Under The Banyan Tree By Sam Awrabi
The AI-Native is the podcast hosted by Sam Awrabi, Founder & Solo General Partner at Banyan Ventures—an AI-native venture fund with nearly $17M AUM and a track record of backing early-stage companies that have gone on to generate over $490M in revenue from pre-seed and seed. Sam draws on deep, first-hand AI infrastructure experience and thousands of founder relationships to explore how the next generation of AI-native products are being built, sold, and scaled. Each episode dives into the tools, mindsets, and systems driving the most important technological revolution of our lifetime.
Under The Banyan Tree By Sam Awrabi
How This YC-Backed CEO Is Reinventing Data Centers With AI—Drawing on Her Microsoft Experience
What happens when data centers built for the cloud era try to handle AI workloads? They break down—frequently and expensively. Margarita is CEO & Co-Founder at Aravolta. She is a trailblazing entrepreneur building a cutting-edge data center management software company. With a solid background in engineering and extensive experience at Microsoft working on Stargate with OpenAI.
Data center operations have remained largely unchanged for 15-20 years, creating a massive efficiency gap just as computational demands are skyrocketing. After building internal tools at Microsoft for supercomputer deployments like OpenAI, Margarita recognized the opportunity to transform the entire industry. She discovered that 98% of data center operators were deeply unhappy with their existing software systems but had few alternatives available.
The technical challenge is immense—connecting to hundreds of different device types across dozens of vendors, each with their own proprietary communication protocols. Aerovolta's breakthrough came through developing a middleware layer that can rapidly integrate with virtually any equipment, from transformers to cooling systems to the GPUs themselves. This foundation of comprehensive data collection enables their AI to analyze thousands of inputs simultaneously, identifying inefficiencies humans simply cannot see.
The results speak for themselves: immediate 2-5% power savings upon deployment, 15% increases in uptime, and significantly extended hardware lifespans. This last point is particularly critical as some data centers are burning through million-dollar GPU clusters in just one year instead of their expected 7-8 year lifespan due to workload spikes and suboptimal operations.
Throughout our conversation, Margarita shares insights from her founder journey—from immigrant parents who taught extreme gratitude and self-reliance, to navigating the Y Combinator experience, to building a company that's now working with the most sophisticated compute providers in the world. Her approach to fundraising, hiring, and product development offers valuable lessons for anyone building in the AI infrastructure space.
Ready to see how intelligent infrastructure management could transform your data center operations? Listen now to discover how small, focused teams are solving massive challenges that once seemed insurmountable.
You've built an AI co-pilot that runs on top of all the different real-time data sets. How did you build that?
Margarita Groisman :The world of 40 engineers working on the same problem is over. I think small and nimble teams really rule, because you have every person going very deep. All of your engineers are actually just building and we could give you all these different data points from the outside. There's so many factors that you can look at and it's much better at pinpointing those exact places where you could have increased efficiency.
Sam Awrabi :What are some visions of the future that excite you the most? Are there things you're talking about from a product development standpoint?
Margarita Groisman :There is now a bunch of modular data centers that are running on stranded compute and in the past that could only be done really for mining, but now you could actually run inference workloads on basically wasted energy, where you deploy smaller amounts of energy across many different regions. We can power so many different types of data center models because the core of what we do is data collection. The easier part is training the model. The harder part is collecting all of the data that you need in order to train the model. Because we focused ourselves as a data collection company, we can power every type of data center that will be entering the market.
Sam Awrabi :Margarita, welcome to Under the Banyan Tree. I'm Sam Arabi, founder of Banyan Ventures, and I'm really excited to have you here today. We're in San Francisco, your home turf.
Margarita Groisman :Yeah, I live like five minutes from here. Thank you so much for having me. This is so perfect.
Sam Awrabi :Love it. Did you walk here or Uber?
Margarita Groisman :Well, no, I still drove. Okay, drove Nice.
Sam Awrabi :Yeah, I'm used to being in New York, so there's not a lot of you know like not a lot of drugs. City bikes though. Yeah, less biking here these days, like it seems like.
Margarita Groisman :Have you taken a Waymo yet?
