Name: Analyst Conversation
Uploaded: 2024-10-01T16:00:00.000Z
Duration: 33 min 44 s

Clips
More from VAST Presents: Enter the Cosmos

Jack Gold

Founder and Principal Analyst

J.Gold Associates, LLC.

Dave Vellante

Co-Founder & Co-CEO

SiliconANGLE Media, Inc.

Rob Strechay

Dir./Principal Analyst & Host

theCUBE Research

play_circle_outline Overview of VAST Data Platform, VAST DataBase, and novel lock management

play_circle_outline Focus on real-time data retrieval, governance, and compliance

play_circle_outline Integration of Vector and graph databases, Kafka broker, and private AI

play_circle_outline Fostering AI Innovation: VAST Cosmos Community Among Leaders like Deloitte CTO, Chan Zuckerberg, and NVIDIA

play_circle_outline VAST positioning itself at the center of AI community with big players

play_circle_outline VAST plays in training, GPU clouds, OEM partnerships, and inference markets

31:59

Info

Analyst Conversation

Jack Gold

Founder and Principal Analyst J.Gold Associates, LLC.

Dave Vellante

Co-Founder & Co-CEO SiliconANGLE Media, Inc.

HOST

Rob Strechay

Dir./Principal Analyst & Host theCUBE Research

HOST

Dave Vellante

>> Welcome to theCUBE special live coverage of VAST Cosmos. We're going from New York City to Palo Alto to Boston. theCUBE is pleased to be broadcasting the latest innovations in AI, data and ecosystems that are forming to create the future of intelligent innovation. My name is Dave Vellante and I'll be your host of this Analyst angle. And joining me are Rob Strechay from theCUBEResearch and Jack Gold, founder of J.Gold Associates. Gents, welcome.

Jack Gold

>> Thank you.

Dave Vellante

>> Welcome to theCUBE. Good to have you. Okay, so we heard today from Marianne Budnik kicked off the program. She is the CMO of VAST, Renen Hallak, the CEO. Jeff Denworth came on. He's the co-founder. He got super technical, geeked out big time, which is awesome. We heard about the VAST Data Platform, which has been their historical product. It's a full stack infrastructure layer, the VAST DataStore, which again is the history of the company. But then last year we heard about the VAST DataBase, which was introduced a year ago as part of Build Beyond, which is this parallel transaction database that does analytics. We heard about this novel lock management, it's got full system observability, bridging the gap between structured and unstructured data, disaggregated shared everything. It's just really back to deep computer science roots for those of you who are interested in that kind of stuff, running anywhere. VAST calls this the OS for the AI age, which is kind of interesting. Rob, what's your take on what you heard today? We're going to get into the news, but what's your overall take on the announcement?

Rob Strechay

>> Yeah, I think the two big things that they talked about was really the Cosmos community that they're building out and really aiming at AI and bringing it together. And I think also was really the engine for insights that they're building out that really is about how do you, I guess you could say, move up the stack and continue to move up the stack from a data perspective and be more of that data layer for people doing AI. And I thought it was a really interesting set of announcements. Not to mention that they are going further with some of the database type stuff for streaming data and things like that with Kafka under the hood.

Dave Vellante

>> And of course we heard it from a ton of customers and partners. We'll get into that a little bit. Jack, there was this hilarious, I don't know if you remember, Saturday Night Live skit back in the day it was Dan Aykroyd and Gilda Radner and they were debating whether this new product was a floor wax or a dessert topping. And then the Chevy Chase came in and he said, "It's both." And it reminds me, a reason I bring that up is VAST. It's like, okay, they were a storage company and now they're a data company, they're a database company, they're an AI company, they're an AI infrastructure and operating system for AI. It could have all of the above. What's your take?

