seemed crazy, but today is just so perfectly positioned you for this insane success that you've seen. What was the crazy idea that you had? I w was it back there that you had that idea, or did you sort of step out before?

The 2015 AI Bet On Data

Speaker 1 3:22

Or it was back then. It was if you think 10 years back or even 11 years back into 2015, um, it was AI. And it was very, very early AI.

Speaker 3:34

That is early. Yeah. To have that even belief that that was going to turn into anything.

Speaker 1 3:39

I didn't know it would be anything like this, but I thought if we can help build the thinking machine, even if it's a very low probability of success, and even if it never actually materializes, it's worth a shot. 2015, Google just acquired DeepMind. Yes. And so it was very, very early days of uh neural nets starting to show promise. But it was even then pretty clear, at least clear to me, that you needed very fast access to a lot of data in order for this to work. And I looked, I was a part of EMC, I looked at all of the different systems that we had in-house. None of them were scalable enough, none of them were performant enough to um enable this next revolution.

Speaker 4:24

So the theory was that AI is gonna have to access so much more data so much faster or s across a much bigger landscape. Yes. And somehow you've got to feed that up to the machine. That's right. And I assume old models were just not built that way.

Speaker 1 4:40

Um No, there was always a very clear trade-off between performance and capacity in the worlds of storage, in the worlds of data, and you had many different types of systems and different tiers. Yeah. And there wasn't one that could do both. And it was again, um for these new workloads required to have not just access to a few numbers, but to pictures and genomes and video and sound, and not just uh to a few CPUs doing analysis, but to now we know it's GPUs at the time that wasn't even clear. No. But today we have clusters of half a million GPUs and they need very fast parallel access to all of this data.

Speaker 5:23

Yeah. I mean the the world has changed so much, but it's almost like the last two years has changed so much. And for you guys, it it feels like you've I mean you you've come out as the as the winner here, um, because you've built something ten years ago with this vision to get to this point, and it's sort of just all exploded. Uh yeah.

Speaker 1 5:44

I think we were very, very lucky to start when we did because the one hand we could see this revolution begin and so we architected for it.

Speaker 5:53

Yeah.

Speaker 1 5:53

On the other hand, we were um a little bit ahead of when the actual revolution happened, and so we were ready when it did.

Speaker 6:00

You're out there waiting for the wave to come. Yes. Exactly. Is it coming and then a tidal wave?

Speaker 1 6:05

Yes. I'm not a surfer, so I can give those analogies, but I'm sure you're a lot better than that.

Neo Clouds And A Rebuilt Stack

Speaker 6:10

Yeah, there's a big wave that's coming. Um we were just, I was at the um NVIDIA keynote this morning, and you know, Jensen's up on stage, and he was just, it's just so crazy how different the the spend has been. Um and then the two things that I noticed, one is he got on stage and he said, you know, this time last year I stood on stage and I said to the audience that I think I can see half a trillion dollars, so you know, $500 billion of spend in NVIDIA in this coming year. And this year, so he's like, um, it's a year later, I can actually see a trillion. So he's actually doubled what he did last year. Uh last year, but between the previous year to last year seemed just unbelievably insane. And now what he's he's actually doubling that again. And he's saying that uh, you know, from an inference perspective, it'll just keep going. And and all the data that he's put forward just, you know, it's mind-boggling teraflops and all these numbers and letters and what have you. But yes, um, he's executed on it last year, it's happening again. And the next slide that I loved, which I thought was really, really interesting, was the percentage breakdown of spend between hyperscalers and neo clouds. And the fact that if you rewound three years ago, I would say, and if you told me that this neo-cloud phenomenon would even occur, then anyone could even come close to disrupting AWS, Azure, GCP, or even getting just onto the stage, I would have said just not possible. I just wouldn't have believed it to be actually doable. And here's 40% of that chart. Now, tell us about your play in that and how different that is to a hyperscaler. Yeah, and and how it's changing literally everything we know about tech.