Sam Awrabi :No, I haven't done that yet. Oh, you have to. Okay, yeah, actually I want to do that today, now that you bring that up.
Margarita Groisman :Yeah, do you have the app yet? No, or?
Sam Awrabi :Okay we could, I might take you up on that. So this podcast is really focusing on telling your story and learning about you and what you're building For the readers and listeners context. We, banyan, invested in your seed round. We're really excited about that. You're building data center management software for the modern age. You're leveraging AI to essentially build a copilot that detects vulnerabilities and keeps data centers compliant, reduces the actual power bill of the data center and brings all of the modern functionality of software and AI to the forefront of data center operations and management. But I would love to hear a quick introduction to you and a quick pitch on Aerovolta and then we can just jump right in.
Margarita Groisman :Yeah, so, like you mentioned, I'm Margarita and I started Aerovolta about a year ago. The reason we started the company is data center operations have basically remained pretty stagnant for the last 15, 20 years since we started building and scaling out data centers for just traditional compute purposes. And as we're scaling out and building out so many new data centers at a scale we haven't seen before for AI data centers, we really need to update the way we do operations and management of the data centers so that we're as energy efficient as possible, as efficient labor-wise as possible and we don't have any interruptions or downtime across our really, really expensive data center assets just because this is the workloads that are training AI. We can't really afford 20% downtimes or some of these crazy figures we've seen in the past.
Sam Awrabi :Yeah, I mean, there's so much to unpack here.
Sam Awrabi :We have six or seven investments in physical AI infrastructure products at Banyan with my AI fund and when I first saw what you were building, you and Jack, I think it was like back in January I emailed you and I think I didn't hear for a while.
Sam Awrabi :But I kept following your updates because I firmly believed that the future of data center management needed to be upgraded and I had looked at the other tooling in the space and you're talking about, like, competitors that are 20 years old. And you know, I know your background, you have some experience working at Microsoft and on really large deployments there supercomputer deployments but how did you get the intuition to say, hey, jack, let's jump ship and go to Y Combinator and start a data center management software? I mean that that like it's not even a space where I'm seeing, you know, 10 other competitors. Necessarily, I'm sure they'll, they'll be coming when they see your success. But how did you get that insight to jump into the space? And then, you know, bring AI along with you for managing the data center. We'd love to learn more about that.
Margarita Groisman :Yeah, you know, at first, when we were ideating and just running the idea past some people, a lot of people told us that basically these people, the industry is older, everyone has been running things a certain way for the last 15, 20 years. They wouldn't really use new tooling and it's kind of all embedded. So what we did is, before we had a product or really knew anything, we went to some data center conferences and just talked to a ton of people and just got an idea of what are they doing for their operations, what does it look like right now? And like, 98% of people that we talked to were deeply unhappy with their existing software systems and a good amount of people were looking for new systems as well. But there were limited market options, or at least limited market options that were modern, usable and actually fast to deploy. So I think that was a big thing, just starting with the people we talked to at a couple of the initial conferences. One of them was like the AI in front in San Francisco. Then we started building a product.
Margarita Groisman :Pretty quickly I think it transitioned because I had built an internal, just like a simple tool dashboard within Microsoft that was used by a couple of teams for just managing data center deployments, especially new tech deployments, and it was just to kind of streamline all of our data points into one simple view to actually make sense of things. But I had wanted to build something much bigger and much more for the whole data center operational front. For a long time, internal to Microsoft, that would have been pretty difficult to do just because it would have taken so much coordination, I would have to ask so many higher ups and it would have taken five to 10 years probably and it would have turned into a completely different project. So, realizing and talking to so many others, we realized we could build this full tool for external and all these other data centers.
Sam Awrabi :Wow, I mean there's so much I could say on that. But things that jump out to me is like why Common Error Talks about build something people want and so if you're hearing 90% 95% of the conversation saying we need this and we're open to actually evaluating new options, but there aren't any, there can't be a better sign to start a company new options, but there aren't any. There can't be a better sign to start a company and you know I've worked with so many different software and infrastructure startups and you know that's something that I've seen not go the right way is you build something that maybe people want but the organization and the market just isn't ready to actually adopt it, and usually what it looks like is deals lacking urgency. So you're pitching this product and it's like, yeah, this is cool, I like it, but you know we it's not like a top priority, it's not a need, whereas what you're building when I met you and saw your pipeline and the customers you were talking to, I mean these are the most advanced, sophisticated, cutting edge and largest data centers and compute providers in the world and it's pretty insane, like even at microsoft, they didn't have a solution internally for for you as a data center kind of management operations person to use. So he's just like, hey, I'll just build it myself.