Jack Gold

>> Yeah, so forgive my pun, but they're trying to move up to stack. They're trying to get more in the face of end users. Before they were mostly an infrastructure play. No one really noticed that. I shouldn't say they didn't notice them, but they were under the covers more or less. They're trying to move up to a level that says, "Hey, without us, you can't do AI from an enterprise perspective." And that's actually probably a pretty good play for them because if they can move up the stack, one, they can charge more. And number two, the visibility also helps them as they try to go to an IPO in the not the distant future. That's certainly not a bad thing. So I liked the idea of trying to again, move up the stack being more embedded but visible to people as opposed to just being under the covers.

Dave Vellante

>> So VAST is one of these companies that's really focused on solving hard problems and coming at it with novel approaches. So let's try to unpack some of that. One of the things we heard today was LLMs are becoming more intelligent, but they're not real time. They're a big focus on real time and I want to come back to that. And they need access to proprietary data and specialized data. This gets into the long tail, the power law of the cube gen AI that you guys, theCUBEResearch and we put out, but specialized models, maybe small language models, we're going to get into agents, small action models or even large action models. So what do you make about that commentary there, Rob?

Rob Strechay

>> Yeah, I think exactly what Jack was saying, as they move up the stack, they're looking to be the platform where you run the majority of that intelligence and do your embeddings. And so I think what they've been doing is subsuming different pieces that you've seen in DBMS and things of that nature. The embeddings is definitely a shot at Vector and at knowledge graph. And what they're trying to do is be able to connect all of the data underneath there so that you can query it a lot faster from through that NIMs layer that they are now also embedding inside the VAST Data construct using containers. So I guess not embedded, but they're sitting in containers on top of the VAST hardware and operating system doing the NIMs NVIDIA stuff. So I think again, you have to want to use that stack under the hood for doing the embeddings and things of that nature. But what that's also doing is it's not only subsuming some of the DBMS, it's also moving up the layer into that compute engine layer that people like a Databricks or a Snowflake or others have talked about as being where they want to compete. And so I think there's an interesting, I guess, you could say dichotomy and of what is the division of labor and what is that? Where are you going to build your action models? Where are you going to build your agents? Well if that's up at the compute layer and all that doing is calling down into the VAST abstracted insight engine, which is NVIDIA under the hood. Again, you're talking that translation there. I think it's a really smart move on their part for being more important .as Jack was saying, up the stack and being more visible in the AI layer.

Dave Vellante

>> It's interesting you mentioned Snowflake because somewhat Snowflake-like in that the value proposition rests upon putting the data inside of the system so that it can be governed and controlled. And that is the promise that if you bring it in here, we're going to be fast, we're going to be more cost-effective, we're going to be governed, we're going to have all the lineage, et cetera, et cetera. And so it is competitive in that regard, isn't it?

Rob Strechay

>> Absolutely. And I think that where they're going and where they're looking from, the query prompt down is to be that server of the data, unstructured, structured, what have you across that, and to be the ones that index that, that do the embeddings that really, I guess you could say, prepare the data to be utilized for AI. And I think it's very AI-specific and where this infrastructure is built, it's not built out to be multi-purpose in that way, it's really built to leverage that. It doesn't mean that you don't go and put, because they were talking about streaming data and things like that. You could think of building an agent on your website for retail that actually needs streaming data and first-party data that is streaming in. Well, okay, you come in through that at the top of Kafka topics, say that 10 times fast, are assigned automatically, embedded. And then you're using that data in coordination with some RAG. I could see that. It makes a lot of sense what they're doing in that layer as well.

Dave Vellante

>> So let's bring Jack even into this conversation. My first question to you is, Vector, they're building Vector into their database, the VAST DataBase. Where do you land on Vector? Is it a feature or should it be a separate thing? I mean is there enough value to have a separate Vector database? Obviously the Vector database guys would say that, but what do you think about it?

Rob Strechay

>> Yeah, of course. Look, I think where VAST is trying to go is to be the data consolidator. And I don't mean that in the traditional sense where we're talking about a data lake house or whatever. They want to be the central point for me being able to get to my data no matter where it is, anywhere in the world, in any format, Vector, SQL, whatever it's going to be-

Dave Vellante

>> Structured unstructured.