Speaker 1 8:00

I think Jensen framed it very well uh a couple months ago when he was in Davos when he said it's a five-layer cake. Yeah. And it's a new stack. All of the parts of that new stack need to be redone. And that's what A is enabling these neo clouds to succeed because they're building from the ground up, they're building a stack. Yeah. Yeah. New power, new chips, new software, putting the models on top and enabling training and inference and fine-tuning in ways that the old clouds aren't able to do this quickly.

Speaker 8:32

Yeah. And which seems like unbelievable that they couldn't have done it as quickly.

Speaker 1 8:38

Yeah. Yeah. Well, I think they didn't notice, and they've gotten so big that it takes them time to move the ship. Yeah. But I think they're realizing now and they're trying to um catch up. In some cases they'll be successful, in other cases they won't. But it's nice it's nice to see this democratization that's happening. Yes. It was in the realm of three big companies, and now you have we're working with maybe a hundred of these neo clouds. And each one is in a different country, and each one is taking a different spin. And it's really nice to see this diversification happen.

Speaker 9:15

Well, I I was at Cisco for many years and always saw the revolution of cloud coming along. And actually, whilst cloud was incredibly good for three companies, there was most of the other vendors that had been spending their lives building out data centers as an example. You think of EMCs and um uh or the net apps of the old days, the sort of storage, and then the Cisco's and the Dells and the, you know, and the all the service, that all sort of was taken away and put into these clouds. But the clouds ended up building most of their own stuff. So the vendor community wasn't really leveraged anywhere near as much. Whereas today, you you see yourselves, you see all these other players like Cisco's and so forth, all coming to the party and all being a part of this big Neo Cloud play. Yes. So for you, this must be like let's say your revenue without Neo Cloud would be very, very small, but with it is just, I'm assuming it's insane. But yeah. I'm assuming it's a lot better than that previous company you went for to three billion.

Speaker 1 10:17

It is a lot better. And we're doing it ourselves also, which is a lot more fun.

Speaker 10:21

Yeah. And as a private company, yes.

Speaker 1 10:24

Yeah. Um, we've been growing fast, we've been tripling year over year. I think maybe that's the reason why it seems like it's been overnight because as you keep tripling, you st you suddenly get to large numbers. So big. But um I agree with you. It's never been the case before that the same company developed the technology and provided the service. Yeah. And uh for 10, 15 years that was the case with the big three hyperskeepers. And I think now we're shifting back to normal where you have a diversity of technologies and a diversity of service providers, and all of them can continue.

Software That Breaks Hardware Limits

Speaker 11:01

And also in countries as well. Yes. And what's also really interesting about your business is it's software. It is. And you decoupled this uh linkage between software and hardware, which is a very difficult thing, I assume, to do in in storage.

Speaker 1 11:16

Um it's not easy, but I think it's required uh it's that five-layer cake. And so you have power, you have hardware, you have software services, and then you have models and applications. And so we want to be the best in that middle layer. And so we started by building a storage system, and then that was good for unstructured data. Very quickly, we realized we need to build a new database that again scales and performs at the level that is required for these new models and applications. We then built a data engine, which is a way to orchestrate compute across all of these chips and across geographies and what we call the data space. And so what started as a storage system is now trying to fill out that software infrastructure stack.

Speaker 12:02

So you're beyond storage now. Yeah, yeah.

Speaker 1 12:05

I like to call it the operating system for this new AI era because in a way we're abstracting these new applications from the new hardware and filling that mid-scale.

Speaker 12:16

And that's what's really interesting. You can actually operate with your pick your hardware in effect. Yeah. Or partnering with everybody. Yes. Yeah. And that solves a lot of problems. Um one of the biggest challenges is supply chains. Yes. And look, supply chains are really difficult, even with the a decoupling of the hardware. But I'm sure that puts you in a much better position comparatively at the moment, because there's at least choice. People have choice, be that choice because you've got a relationship with the vendor, or there's supply, or you've got some sort of financial arrangement, or whatever it may be. For us right now, it's really just about supply. How quickly can we get the infrastructure? We actually don't care anymore. We just like we need to get the infrastructure out to customers. Yes. Yeah.