Sam Awrabi :So I guess, on that journey, what have been some of the challenges? You didn't know about that. You wish you knew. Now, either from like a sales point of view or a product management point of view, would just be curious, like have you seen anything so far? You're like, oh wow, that's like insanely hard, but you're cracking the code on it. Or just would love to hear from you because this is such like a new and fascinating space on how you're approaching it.
Margarita Groisman :Yeah, I mean this is obviously the first company I've built and honestly everything's a challenge. I'm going to be honest, when you're building a company, you're kind of caressing and cajoling something from nothing and non-existence into making it real in the world and kind of making the world know about it. And when no one knows your name, you're trying to get so many meetings and kind of build up this pipeline from nothing. And so I think, just like I think there's been a lot of challenges that are kind of on the engineering front, how do we incorporate hundreds of different device types? Luckily, our engineering team is, like has been, one of the only teams that I think could really tackle this problem. You have to go very deep into low-level hardware. And then, outside of the engineering front, how do we build our sales pipeline?
Margarita Groisman :We tried different things, like how do we get people to actually know about us? It turned out conferences were kind of the best approach, in person and talking to real human beings. So I think that was. It was just sort of figuring things out as we go and trying a bunch of things, and you know, a lot of things didn't work that we tried. But I think once you find the thing that works. You just kind of iterate on it and do it again, and again, and again.
Sam Awrabi :Love that With the two questions I have, like the lower level integrations, because I'm just like thinking about it. You go into a data center. There's different regions, different constraints of that power center sorry. Positron Recognize SpinCloud. All three are inference providers. All three are rapidly scaling their offerings, like Positron just announced their $51 million Series A yesterday Shout out to my Tash and team. But the point I'm getting at is like it's not all the same. You know, a software usually accounts for X amount of variables and then you add over time, whereas here these are like hardware-based dependencies and it's a software product and we're going to get to this too and you're adding AI into it and helping the data center. But how do you on the engineering side, like how do you actually account for all that? And you know, I think you're two or three, maybe four employees at this point, so yet you're working with the biggest customers in the world. So how are you like doing that? I mean, it's it just blows my mind personally and I just love to learn more.
Margarita Groisman :Yeah, I mean, first of all you just need some of the top notch engineers who put every effort into solving very hard problems. This isn't like you could vibe code. You know together a nice front end you actually have to connect to like we had customers with custom chips and customers with very non-standard data center deployments and you basically have to go in and find a way to integrate these devices from the actual UPS to the transformers and the busways and the tap boxes and all these new device types. So it's not just the servers and the switches and the PDs, and so in order to do that, we have to go really low level and connect to those devices.
Margarita Groisman :Honestly, we didn't know how to do that two, three years ago, but basically we've built up our own middleware layer. It makes it much easier to integrate devices much faster. The reason why a lot of DSIMs and other deployments of data center software have failed in the past and even BMS systems or building management systems don't do this is because there's so many vendor types and it's very hard to be vendor agnostic because every vendor has their own kind of language by which you can communicate to the device. So our middleware is basically what we started with. We started with how do we actually integrate all these different device types, and that was kind of the main technical challenge, that we've kind of built up a way to really really quickly integrate very many different languages or ways of interacting with those devices.
Sam Awrabi :Yeah, wow. Well, I mean that, with my limited but somewhat knowledgeable understanding, it makes sense horsepower and extreme focus on details, and then just finding a way to build out this middle layer that can be flexible enough to not be constrained by the customer requirements, and then constant adaptation to be able to service that customer. And the way I kind of understand it is you know, a human can only work x amount of hours a day. They might, might be tired, their kid might have kept them up all night, or their puppy, and so people, you know we're people, we're not perfect, whereas AI I'm not saying AI is perfect, but AI is a hustler AI is always working, just like 24 seven.