Rob Strechay

>> Structured. Yeah, exactly. I don't care. Video, audio, text. And if they can pull that off effectively, that puts them really at the central control point for all the data I'm going to be using in my AI world. And I think that's really where they're trying to get to. Now can they do it? It's not an easy thing to do, they're well on their way to do it, I think, but they're trying to be the consolidator that allows me to get any data in real time no matter where it's located around the world. Sounds like a really big problem.

Dave Vellante

>> Yeah, I mean if nothing else, VAST is ambitious. Let's talk about RAG because people are excited about RAG. Everybody's talking about it, Dell talks about it, Oracle talked about it in a CloudWorld we heard that and everybody's excited about it. But when we look at the data, Rob and Jack, from our partner, survey partner, ETR, not a ton of people are doing RAG. And those that are, it's limited. I mean it's not of great value, it's complex. And so I'd be interested in your thoughts on, let's start with Jack, how VAST is solving this problem and how will it make a difference to customers?

Jack Gold

>> Yeah, the problem with RAG for most people is just accessing the data to help fine-tune, not fine-tune, but RAG the model, if you will. It's hard to get the data in. It's hard to get it in real time, which is really what you need to do if you're going to use AI in any production environment, it's costly. You may not even know where the data is. So what VAST is trying to do, I think is great, which is I don't care where your data is, just point us to the data will be the consolidator, we'll bring it in, we'll put it into one format, we'll be able to do it in real time. We're breaking down the silos of data into more of a plane if you will, a mesh, that's very valuable for RAG because that's what gets you to the real-time environment. If you can't do this in real time, then RAG doesn't work for most people.

Dave Vellante

>> Well, let's talk about real time. And there's a huge focus in the announcement on real time and the difficult challenges of AI retrieval and they've announced an in architecture vector database. They've got this globally, because they talked about globally consistent, atomic file system, which eliminates the whole batch operations and it's scalable. So real time, we talk about real time digital representation of your business all the time. Rob, what are your thoughts on how real this is, how important it is and can they pull it off?

Rob Strechay

>> Yeah, I think that them being able to use streaming in Kafka to inject into tables on the fly and be able to index it, embed it, and do all of these really, I guess, you could say the heavy lifting of the DBMS for both real time and bulk transactions, that really helps them. I think where people were looking was, hey, how do I get the two sets of data to be coexisting so that when their coexisting, to your point, Jack's point here is that really I'm not pulling all of my data from a RAG which could have a shelf life on it. And I think that's the big piece of what Jeff was getting at as well in his part was there's a shelf life to normal RAG. And you have to be able to version that stuff as well. But at the same time, you need to bring in other data from other systems to be able to coordinate that so that you actually answer the question. Because you may or may not have all of the data domiciled inside the VAST engine, I guess you, could say.

Dave Vellante

>> So let's talk about the news. We saw Renen Hallak, the CEO and Co-founder sat down with Jensen at Jensen's offices and talked about the big announcement here, the VAST InsightEngine with NVIDIA, which unlocks insights from all enterprise data, the world's first solution to securely ingest, process and retrieve all, to Jack's point, enterprise data files, objects, tables and streams in real time. Talked about the first application workflow to run on the VAST Data Platform. So we were told that this could be available or will be available in the first half of 2025. So pretty aggressive schedule here. It's things like this that I think potentially can help us get better ROI out of AI, which a lot of people are concerned about the whole process and retrieval engine for enterprise data, hosted natively inside the VAST Platform. So let me start with you Jack. What's your take on the announcement?