Speaker 1 13:01

I think there's a big shortage right now because of this AI boom, all of the data needs to move into flash-based solutions. The flash vendors haven't built enough. And so there's a gap between supply and demand. And like you said, having the choice of whichever one you want, if we are to be the operating system, we need to be working with all of the different hardware vendors. And what I found, especially in the last six to nine months, as this shortage becomes apparent, that the fact that our system is very efficient and that you need a lot less underlying hardware to have very fast access to a lot more information, uh, enables people enables people to leverage us in a way that maybe Yeah.

Speaker 13:46

I remember you had some very clever compression algorithms, or is that the appropriate?

Speaker 1 13:52

It is. Um and data. Exactly. We take all of this unstructured data, pictures and video, and we find similarities across a global namespace, and then we reduce it down based on those similarities.

Speaker 14:08

Which is and it's material. It's it's like on average three to one. Yeah, wow.

Speaker 1 14:12

And that's on top of a very efficient uh data protection scheme that doesn't waste a lot of space, and that's on top of a data placement scheme that allows us to use very, very low-cost SSDs. And so you can get up to 10 to 1, 20 to 1 savings when compared to other systems just by using our software.

Speaker 14:32

Aaron Powell So that and the other thing I remember is that it's one software platform and it crosses all the all the types of storage.

Speaker 1 14:39

To try. Um we provide very uh diverse set of interfaces. And so for storage, we provide block and file and object. Then for database, we provide a SQL interface. For streaming, we provide a Kafka interface. Uh, we have Lambda functions in the system. And so really try to be as standard as we can on the phone.

Speaker 14:59

And one platform across all of that. And I think other vendors you have to choose or pick, and each one you can't cross them over. Correct. Um and that's because you've built that from scratch with that expectation.

Speaker 1 15:11

And we had to because every one of those pieces we build on top of our architecture, and it's the architecture that allows us to build at the scale and performance and resilience that's required. And so if we were to take open source or try and integrate other systems in, it would dilute them out.

Speaker 15:30

It's crazy. It's it's awesome what you've built. That the scale is just mind-boggling. How do you wake up and triple, triple, triple? Like, how do you I mean, as the CEO of a company, yeah. People are aiming to double maximum their business and you're triple, triple, triple. Uh, that's that's it, just it changes everything. Your supply chains, your your hiring, your staff. What have you got staff-wise?

Speaker 1 15:56

We're just around 1200 people which is not a lot.

Speaker 16:00

No. For the for the for the revenue that you're doing.

Speaker 1 16:02

Yeah. We're growing fast. We're growing efficiently also. Um, so we've been generating cash forever, uh, in parallel to growing this fast. Investors would love you. Yeah. Yeah, they're not complaining, but it's not easy. Uh every six months, it's a different company. It's a different company, yeah. And we need to keep redoing our processes and figuring stuff out over and over again. Yeah. It keeps it interesting. It's a lot of fun.

Speaker 16:30

Oh, that's so cool. So you're a private company. You probably hasn't had to raise much if you're if you're actually pr spinning off cash. Yeah. Um, so your investors are going to be loving you. Do you stay private? What's the long-term plans for the business?

Speaker 1 16:43

That's a good question. I think as we get to a certain size, it makes a lot of sense to be public. Yeah. Especially now as our customer base shifts from neo-clouds and frontier model builders into very large enterprises.

Speaker 16:56

Yeah.

Speaker 1 16:57

They, in some sense, expect us or have more confidence in us if we are public.

Speaker 17:03

Yeah, they do prefer a public. But there's plenty of big private companies.

Speaker 1 17:08

There are.

Speaker 17:09

So I don't I think you could make your choice either way there.

Speaker 1 17:12

We're we're not in a rush to be public. I don't we're definitely not going public this year, but we have been preparing.

Speaker 17:18

Yeah. Uh two years ago.

Speaker 1 17:21

We're trying. Um it's getting to be very big numbers, and I expect that as we get to bigger and bigger numbers, that uh crypto will double.