Sam Awrabi :And my understanding is like hey, you've somehow built an AI co-pilot that runs on top of all the different real-time data sets that are being produced from the data center and you're finding these extremely tangible suggestions, and I'd love to learn one how did you build that? Tell us more about that. And two, are we going to be in a world where it's like AGI for the data center and you're making real-time adjustments at Aerovolta without even needing the customer's permission? Or what's the frontier going to look like in the future, and where are we at today, and how'd you even do that?
Margarita Groisman :Yeah, so I think there's a lot of things there that I want to touch on. I think I want to first step back into your last question, which is how do we do it with such a small team? I think one of the big things that I believe in is the world of 20, 40 engineers working on the same problem is kind of over. I think small and nimble teams really rule, because you'd have every person going very deep into their ownership area and building something incredible. You're not spending all of your time coordinating between your engineers. All of your engineers are actually just building all the time. So that's one big thing.
Margarita Groisman :The next, in terms of the AI component, and how do we actually train a model to make these data centers more efficient. So there's been a couple of approaches, and I think there have been companies in the past that have tried to build so-called autonomous data centers. I think AI is really good at this specific thing, which is can we hold thousands of data points that uh for in our head all at once? A human isn't really good at that. We can maybe think about like seven numbers and memorize.
Sam Awrabi :I'm like a two or three.
Margarita Groisman :Exactly and and kind of make a decision based on those data points. We can hold some things in. I'm like a two or three, for example. We could make a bunch of recommendations in our data model and say, okay, we have like this system and these are how everything is interacting. This is the model of the data center that we could show you and tell you and we could give you all these different data points.
Margarita Groisman :This is how much energy this equipment is using. This is how much energy this other point of equipment is using. And now it's holding hundreds of these data points and it's saying, okay, where am I being inefficient? Where am I losing actually some power? Maybe between going from the transformer to the UPS, to the busways? Where can I gain energy? And it could even be in like the fan speed settings, the actual way the cooling is running and when to run it during the daytime, and when do you have natural cooling just from the outside? There's so many factors that you could look at and it's much better at pinpointing those exact places where you could have increased efficiency.
Sam Awrabi :Makes sense.
Sam Awrabi :We at Banyan and I'm a solo, so I say this we think that I'm running the fund and the thesis is all around AI, native technology, being the dominant force of our lifetime.
Sam Awrabi :That will create more value than web, mobile, even compute combined, and we're talking about at least a 75% increase in each person's productivity in the coming years, and I think that could be on the low end. And so now we're looking at the data center space here and applying AI, and that is a theme I see is where there's a lot of structured, repeatable inputs into the AI. You can get structured, repeatable, high quality outputs and so you've put that together repeatable, high quality outputs and so you've put that together. And I'm curious when you and Jack you know your co founder and CTO, or just you know having a beer ideating after work what are, what are some visions of the future that excite you the most? Like, where could this go on the data center space, applying this AI native angle on top of the structured sort of outputs from each data center? Like, are there things you're talking about or really excited about from a product development standpoint?
Margarita Groisman :Yeah, I mean there's a lot of things we're excited about. On one hand, I'm excited about all the new data center type of models that are coming out, and by models I mean there's now a bunch of modular data centers that are running on stranded compute, and in the past that could only be done really for mining or these other data center purposes, but now you could actually run inference workloads on basically wasted energy. And there's now a hive model of data centers where you deploy smaller amounts of energy or compute across many different regions and you use less energy, and so we can power so many different types of data center models. Because the core of what we do is data collection, the honestly easier part is training the model. The harder part is collecting all of the data that you need in order to train the model, and it's pretty difficult to do that because, like I mentioned before, you have so many different equipment types and so many different vendors.
Margarita Groisman :And so because we focused ourselves as a data collection company in some ways across every single data center type, we can power every type of data center that will be entering the market, and I've met, I think, at this point, like tens of new data centers that are coming up, people who are building a 50 megawatt development, people who are doing like 10 sites of 5 megawatts, people who are doing immersion cooling, people who are doing liquid cooling, and all these new, innovative ways to make data centers, and I think we're really, really excited to be the company that makes all of that possible. Like, these data centers wouldn't be operating without their operational software, and so there's all these different data center models and we're just more excited to be able to, yeah, help them be like, be alive, like the hardest part is the operations for these data centers, and now they could just say oh, that's covered.