Jack Gold

>> Yeah, I think it's certainly a big deal. It's very NVIDIA-centric because they're working with the NIMs platform, NVIDIA stuff as a microservice. But if they can pull this off, what they're trying to do is they're trying to say, look, if you've got data out there that you're updating once a week, once a month, once a year in batch process, we're going to be able to pull that in real time and you're not going to be just updating it once a year or not have to load it up to a cloud somewhere into a central data lake, a data warehouse somewhere. We're going to be able to build a mesh that allows us to pull all that data in real time, put it into the NIMs system, which is a way of actually trying to do real time updates of AI. And we're going to give you a RAG system if you want to go RAG, there are other ways to do this as well, that will allow you to get the data you need to get the work done that you want to get done. And we haven't talked about cost yet, that's a whole different issue, but in a reasonable timeframe. Sure. So I think it is certainly a big deal for them if they can do this. I'm assuming they can or they probably wouldn't have announced it or Jensen certainly wouldn't have been on stage with them if he didn't think they could do it. I think it's going to be a game changer for these guys.

Dave Vellante

>> So the DataBase, they also announced the VAST DataBase supports of vectors and graphs, so the vector embeddings and then graph database. Also, there's a Kafka broker inside the VAST DataBase, so they're eliminating all the copying and the ETL-ing and all the movement into a data lake. They really stressed, Rob, governance and the compliance. So it's this all in one. It's all there as Jack's been saying, any data, anywhere, encompassed in this platform.

Rob Strechay

>> And this goes back to them moving up stack that we both talked about, Jack and I, around how they're moving more into that layer, even using AI. And like you said, the NeMos and NIMs retriever technology to create those embeddings that solves for that vector database and the knowledge graph, but also the Kafka for the streaming and how that integrates. Really, it reminds me of everybody who's talking about private AI. And when you start to look at private AI and the easy button to get to AI, this could be a big accelerator because what you're then talking about is, Hey, I'm bringing in some other NVIDIA kit on top of this that's running my PyTorch and other stuff and my AI agent layer you could say. And then those agents are making those calls in say, today, SQL, and maybe in the future they were talking about Python, down into that data layer to then bring out what they need out of the data layer instead of having to go to a data lake that's on other hardware or into a cloud. And so it could simplify significantly the stack for those consumers of this. Like the HSBCs who were showing off what they were doing or Grok and xAI folks and things like that where maybe you can do more as we would talk about, agentic type stuff up above that. And that becomes more complex where you're looking and you need more compute to actually make those queries down to that data layer. Well, if your data layer gets smarter and has some of that AI technology in it like the NIMs and NEO and the retrievers and stuff like that, you're already ahead of the game for pulling that data. And to Jack's point, it becomes more real-time, it becomes more a faster retrieval of that that helps you achieve and you can do new things, solve for new problems, which to your point on ROI is I think one of the big problems is that the data changes so fast, how do you keep up with it, so you're not giving the wrong answers?

Dave Vellante

>> And I want to come back to agentic, but before we do, we heard a lot about atomicity, the atomic capabilities. You remember in the spinning disk drive days, that was the big bottleneck and you'd have to go through these chatty SCSI protocols to actually get a write done. And if something went wrong, then the system would take a long time to recover. And then with flash storage, you got atomic writes NVMe, and that was a big change. And it was interesting to hear Jeff talk about that because in computer science, atomicity is something that ensures that transaction is treated as a single unit, invisible unit so that if something goes wrong, you've got recovery mechanisms in there. David Floyer taught me this decades ago that he was always asking, "What happens if something goes wrong?" And a lot of times the answer would be, well, we have this eventual consistency thing, and so VAST is taking care of all that. And that was something that Jeff spent a fair amount of time, he really didn't get into what it all means, but I think there's underlying technology here that ultimately is going to build trust with customers. Your take, Jack.

Jack Gold

>> Well, and that's their secret sauce. That's where they came from. It was all about that when they first started out with the new algorithms and with the NVMes, that's exactly the problem they were trying to get around. And now they're extending beyond that. So I think that it actually makes an awful lot of sense from an ROI perspective because look, if one data bit goes bad, if one Vector goes bad versus a thousand Vectors, you're trying to write a thousand Vectors all at once, that's a much easier problem to solve than it is if you're trying to do all a thousand of them, and correct those.