Speaker 17:30

Yeah, that's fine.

Speaker 1 17:31

It's not happened. In fact, we grew faster last year than we did the year before. Oh my gosh. And so growth has been accelerating. Accelerating growth. And it's all because of this AI wave. Um, everybody's realizing that the old stack isn't good for these new applications, and that's driving them over to us.

Power Space And Supply Bottlenecks

Speaker 17:50

And the demand is, I mean, people can't actually deliver enough. So even though you've triple, tripled, tripled, you're landing with a lot of the neo cloud providers. Yes. Announcements came out today, I saw as it um Nebius just added um some crazy amount just for one customer being better. Uh, but none of them can actually fulfill enough. No. And I think if Jensen could make more chips faster, we'd still not be able to give enough, which is why I think his his prediction for the trillion, and that's just his revenue, everyone else scales with that. You guys will flow through with that. Yes. And it is not slowing down. The only limiting factor seems to be just the power and space. Um maybe the chips, maybe how many chips they can get out. Yeah. Yeah. It's all physical.

Speaker 1 18:36

It's hard to build a fab, it's hard to build a power plant, it's hard to do these things that are required. But as everything shifts over to AI, you're gonna need a lot more of this. Yes. And especially as you see these multiple exponents one on top of the other. In our world, the world of data, they're building these AI factories, and then data starts to accumulate in each one. Yes. And nobody wants to delete anything ever. No. And so it just keeps growing and growing and growing. Yes. And now we have all of these.

Speaker 19:06

It's funny because it gets sticky. It's like we always said uh data is like gravity or wherever you're storing it. It's like it's so hard to move.

Speaker 1 19:14

Yes.

Speaker 19:14

And you've just landed all of this. You probably haven't even had the net retention effect of the expansion of these things having been landed, installed, and expanding. You're probably gonna get that. It's starting. Depends on when when who we're probably put in, you know, like a core weave or something may have gone earlier. Yeah, yeah.

Speaker 1 19:31

Our net retention is uh in the 200% runs. And so it's growing fast.

Speaker 19:37

But I I think you just that's even gonna wind up because there's so much more opportunity on that. It's all these mind-boggling numbers. It's so rare because we've never had a period in, well, certainly in my history, where I've seen anything scale or grow like this. Cloud was fast, but nothing comparative. This is just so much. I mean, they're almost building bigger clouds in the space of a few years. And if you think about the amount of power these things require, it's obliterating anything that was ever put in place, you know?

Speaker 1 20:09

Because they have to. Everybody's realizing that when you make an infrastructure investment, it's for five years, it's for seven years. And nobody in their right mind will say that they're not going to do AI in the next five or seven years. Yeah. And so everybody needs to move that infrastructure as fast as they possibly can.

Speaker 20:27

Aaron Powell Which and comes with the power challenges, the cooling challenges, that just how they get the power to each rack. We've got so many data centers. I mean, we we live in so many data. We're in a thousand data centers for the network side and 26 for the compute side. And um, trying to get someone that can deliver you over about 17 kilowatts in a rack starts to get hard. 2020 is like, oh, you're pushing the boundaries. And here we have 75, 150, and now Jens is talking about like a megawatt in one rack. In a rack.

Speaker 1 20:57

Yes. You have to build it from scratch.

Enterprise AI And The Adoption Gap

Speaker 21:01

Not that you have to throw it all out. It's in it's just like insane numbers. Uh it's it's it's so impressive. All right. So um you get to see the most interesting AI companies. A lot of enterprises still trying to figure out what to do with AI. And and the reason I'm so bullish on the future is like we haven't scratched the surface yet. The customers we serve for AI are all these AI companies building AI. And they're just going crazy with it. But if you look at the enterprises, they're still going, I haven't figured out what to do with that. You see all these different model providers who are building sort of a private model for the enterprise. And they're getting some traction, but just not what I think will any anything like what we will. And I look at my own company, like we haven't replaced all the teams inside where they are or the structures or anything yet. Yes. Um it's landing, I see it, and I think the world sees it in SaaS platforms that have leveraged it well. And a great example is like Claude for all of our developers. Yes. Everyone is living and breathing in AI. But what are you seeing really interesting? I mean, everything's so interesting, but what are some of the interesting companies you're getting to see? I mean, you talked a little bit about genomics and all these things, which are literally, this is the stuff that will really change the world. Because I think we've we haven't yet changed the world with it yet. I think we've got something that will hopefully solve cancer and all these sorts of things.