Sam Awrabi :Like we already have air volta yeah, I mean, and just to give the listeners context, um, one of our investments is ori gpu cloud. Um, we were, are we being? I was like one of the first, I believe, five investors after they transitioned into the GPU cloud space and they were one of the earlier GPU clouds based out of the UK and they had a really incredible team like the initial team and investors, even like Lee Fixey was an early investor in their original Ori vision like a distributed cloud computing platform, so they had an expert team on Kubernetes from the jump and then when they went in the GPU space they were full stack and they support multiple very high profile clients both in the Middle East and the US. I won't share any names without their permission, but they're behind some of the most well known and high scale kind of providers of compute out there and they're full stack.
Sam Awrabi :So in Ori's business they are responsible for what could go wrong, like if there's an outage of power. They're on the hook to provide their end user. Whoever's paying them for this hardware, thesepu kind of you know equipment they're responsible to give them a refund and so they they call it an sla, a service level agreement um, it's massive dollars. I mean for them a small deployment's 30, 40 million dollars, like that's like, that's like a small, uh and that and that's paid back to the debt provider on their business uh, usually on like a small, and that's paid back to the debt provider on their business, usually on like a two to three year clip, and then they own the hardware for seven to eight years, similar to the CoreWeave business model.
Sam Awrabi :And so I just just for listeners like that, what is at stake here with this space is it's mission critical. For every minute of downtime it could translate into two or $300,000 in losses. And then there's other downstream effects because now that customer, which is all built on trust, might not have the right level of compute. So if you're in customers a bank, they don't have compute. Now their customers can't use their banking. So I completely agree this is helping them stay alive and handle everything.
Margarita Groisman :Default alive Right.
Sam Awrabi :Yeah, I love it. I think one of the things too is right off the bat, when you deploy your software. It's like a 2% to 5% power savings.
Margarita Groisman :Yeah, that's what we've seen just from initial deployments and then estimating what were the gains that we saw immediately just from actually implementing the recommendations.
Sam Awrabi :Wow, any other like big proof points you're seeing in the market today when you like the before and after I love that framework of before our software was here and then after like are you seeing any other big before and afters that you're really proud of?
Margarita Groisman :Yeah, I mean a big thing is just uptime, like I said. So data centers will show 99.9999% uptime if they have to be.
Sam Awrabi :That SLA agreement right.
Margarita Groisman :And that's a requirement. But the way they can actually do, that is, when a server goes down or a port goes down or another issue happens, they just move the workload to something that's actually working.
Sam Awrabi :Right. It's not like the actual. They beat the cost Exactly and behind the scenes. That has a massive downside to you as the GPU provider.
Margarita Groisman :Exactly. So that's very painful and expensive, because if you have 20% of your hardware down, you need 20% more hardware to run the same amount of workload, and so that's, like I said, expensive.
Sam Awrabi :And just for context, for listeners, just to finance that, depending on your operation, that could be in the hundreds of millions of dollars and spare compute you just have lying around for this 99.99% sort of guarantee you're giving your end user.
Margarita Groisman :Yeah, spare, compute, spare parts all of that is a huge component, and so we like to say that you're going to see 15% more uptime, or?
Margarita Groisman :actually much, much more increases in getting the availability much high. And then another component is a lot of these customers, at a certain scale they're just running the GPUs to basically have six-month lifetimes or one-year lifetimes. We saw that with a big customer that were just running their GPUs at 90% capacity for training workloads and then after a year you just have so many issues and problems with the GPU that it's just basically you need to buy a new set, even though they should have much, much longer uptimes or lifespans. If you're running them at, you know, 60%, 50% capacities and because you're having so many spikes on the GPUs over time, it just wears it down. And so because of that we can say how do we actually prevent those types of things, how do we prevent spikes in different issues? So it's all on the operational software front and making those optimizations.