Rob Strechay

>> Yeah, and I think one of the things was they had CoreWeave on as well. Coreweave talked about the checkpointing and the fact that they can write down all the checkpoints all at once to that exact point. And I think that's where it comes in for AI because if you have to redo a training run because your checkpoints get all screwed up or you lose checkpoints, that's big time badness.

Dave Vellante

>> The other big announcement is the VAST Cosmos tech community with the mission to simplify and accelerate AI adoption, exploration and discovery. They're bringing together all these organizations who care about AI and have expertise to build an advance AI by creating a comprehensive and supportive environment to nurture innovation. The community is built of AI practitioners and for AI practitioners. So it's of and for researchers, technology partners, service providers, solution integrators. So we heard from a number of folks, you mentioned CoreWeave, we heard from CTO of Deloitte, Chan Zuckerberg was on the list, Jeetu Patel from Cisco, Manuvir Das as well as Jensen from NVIDIA. My man Jon Lin from Equinix was on there, Andrew Ng, you had investors from like NEA, Supermicro and many, many more, many. What do you guys make of this community push? It's interesting that VAST is putting itself at the center of that community and it was able to coordinate so many big players, big thinkers, big brains to come together. What were your thoughts on this?

Rob Strechay

>> Yeah, I mean I think it's a very interesting play. I think that maybe part of it was the fact that the AI Alliance, which does not include NVIDIA, has Intel and AMD and IBM and a number of other research organizations as well as commercial. I think maybe this is a play to really offset that non NVIDIA-based. And NVIDIA is doing a lot. We see them, they'll be at the CNCF at KubeCon + CloudNativeCon talking about how they're bringing stuff to Kubernetes and stuff. But I think again, it's a place for them to play as well. And I think you see some of the others who are in that community like Dremio and others really are helping out bring it to be more of a data community, which I find very interesting. It'll be interesting because Meta's in the other one, and where do the model providers come in when you talk about this community? So I think it's going to be interesting to watch.

Dave Vellante

>> I thought they played it very cleverly though, Jack. I think there was a lot of FOMO in there. Oh, they got Jensen. Okay, he's in. I better jump in here because I don't want to be left out. What are your thoughts on that?

Jack Gold

>> Well, Jensen's in everything of course, if he's not on stage, then you're not making an AI announcement these days. I mean that's just the way it is. Look, generically, I think it's a really good move to have as many outside sources as you can coming in and increasing the knowledge, the AI knowledge if you will. I mean, when we build a big AI model, what do we do? We try to bring in as many data sources as we possibly can to train that model so that we get the best answers possible. This is on a human level, the same thing. There are other OpenAI type initiatives that Rob talked about and there'll probably be even more. This is really, I think, about VAST and NVIDIA getting together and saying, we know this is going to happen. We know they're going to be open initiatives. We know they're going to be groups out there. Let's see if we can focus that effort in our space as opposed to just having a generic open source kind of a community. And so if it works for them, if they can do that, it's at their advantage, because people will be concentrating just like NVIDIA concentrated on CUDA. They're trying to say, they're not saying it obviously, but they're trying to push it towards if we can get everyone working on the VAST Platform with NVIDIA in the background, then we win long-term.

