The Exchanges

Q Okay, so maybe to, uh, double click for the more technical people, um, in the, in the audience. So, like, how does that, how does that work? How were you able to, um, sort of do this no compromise kind of approach?

A Sure. So every scale out system that I am aware of before VAST is based loosely speaking on a concept called shared nothing sharding. You have a lot of nodes in a cluster and each one of them has direct access to a bit of the namespace and each one of them has responsibility for a piece of the pie for a piece of that namespace. And we realized that that architecture is reaching the end of its rope. Um, it starts to see diminishing returns in performance as you scale beyond a certain limit. Uh, resilience is very problematic. If a node fails, you need to recover that node's responsibility, and it can take a week, and during that time, you can't have another failure, um, and so that limits the scale of it. What we needed to do is the opposite. Instead of direct attached having drives in the nodes, we disaggregate. We put the drives on one side of the network, we put the logic on the other side, and we leverage a new protocol called NVMe over Fabrics to make it look like All of those drives are directly attached. That allows us to scale capacity independently from performance. It allows us to scale with dislike parts over time, so you never have to migrate your data between systems. More importantly, it allows us to move from shared nothing, sharding, to shared everything. Every node can now see the entirety of the data set. Data on low cost flash, metadata on storage class memory…

AI assessment note: “Instead of direct attached having drives in the nodes, we disaggregate.”

Q So the database is to query the metadata, is that what you're saying? That's how it started, and everything expanded into.

A Yeah, everything in VAST is multi-protocol, and so, for example, you can write a file and then read an object. Everything is the same underneath the covers, and so the database is the same. You can write a parquet object and then query it using SQL natively off of our platform without needing any higher layers on top. Um, and so now people are using it not just to analyze their metadata, but also as a data warehouse. And in the same way that we broke those fundamental trade-offs, it grows to extremely large scale without compromising on performance, without compromising on consistency and ACID requirements. And so we find that in the database space, there are more trade-offs to be broken. Ah, row-based databases and column-based databases. That all stems from storage. It stems from hard drives needing to be sequential. If we can give different views into the same information, then you don't need that complexity. You can consolidate very similar to what we did, ah, in consolidating those tiers.

AI assessment note: “now people are using it not just to analyze their metadata, but also as a data warehouse”

Q All right, so data store, data base, and data engine, that's the compute part?

A Yes, that's where we bring it to life, and so both the store and the database are static in nature. You write to them, you read from them, you query from them. Our customers didn't like that. They don't like the fact that their application is written in that way. In fact, they want it to be, I'm going to put a plug for you now, data driven. Um, they want everything to be, ah, based on the information as it flows in. That genome file came into the system. It should trigger that inference function that runs on it. It should trigger an incremental training, ah, job that runs on it. That should run on a low-end GPU. This one should run on a high-end GPU. Um, if we can build that language of triggers and functions where you can then Understand more and more about your information, and put that in the database, and that triggers more functions, and you have this recursive machine that's all data-driven, um, then we have the full stack, and that allows us to do things that couldn't be done before.

AI assessment note: “Yes, that's where we bring it to life”

Q Interesting, yeah. I guess the timing was perfect, right? Because you were, you caught the alternative data Uh, trend in hedge funds at the time. Okay. Very interesting. All right. Um, how do you sell today? Like you have a sort of a classic kind of, um, high touch sales organization kind of thing?

A We do. Uh, we are very close with our customers, and I would say the two things that we do well is R&D, building a product, and sales, selling the product. Uh, everything else we're not very good at. Um, We want a very intimate relationship, especially with the first several hundred customers, uh, which is where we are today. And, um, now through partnerships, we're able to get reach into everybody else. And so you mentioned HPE. HPE has, uh, tens of thousands of sellers out there, and they've decided to standardize on VAST as their file offering and, uh, soon to be other parts of their stack. And so they are Giving us reach, uh, beyond those most data intensive organizations. We're partnering with, um, AI clouds to get reach into these, uh, smaller AI startups before they get to a hundred petabytes. Um, and again, the analogy of the operating system, we want to, on the one hand, make it easy for applications. On the other hand, have everyone underneath us help us sell.

AI assessment note: “We do. Uh, we are very close with our customers”

Q data. You said that, uh, several terabytes was, uh, hundred terabytes was too small, and you're interested in the petabytes. Um, how does, how does one do that when you're a small company, uh, and you go after the biggest customers, and you have to convince them to use your product that has been around for like three years? Who were the first customers, and how did you get them?