Sam Awrabi :Yeah, I want to ask you about this too, because when we've, you know, my AI fund, one is constantly looking at new opportunities and infrastructure. That's AI native, so it's essentially improving the efficiency of X problem, whether it's inference or cooling or many different areas. So, kind of. From your experience, though, we've traditionally seen a GPU supposed to last seven or eight years, this is actually the first time I've heard someone saying, hey, actually we're seeing it last a year. That's a huge issue. So, like taking a step back.
Sam Awrabi :One of the LPs in my fund, a friend of mine. He runs a really large debt financing service. I think you've met him. We won't say his name out of respect for him, um, but you know they're in like the billion dollar range of financing per year and, uh, you know, so they're loaning out a billion dollars a year so companies can buy gpus. They take the risk. If it all goes goes to shit, then they have to take over the hardware and resell it. Really interesting business, um, but that would keep him up at night if he knew any of his people that he was financing were only going to get one year out of that equipment. I mean, that's an automatic.
Sam Awrabi :Let's not work with that customer yeah because you, you, you need this thing to be in service a long time. It's like buying a brand new car and then hitting it on the curb and smashing a wall and buying a new one and it's really ineffective. So tell us more about that. Are you really seeing that?
Margarita Groisman :So this was a specific case and the reason is is when you're spiking the load so much and the GPU is just running really hot and running into issues consistently. So you spike the workload, it runs hot and then you stop again and that's pretty common for training workloads It'll go dark for you know a couple of days and then suddenly again it spikes. If you're doing that for you know a year on end, yeah, that life cycle or lifespan of the GPU is going to be not great. They're meant to be running at small consistencies, at around 40% capacity, at consistent energy levels, and that's kind of how you increase that lifespan and make sure everything's going to work properly long term. But, like I said, if you're running large training workloads a lot of the time, you kind of have a lot of maintenance issues on those servers and GPs.
Sam Awrabi :Do you think in a world where because we're still early days in AI, but as more workloads come online and specific inference workloads and you know like probably the most clear example was when OpenAI released one of their prior models Clear example was when OpenAI released one of their prior models, they had something like an insane number like millions of signups per 10 minute intervals and I'd imagine their compute. I mean, he said our Sam Altman said our GPUs are melting and they were only constrained by GPU availability to drive new user growth. Do you see a world where that continues to sort of degrade the lifetime of the GPU or do you think it's mostly going to stay in the training realm?
Margarita Groisman :I mean it could be for the inference realm, especially if you just don't have enough capacity, then you're really just spiking the capacity on what you have and then you're kind of running into the same issue. It just depends on how much you have and then how much demand you have. And I think, as you know, nvidia and every new chip comes out, suddenly the demand spikes for that specific compute level and the supply still hasn't caught up, and so you see initial problems with that specific GPU, and that's a pretty common scenario, until suddenly the demand fits the supply of the chip and now you have enough and now you're running at 40% availability of the over 40% workloads, and so I think it's going to be a continuous problem. And that's just kind of the cycle of chip deployments or how they've worked so far in the last couple of years.
Sam Awrabi :Yeah, that makes sense, shifting gears. I'd love to you know, go on a more personal note. So you're a Y Combinator founder, you're a female founder. You're in an extremely complex space in terms of complexity of software, complexity of sell cycles. When did you know you would be a CEO? Like when did when? In your journey?
Sam Awrabi :You know, I've kind of noticed like founders they kind of had this aha moment, maybe when they were kids or at some point where they're like hey, I know, I want to jump into things. And you jumped into a space where I'd argue it's male dominated and you're doing it and you're in a for listeners context. I've worked with dozens of CEOs over my my career and it's hard to find someone more resourceful than you and diligent and just a doer. You know, no talking, just you know, when we made that investment into, into Aerovolta, immediately you were tapping my network. You were looking for any sort of feedback on the sales process that you could find and just trying to bring the company to the next level with such high urgency, and that's rare in a founder. I mean that to me is what makes like a compounding difference over time. But I would love to just learn more about your background. You know from childhood till now and you know, just learn from you about that.