Speaker 1 22:25

Once the AI gets to a level of intelligence that it can solve those hard problems, I think we're still a ways away from that, but it's going in the right direction. Um, the reason I think the enterprise hasn't adopted it yet is because the applications aren't there. Yeah. And the enterprise doesn't build their own applications, they need somebody to build it for them. Agreed. Yeah. And the first application, of course, was Chad GPT, Chatbot. Yeah. And now we know that AI is possible. And now we need this operating system layer in order to make it easy for everybody else to build new applications. Yeah. Software is the first one beyond Chad GPT because software developers are looking at themselves and trying to make their own life easier. Yeah. But I think everything will get disrupted with this new technology. There's not going to be we're seeing lawyers and accountants and we're seeing uh graphics people and every single industry.

Speaker 23:23

Yeah, medical industries. Yes. Yeah, there's it's it's it's almost harder to think of something that won't be disrupted. And the the the next step, the problem is, because I would have said, oh well, it's you know, um, there's a great sign I saw driving down the street and said, Chat GBT, build me this building. And it was sort of the statement to say that builders are still required. But for now. But when we get to robots and when we get the AI, this sort of loop robots will build more robots and they need more power, they'll build a power plant.

Speaker 1 23:57

And there's not anything left that maybe we will be able to do better than the AI. We'll still do stuff, but yeah, they'll be better at it than we are.

Speaker 24:07

So what what do we do? What does the future hold? Um I'm we're a publicly traded company, and a lot of our investors asked us because we had this big SaaS crash and they called it SaaSpocalypse and sort of the end of the world and all the rest of it. Um so everyone's very worried about what's gonna happen, everything's gonna get disrupted. And I think it's funny because um this SaaS pocalypse happened about four weeks ago. Yes. And the funny part was that um investors sort of came forward and said, Oh my gosh, this AI could disrupt all software. It can. And it's like, well, hang on, can three years ago we knew that? Yeah, exactly. Did we just wake up to this fact then? It's like, um, I thought we knew this. Uh so yeah, I mean, obviously that's it's it's changing markets, it's changing how companies are getting valued, it's changing how people, well, you know, what you know, how people react. One of the interesting feedbacks we had was people are struggling to pick the winners. Yeah. And then they can't pick the losers, or they don't know if they can pick the loser or the winner. So instead of in the absence of being out of the pick either, they've sort of exited out. And so what you've seen is this sort of flight to safety um in investments across the world, which is resulting in also on this sort of SaaS crash or whatever they've referred to it as SaaSpocalypse. Um, and it's funny, that's why we've seen gold go through the roof. People have gone to mining. They're like, well, if all the robots are here, what's left is like mining or what have you. What's your um what's your thinking on where the world heads here?

Speaker 1 25:35

I don't know. I think, well, if AI really can do everything better than we can, then there will be a lot of abundance. Everything will be available to us. Um personalized everything, personalized med medicine, personalized software, personalized Exactly. Um if there's still things that we are able to do better, then I think we'll just have a lot more done. Yes. Uh we will be able to leverage the AI to become more productive.

Speaker 26:03

Yeah.

Speaker 1 26:04

And uh one person will be able to do what a hundred people could do before.

Speaker 26:08

Yes.

Speaker 1 26:08

But there will be a hundred times more things to do. That's because that's that's the way we operate.