A Having a lot of data does not necessarily mean it's a big company. We just sold an exabyte, uh, to a company with less than 30 people. Uh, that's not common, but it exists in 2024, especially around AI. Um, when we started, we went after early adopters. Uh, we went after hedge funds, uh, in the, uh, in the financial sector. Now we're selling into a lot of, a lot of big banks. Uh, in the early days we started with Uh, research centers and universities in the life science space and medical imaging. Now we're selling into large pharmaceutical companies. Um, and so we started with early adopters with smart people that we could work with and with people that did not have a lot of process and bureaucracy. We wanted direct access into the end users and again, learn from them what they needed us to build. Um, we started working with them a lot before we had a product. And by the time we had something to sell, it was theirs as much as it was ours. And so it was a very natural process of them moving from testing to using in production once it was generally available.

AI assessment note: “When we started, we went after early adopters. Uh, we went after hedge funds”

Q Any lessons learned in that kind of like bi-continental culture, uh, where you have, you know, a strong group in Israel, presumably, and then there's Strong group here. Like, how do you make it all jailed together culturally?

A That's a really good question, and there are differences. For example, here, um, people work, uh, from home mainly. Um, in Israel, everybody's in the same office together. They won't have it any other way. Um, in Israel, uh, the culture is one of, um, finding shortcuts and doing stuff very quickly, sometimes quick and dirty. No, you don't think about tomorrow as much because who knows where we're going to be tomorrow when you're in the Middle East. Here you build for a hundred years. You see these buildings that were built a hundred years ago and a 150 years ago, and it's a very, I think the art of building a company is to knowing how much to draw from each of those extremes, and when do we need to build for the long term, and when are we just experimenting and Uh, we need to move fast, and, and it's okay if things break along the way. I think that juxtaposition of, um, cultures is the vast culture. Um, and it's very, very important to articulate it. In the early days, everyone's in the same room. You don't need to talk about these things. They just happen. As you grow, and as you grow at this pace, you have to explicitly tell people, Uh, that you can leave a meeting in the middle, and that you shouldn't waste time on internal crap, and that titles don't matter within the company, and that we don't have a hierarchy. We have a customer at the top, and everybody else that's servi…

AI assessment note: “the art of building a company is to knowing how much to draw from each”

Q time, you know, there's a number of other companies that are hoping to be the software data layer for generative AI, like that's the Databricks of the world, there's the Snowflakes of the world, plus, you know, the hyperscalers. Um, so, you know, and that was very much already the case in 2016, so what was the fundamental insight that led you to think, well, actually we can do better?

A It's a few things. The first one is that, um, It was very clear back then that old ideas, old algorithms are starting to have value. Um, and again, eight years ago, uh, it was finding cats on YouTube videos. It wasn't yet generative AI or large language models, but it was already clear that fast access to a lot of information is the key to, uh, generating benefit from these algorithms, and we wanted to enable that. We wanted to enable that for Everyone else. Obviously, Google knows how to build it for themselves. There are a handful of companies who do, but everybody else needs someone to do it for them, um, and so we started by breaking those fundamental trade-offs that have existed in the storage space for so many years, and then we realized that in order to do that, we need a new architecture, and that new architecture enabled us to break those trade-offs, not just for storage, but for the entire stack, and so I think our The biggest advantage versus those other companies that you mentioned is that we started late, and that allowed us to see this deep learning problem in front of our eyes versus machine learning, again, a lot smaller, a lot slower, and it also allowed us to leverage underlying parts that didn't exist, uh, before we started, and so starting late is our biggest advantage.

AI assessment note: “our biggest advantage versus those other companies that you mentioned is that we started late”

Q expert, but, um, the little bit I know, like, the, the problem is that you have different tiers, right? And you put, like, the historically before you guys, so you put the new thing in the first tier and the old stuff in the second tier, but the problem is to process it is, is hard. Is that, so, so that's the historical thing, and you guys completely broke that?

A We tried to, yes, collapse that pyramid because Um, historically, it was relatively easy to say, this is two days old. I need fast access to it. Now I'm moving it down to a mid-range system. It's two weeks old. Now it's two months old. I'm never going to touch it again. I, I can put it in an archive, but AI doesn't conform to that model. You need fast access to everything in order to build an AI model. And once you have it, you want to infer on everything to generate value out of it. And it turns into this loop of, Random reads all over, ah, over and over and over again, and so you want a new system that enables these new workloads.

AI assessment note: “We tried to, yes, collapse that pyramid”

Q Still in the sort of product, um, rabbit hole, uh, roadmap, anything you can share about what you're building and what is next?