Margarita Groisman :Yeah, I mean I think there's a lot of things. I think I knew I was going to be start my own company when I was pretty young. I didn't really have the idea, I didn't know what like CEO was or kind of like what a startup was, but I just had an idea that I eventually would want to run my own team and kind of I pictured this beautiful scenario of how my life would be. But I think so there's a couple of things from my childhood that I think made a big difference. One my parents immigrated to the US right before I was born, so my dad from Ukraine and then my mom from St Petersburg, and one of the biggest things they always would tell me and teach me was extreme gratefulness and extreme can do. They also happened to be Jewish and their family had gone through so much during World War II, and so one huge component was, like they always said that I always felt so privileged to have everything I did in life, even though you know we lived in a nice town and it was a great spot in New Jersey. It was just like the idea that I had the world kind of at my fingertips and that's something they taught me and I think a big part of my childhood was letting me be really free, maybe a little bit too free, like I started walking to school when I was maybe like seven or younger and just like I would walk around barefoot and I was just kind of a wild child. But they really taught me that if I did whatever I wanted and made my own achievements, then I would kind of see the rewards from that. And they never kind of over overpraised anything I did. It was all based on my own achievements and so I kind of grew up with this feeling of, yeah, like you can just do that or I could just try to do this and like see what happens, and kind of making my own decisions. So I think I just never had that fear.
Margarita Groisman :In some ways, I always thought things would eventually work out if I tried hard enough. And I always took the approach of what is the worst case scenario? Like what's the worst case scenario for quit my job and the startup fails after two months? Ok, I'll like get a new job, I'll figure it out, it's, it's fine.
Margarita Groisman :And I always kind of take that approach when I think of something really challenging I have to do in order to make my startup work and so, yeah, even just being what you said about a woman in this space, it's like, I think a lot of, I think it's been an advantage in a lot of ways, just because you're able to stand out really fast, even in like conferences and booths. You're like one of the only ones in like a sea of people and it actually, in some ways, is helpful. Actually, in some ways is helpful, and so I always just I'm kind of a glass half full type of person and I always just think like what's the worst case scenario? And so just go ahead and do things. I think maybe it's a good trait, maybe it's a bad trait, but that's kind of how I've been for a long time.
Sam Awrabi :Yeah, I personally I think it's an excellent trait and looking at what you can do versus anything in your way and then giving yourself that sort of relief of if it all doesn't work out or doesn't go as planned, worst case scenario, we'll be okay. And I didn't know you're Jewish for one or come from a Jewish heritage. I didn't know your parents were immigrants. My dad's actually from Lebanon, okay, yeah. And so you know he grew up sort of poor and Saida is where he's from Many, many brothers and sisters, and you know he just sort of scrapped his way along in school and did really well and the civil war broke out and he had the opportunity to work abroad and then eventually apply to colleges in the US.
Sam Awrabi :He's the only family member, family member to make it to the US. So the Arabi name I don't think there's any other Arabis in the US. And basically bringing all this up, because it's like similar to you, he'd always emphasize, uh, and my mom just independent thinking, think for yourself, own your decisions, and like it's all good, you know, just have confidence that if things don't work out it will be okay. Um, I'm curious, what'd your parents do like? Did they work in technology at all or where, did you kind of get involved in computer science, and, and you know how did that journey go for you yeah, I mean.
Margarita Groisman :I mean so my dad. Actually he didn't like speak English when he came, so he started out doing these like base electrician jobs for a while, um, but then eventually he made his way up and then became a software engineer at Bear Stearns and then worked in like banking and hedge funds, yeah, so eventually he learned English somewhat and then um kind of grew from that.
Margarita Groisman :Then from banking and hedge fund software engineering he moved to like government contracts and started doing a lot of that. So he kind of he just pulled himself up by the bootstraps. I hate to be a cliche, but he really really did. And like I'm so proud of my family, especially because it was really hard for them, like I can't imagine moving when I'm like 30 years old to a place where I know nothing. And then on my mom's side she similarly started with some kind of I think she was like a cleaning lady for a bit and doing things like that. And then eventually she also made her way up. She took some classes English classes, computer classes while she was here, and then she became, she started working at nonprofits. So she would do like nonprofit operations for a long time in New York.