Speaker 26:14

And I think that's one of the one of the things I've always seen. And humans have this, we always manage to do this. And whenever something makes us move faster, all we do is turn the treadmill up and run faster. Yes. And so even a great example, and this is very sort of very small, is that in our business, we're probably coding 10x faster than we were. Maybe it's five, maybe it's somewhere in that, but it's it's material. And we have hired more developers because we're doing so much more and it's flowing on to so many more, and we need more and more and more. And even when we're looking at automatically code checking, we're still needing more humans to be a part of this treadmill that just runs 10x faster. Um, and so we have this habit as humans, instead of relaxing because we invented the calculator or a spreadsheet, we don't know what we just run. Yeah. So I I feel like we're just gonna keep running and things are gonna keep exploding in different ways, and we ended up having more and more and more of everything. It's getting super fast as it is. So I feel I feel less um, I think when people would say it's the end of the world and everything's gonna be replaced. I just find that humans have this great way of running faster in so many different directions. I just think that's what will end up happening.

Speaker 1 27:29

I think it's the beginning of the world. Yes, it's a new world that's starting now. Yeah. It'll be hard to recognize the old world 20 years from now.

Speaker 27:36

Well, we can't recognize two years ago. Like honestly, uh if you had to sit there two years ago or three years ago before the neo-cloud revolution and say that you could that that could even be delivered, it just it was it's not possible. Um I I I'm so fascinated that that was built because there's only really one. I mean, Nick Jensen created that, literally created this entire new industry. Um, and in theory, the clouds could have could have probably managed it with their CapEx profiles, but he needed to create diversity in that to de-risk his own position with all the chips that are being manufactured or that will be competitive. And his job now is to just run faster and faster, so it's so hard to catch up. And it's it's already happening.

Speaker 1 28:18

I think NVIDIA is really, really good at building an ecosystem around. We see that here at GTC. Yes. They're elevating all of their partners.

Speaker 28:26

Agreed. And on just one. Yes, yeah. That's a really good um observation. Yes. And then everyone's on board. Yes. And everyone's thinking bigger together. Yeah.

Speaker 1 28:36

And that's what Microsoft used to do. Yeah. They were elevating all of their partners. I remember seeing an interview with Bill Gates a long time ago where he said for every dollar that Microsoft makes, somebody else should be making five dollars. Yeah, yeah. True. That's what's true. Yes. And I think Jensen is doing that very well.

Speaker 28:53

Well, it's working. Yeah. And you guys have captured, I think, 90 something plus percent of all of the AI from what we can see. Um, we've got so much demand coming for your platform. I mean, we're excited to roll your platform out to our 26 different locations coming soon. Um all on demand or d delivered in seconds. And actually, as we've been talking about that with our customers, so many customers are just lining up trying to take more. So it's just this insatiable demand. Yes. But um, it's astounding that the difference between what has been, I think I know you've been around 10 years, but it's still a new entrant in that in that so I guess storage. I know you're more than that, but yeah. It's pretty amazing that you've disrupted this space.

Speaker 1 29:39

Um when you grow on this exponential curve, then it it it happens.

Speaker 29:47

It's wild.

Speaker 1 29:48

Um I've been talking to a a lot of people today. Uh today is the first day of GTC about um just how much capacity uh is being managed on our platform. And it's getting to numbers where we're we're up there.

Speaker 30:07

It's incredible. Um so AI is a good thing. You guys are not even seeing probably the the start of anything to slow. It's just gonna continue to explode. I think so.

Speaker 1 30:18

Um for the next decade for sure.

Speaker 30:20

Decade, yeah. It's amazing. Um how do we build enough power and space for that all? That is probably the only thing that's slowing it down. Can we solve that equation?

Speaker 1 30:30

I think we need to become uh better at building things, building physical things. Yes. For the last 50 years, we've been better and better at building software and not at building physical infrastructure. Um, this is the time for us to figure that out.

Speaker 30:46

That's so cool. And so as you position, are you also going to add um take a s a much bigger step towards all the enterprises and more of a direct sales approach to the enterprise market?