A Yeah. So we're just getting started on a lot of these things. The database just came out early last year. We're still in the early innings. The data engine is actually only coming out later this year. Uh, the data space is just at the very beginning of it. And so I think building out and filling out this vision over the next three to four years is what we're working on. And then we'll expand beyond. So, um, in the same way that, um, 2019, when we started selling, we exposed universal storage, and then it took us a few years to get to it. Now we're doing the same with this data platform. Eventually, I think what is really required is an operating system. Every technological revolution that we've had, whether it's a personal computing revolution, or the internet, or mobile, it wasn't the hardware, it wasn't the chips that made it widespread. It was an operating system that made it easy for everybody to use, and I think today, in many cases, you still need a PhD in computer science to operate a large AI cluster. We want to make that easy for enterprises, easy for clouds, easy for everybody, and that's the operating system that we're eventually, uh, trying to bring to market. Um, that's, that's our, that's our mission.

AI assessment note: “The data engine is actually only coming out later this year.”

Q Okay, so maybe going back to 2016, so you started the company with co-founders. Um, how did, what did you do before? How did you know each other?

A So, interesting story. I was at, ah, a company called EMC before. I was acquired into EMC, ah, through a company called Extreme IO. I was head of engineering there. And, um, I felt like they didn't need me anymore. Ah, sales were going well. We hit three billion dollars in two years as part of EMC, of course. Um, and I felt like we needed to build something new. Um, in order to address challenges that were coming in large-scale analytics, in AI, I felt the new architecture was required, and building something new within EMC was practically impossible, and so I left, and that's how VASC got started. Um, in the very early days, it was just me, but very quickly, uh, we assembled a really, really good team of co-founders around me.

AI assessment note: “in the very early days, it was just me, but very quickly, uh, we assembled”

Q And what have you found the hardest, uh, on your journey as a, as a leader and as CEO?

A I think it's all hard. Um, if somebody knows of an easy way to do this, um, I, I, I'd like to, to get that email and, ah, I was telling someone, ah, before I have 4000 unread emails that we don't have time to, to do all of the things that we need to do. As much as we try and as much as we work overnight and over the weekend, um, it's difficult. Ah, we're trying to do something that shouldn't Be possible. Um, the big companies that we're competing with have more resources than we do. They have a better brand than we do. They have everything better than we do. And we, through a fighting spirit and through, uh, a weird, uh, culture that we've built and, uh, uh, it's an island of misfit toys, I call it. A lot of very, very interesting people that together are doing something that Shouldn't be possible. That's what drives me to continue. I love, I love the innovative aspect of it. I love the, um, the people that are working at Vast. I love our customers, um, but it's all very, very difficult.

AI assessment note: “The big companies that we're competing with have more resources than we do.”

Q And it's declarative? Like, I, I, I tell, uh, the compute engine where it needs to go based on certain characteristics of the data, or does it infer from the data where it should go?

A It's, okay, so, uh, the triggers are based on actions that happen, so you can have on new file, uh, and then you can have a filter on that. Uh, it's not every file, but only files of type genome, and only that belong to people, uh, above the age of 50, and then you call that function if that happens, or, uh, as things get updated, you call another function. Um, the functions themselves can be whatever you'd like, Um, and of course, um, once we understand things about them, like the length or, um, the urgency, then we can schedule them in a much more efficient manner. The, the fourth piece of the platform, in addition to the data store, data base, and data engine, is what we call the data space, and that allows us to go across geographies, and so now you can run a vast instance in, uh, public cloud AWS region east, or In CoreWeave or on-prem or at the edge, and we stitch all of those instances together into one global namespace, and that allows us to schedule close to where the data is. It allows us to move, uh, functions rather than moving data across, uh, country lines. In some cases, it's not allowed to move data out of Germany, but you can run a federated training job That's global, and so that again gives us an ability to do things that couldn't be done before. For example, break the speed of light. Data has gravity. Compute is a lot more lightweight.

AI assessment note: “the triggers are based on actions that happen, so you can have on new file”

Q But how did you even know to hire the good people when you started and you need to hire your first salesperson?

A Um, so the big advantage that I had coming from EMC is that I had first, uh, I, I saw how to build a business. Um, previous to EMC, I was with a lot of failed startups where we built really cool technology that nobody ended up using. Um, and then I realized at EMC that building a business is so much more interesting because then you get to innovate over and over and over again. And you get this flywheel effect where you can, um, again, ride an exponential curve. The fact that we're worth nine billion dollars now, which is staggering, is because we keep tripling. Um, and if you start over, then you start from zero. That's not any fun. Um, I had really, really good people that worked with me at EMC that were on the sales side, and they were, they were able to help, uh, direct me in the right direction, and also a lot of them Um, joined Vest over the last few years.

AI assessment note: “I had really, really good people that worked with me at EMC that were on the sales side”