Dave Vellante

>> Yeah, and I want to come back to something that Jensen said at GTC in the spring. Jensen, see sees around corners the head of everybody on this. He talked about reasoning. He said, AI in the future, you're going to give it a query and it's going to go off. You're going to allow it to go off and maybe for a week or two and you're going to say, I got $2,000 to spend on this go reason. Now we've seen glimpses of that with ChatGPT 1.0, where it'll take 30 or 40 seconds and it says, "We're reasoning." And reasoning on the data, which is really interesting. And they sort of also talked about these agentic workflows, which we've written about a lot and researched a lot here at theCUBEResearch and all about sort of learning human interaction. So when I think about the knowledge graph capability, you've got to be able to, and Jack to your point about any data, well, if you have any data, you've got to be able to harmonize that data so that revenue actually means revenue. It doesn't mean bookings, it doesn't mean NRR. It doesn't mean AI. It means revenue. And so you're not arguing in the meeting about what is this? This is a very simple example. Imagine with any data anywhere how complex that could be on any system. And then as well this agent control framework that we've talked about, which seems like that's NVIDIA in this equation, but is this, Rob in your mind, are we seeing the future of the application stack that the blowing away of the three tier system, the database, the server with some middleware and the UI. UI is going to be a natural language interface and this idea going beyond microservices which are hard-coded and fragile and brittle, and they break and they change and you got to hard-code them. Are we seeing the age of where computers and intelligent agents are going to observe what we do, all that sort of dark matter and dark data, and actually create new workflows and new applications on the fly?

Rob Strechay

>> I mean, I think that's where it's going. I don't think we're anywhere close to that at this point, but I think that when you start to look down the path of where VAST is building in, they're definitely building up into that. And I think to your point, you get above the NIMs layer, that harmonization layer that we always talk about of what is revenue and how do you define it and maybe defined in one set of data differently. Right now what they've built is connecting all of the data together and it will know where revenue is quoted all over the place. You still need to harmonize that and define and have that definition on top of that still at this point in time, but you can easily see them moving up that stack and bringing stuff like DBT into there so that you're starting to do the data modeling at on top of the NVIDIA software and the NeMo stuff and the retrievers so that you have an enlightenment of those retrievers and embeddings. I don't think it's there yet, but you could see down that path because that's really needed, to your point, to get to these agents, building agents, building the action models, the small action models, and then longer-term large action models, I would assume.

Dave Vellante

>> Yes, from LLMs, SLMs, SAMs and LAMs, and it's going to be a very interesting transition, but that's really where it starts to get interesting and where we might start seeing much more tangible ROI. And so I want to end on economics. A lot of this business today, you've got hyperscalers buying from NVIDIA and sending the stock market rocketing. You've got companies that are infrastructure companies like Dell and HPE hopping on board and OEMing GPUs, but there's just not a lot of mainstream businesses. It was interesting. That's why we heard from HBSC today and some mainstream enterprises that are saying, yeah, we're actually using this platform and presumably it's starting to throw off some cash and be self-funding. But to point we were talking about earlier, Jack, I mean AI is expensive. And the ROI today has been little singles, I'll call it. How do you see this playing out, and what role will VAST have in really driving AI to be self-funding so that we can break out of this? Frankly, the market not growing, the macro, not growing, the IT spending, not growing through the roof. The AI piece is growing, but you got this barbell effect. You got AI is hot, everything else is like let's steal from everything else so we can fund the AI. When does it become self-funding? What are your thoughts on economics of AI?

Jack Gold

>> Yeah, so I think we've got probably another two to three years before it becomes economical for most enterprises and even big enterprises. I mean today, if you look at what AI, HBSC is a great example. They're multi, I don't even know what the numbers are, huge billion-dollar corporation, right? They can afford tens of billions of dollars to throw into something like AI to see if it's actually going to help them long-term because look, they're a bank. If they prevent 1% less of fraud on their systems, you're probably talking about tens of billions of dollars. So putting a few million dollars into something like that is not a problem. If you're a general company, if you're a supermarket chain, if you're auto dealership, they're not small anymore either, but they're not going to put tens of millions of dollars into AI. So it's got to get to a point where the ROI is reasonable for me. If I want to get a 10% productivity improvement for my people, I can't spend 20% of my revenues doing that. It's got to be something reasonable. And I think we're still two to three years away from that for a couple of reasons. Number one is building models is very expensive. Very few companies can afford to build their own custom models, so that means that I've got to go to a hugging face or somewhere else and try to pull a model that will fit me. Then I have to fine tune it or RAG it or whatever I've got to do to it to make it for me and not for just the general purpose marketplace, and then I've got to find someplace to run it. I'm not going to go out and spend billions, literally billions of dollars on 100,000 H100s. I can't afford to rent them either because it's costing $3 million an hour to rent. So it's got to move to a model where we're actually talking about smaller AI models. We're talking about a lot more data that we're bringing in from multiple places, which is where VAST comes in, and we're probably going to be talking about mostly inference at that point because most of the models are already going to be pre-built and we'll just modify them slightly. This is going to take time to accomplish. VAST has a piece of that. The other players, not just NVIDIA, but there are a lot of other chip players, they're going to have a place in this as well because there's a lot of concentration on custom AI inference type chips that will reduce the cost. And so all of that's going to come together, but it's not going to be tomorrow.