Speaker 1 30:56

Yes. Um If you look at our customer base today, more than two-thirds of the logos are enterprises. Yes. Um, they're relatively more small than because the scale is so much larger in some of those neo clouds. And because the enterprises started uh later with us and it's a land and expand-type motion. But uh definitely we're seeing AI move into production environments within the enterprise. Yes. They need it to be simple, they need it to be secure. Yes. That's why we've been putting a lot of focus on security over the last couple of years. And we're trying to enable them to take advantage of everything that the big AI companies have.

Agentic AI And What Comes Next

Speaker 31:37

And you haven't even scratched the surface on that side. Yeah. That's why there's just so much momentum behind it. Well, it's been an absolute pleasure to to chat with you. Is there anything else you wanted to share? Anything interesting you've got on the horizon? What's something cool you've worked on that you've seen that's uh fascinating?

Speaker 1 31:54

I think all of this agentic stuff. Yes. Um, people think of agents as the new applications. I think of agents more as artificial people. And especially as you start to have physical AI and robots and you can embed these agents within a physical being. There's all kinds of questions around how do they interact with each other, how do we observe. Yes. How do we control, make sure nothing bad happens? Yes. Um, how do we enable them to uh grow and grow at the pace that we want for all of this abundance to really become available? And yeah, that's I think where we've putting where we we've been putting most of our focus leap.

Speaker 32:36

We had um in inside the latitude business, we have CPU, high performance CPU and GPU on demand. It's in 26 locations. It can just spin it up in five seconds. And so people don't have to think about it, they just go and deploy. Um when the memory situation occurred, we started to see you know a lot of folks just go on a rushing. But what we saw the most was the very, very high performative, high-memory CPUs that have been run for these sandboxes, and they've just gone through the roof. Yeah. And then it's only in the last two to three months that we've seen this take up. It's sort of like you're watching it change because it it's sort of so aggressive. Yes. Um, two to three months ago, not much, now just insane. And on each CPU, they can run sort of, I don't know, 500 virtual machines inside the amount of memory or whatever it is. And each one of those, I'm told, yes, is an agent. And each agent is doing something. Yeah. And this has just gone through the roof. And this isn't in GPU. This is a CPU with very, very high memory. So that is something we've seen two months that we never predicted. Um, yeah, three months ago. You know what I mean?

Speaker 1 33:53

And people were talking about it three months ago, but it it takes a little bit of you know, that hockey stick to go up and then suddenly it explodes.

Speaker 34:01

Oh my god, it explodes. Yeah. And there's so many different companies. We it's not like one or two companies, there's so many different ones just turning up, just pumping out these AI sandboxes. And we're actually ordering a whole range of infrastructure just to go and support that um explosion. Yes. Which is sort of a I guess that's the agentic component flowing through, and everyone's figuring out what agent does what job.

Speaker 1 34:25

Yes. Um and you start to have these agents that learn and that they fine-tune their own models. Yes. Each one of them now has a slightly different intelligence.

Speaker 34:34

Yeah.

Speaker 1 34:35

This one is good at this and this one is good at that. And then they will start to interact with each other.

Speaker 34:40

It's like humans in a company almost. Exactly. Um, which is why yeah, trying to orchestrate. The problem with humans is you've got to constantly face them in the right direction, yeah. Make sure they're running in the right direction, doing the right things. Same thing for these agents, I think.

Speaker 1 34:53

Um you'll have manager agents, and they'll you're doing the wrong thing. Exactly. And then they'll disagree with each other or they'll misunderstand each other, and that's where ideas come from. This is fascinating. It'll be very interesting to see.

Speaker 35:08

That was three months ago, wasn't there? So I think if we do this in a year's time, what on earth is going to have changed? It's going to be in just so mind-boggling. It's an absolute pleasure. Yeah, thank you for the partnership with us. We're really excited to launch your products globally and make them available to everyone in seconds. Um, and just the mind-boggling success that you've had is I don't know, that there'd be very few companies in the world that have ever done anything like that. Um, there's a couple that are that we all know of from an AI perspective, but you're right behind them as well. So congratulations, man. It's awesome. Thank you. Yeah. Cheers.