Dave Vellante

>> So VAST can play in the training game, it can sell to the GPU clouds, it can OEM its file system to HPEs of the world and get revenue there, and that's a bridge to when the market for inference actually overtakes the market for training. Not the training's going to go away, we don't believe that, but how do you see the economics? Let's close there.

Rob Strechay

>> Yeah, no, I think Jack is dead on with this. I think for it to come down market and get out of the HSBCs and Chan Zuckerberg type where they're building their own. And they're AI companies already because they're doing drug discovery and healthcare and fraud detection and they've been doing AI for decades.

Dave Vellante

>> xAI,

Rob Strechay

>> I mean, Grok and all that, you start to look at that. Again, to come out of those really, hey, where I can spend 10% because I'm going to make back 30% off of that because it gets me some advantage. I think you need something. I got to say, I'm interested to see the TCO on what they're doing here because you should be able to compare it to going in, hey, I'm going to go to a Snowflake. I'm going to go to a Databricks. I'm going to go and roll my own data warehouse or go and use one of the other ones DBMS that's out there, or to Jack's point, go out to another hyperscaler and just do it all there. This should be able to have a TCO. Now, you still have to buy either the H100s or maybe you can get away with L's above it because the compute layer, maybe it's simplifies the compute layer by having all, you already have NVIDIA in the box for doing the VAST stuff and doing the NeMo and the NIMs and all that. I think that's yet to be seen if that will help bring the stack down market or not, but something has to happen there. We know that the economics just isn't there right now.

Dave Vellante

>> I mean, you have virtually unlimited capital right now funding this from the hyperscalers and others. But guys, great analysis. Thanks so much, Jack for coming in, Rob being here. Appreciate it. And thank you for watching the CUBE special coverage of VAST Cosmos. We're seeing the evolution of a new software layer that is pointing the way to the future of AI infrastructure. So often we see a startup that reaches escape velocity because they've created a solution, as we were just talking about the economics. Can VAST be that 10X better at one-tenth the cost, and do things that were previously impossible or just not cost-effective? Will VAST be that next great AI infrastructure company. Right now, VAST is in a good position to be just that with a healthy balance sheet, strong growth and cashflow positivity, which is exceedingly rare for a young company that's growing as fast as VAST. We're super excited to hear, to document and report on the ascendancy of VAST. As always, SiliconANGLEs, we got you covered with the news, theCUBE Digital Video, and our team at theCUBEResearch. Thanks for watching and we'll see you next time.

Analyst Conversation

Dave Vellante

Jack Gold

>> Thank you.

Dave Vellante

Rob Strechay

Dave Vellante

Jack Gold

Dave Vellante

Rob Strechay

Dave Vellante

Rob Strechay

Dave Vellante

Rob Strechay

Dave Vellante

>> Structured unstructured.

Rob Strechay

Dave Vellante

Jack Gold

Dave Vellante

Rob Strechay

Dave Vellante

Jack Gold

Dave Vellante

Rob Strechay

Dave Vellante

Jack Gold

Rob Strechay

Dave Vellante

Rob Strechay

Dave Vellante

Jack Gold

Dave Vellante

Rob Strechay

Dave Vellante

Jack Gold

Dave Vellante

Rob Strechay

Dave Vellante

>> xAI,

Rob Strechay

Dave Vellante