Sam Awrabi :Very cool. Yeah, my, my dad's journey is not too dissimilar. I mean, he, he, I think he was like a shoe salesman in college. He went to Fresno State and you know he taught himself how to type really quickly in college. He went to Fresno State and you know he taught himself how to type really quickly in English. And that his initial job outside of Lebanon they brought him on to do just that because at that company the person who was responsible for that sort of she became pregnant and so they had this opening and that's how he saved up money and then came here and, uh, you know, eventually he got his degree in civil engineering and became like a home builder. Um, but anyways, I mean it's just cool to see that, that upbringing.
Sam Awrabi :I think immigrants tend to look at the glass how full, be optimistic and work just extremely hard, no excuses. And it makes more sense now, uh, your approach to running your startup and I noticed it just from the first call where I met you and just seeing that carry through from the time we've worked together so it's really cool. Is Jack's family, your co-founder, immigrants at all, or? Or? Uh, and how did you meet him? I mean, I've always thought that was an interesting question, because YC is like this. I mean, it's a juggernaut. You got like hundreds of companies a year and some of these founders have never even met before they've done it, and they pivot 28 times. So I'd love to learn about, like, how you and Jack met and then maybe we can talk about YC a bit. But yeah, I'll hand it to you on that.
Margarita Groisman :Yeah, I mean, I think Jack's family has been here for a long time. They're from.
Margarita Groisman :Indiana and I don't know where from that. But I think a lot of people underestimate Jack because he's this like really nice soft-spoken white guy from Indiana but he's like a killer and I don't know really where that comes from from him. He would have to probably speak for himself in terms of where that like intense motivation is, to probably speak for himself in terms of where that like intense motivation is. But it is uh pretty crazy in terms of how deep people go and, um, the way he works is really different than me as well. He like needs absolute silence and he'll just like lock himself in a dark room and just like tackle a problem for 12 hours straight, just like on the computer. I can't do that. I need like a lot of like meetings and to talk to people and I like kind of like to work, talk and work through things.
Margarita Groisman :Yeah, exactly, but he just kind of like tackles problems. So yeah, he can speak for himself in terms of where that motivation came from, um, and then in terms of how we met, we met like a while ago super, super randomly, it was like at a hackathon, um, but it was just I was living in Seattle, he was living in Indianapolis at the time, but we happened to both be in Berkeley and just meet.
Margarita Groisman :So it was very random mutuals, but we stayed in touch. We had worked on some random side projects together when we both ended up moving to SF and it was more so like we were really close and so when I was showing him stuff I was working on when I was still at Microsoft, he basically got really interested in the space with me. We started I think we started with the conference and we sort of it was just a very natural oh OK, let's start building this. We didn't think of it as a company at first, we were kind of like thinking of it as a project almost, and then it sort of just evolved on its own.
Sam Awrabi :Very cool, yeah, and, jack, I agree, I mean the precision there is impressive. When you both were going through Y Combinator, what impacted you the most? Like you know, when I've looked at experiences in my life, there's always kind of like it boils down to a few things that I really took from from it. You know there's so many legendary companies that came from the YC platform and it's just so such a great ecosystem to be a part of. So, like, what really did you take away from it and maybe any other advice you'd have for founders considering it? You know, just would love to hear more about your journey there.
Margarita Groisman :Yeah, I think with YC it kind of hit the nail on the head, which is it's the ecosystem, it's you're surrounded by, first of all, all the alumni which are way ahead of in the game. As you are, they've been at it for like 10, 15 years and so you meet all the successful founders. But they tell you about when they started, just like you, and all the issues they ran into and all the mistakes they made, and so you feel like, oh, it's not just like the CEO of Airbnb or these different places aren't just these different level people, they're just like just same as you Exactly, messing up a lot of things and learning.
Margarita Groisman :Yeah, exactly, exactly, and so you just feel like it's kind of so it almost humanizes the approach to like back to this cup half full.
Sam Awrabi :What's the worst case scenario? Yc in a lot of ways helps founders. Just humanize it and say, hey, chill out, it's okay, a lot's gonna go wrong, like brian at airbnb, you know, I don't know their market cap 40, 50 billion. A lot of things went wrong there and don't, don't be deterred. It's like humanizing that. Yeah, I mean empathy for it.
Margarita Groisman :Exactly, and you also learn about so many people's mistakes, so that you don't have to repeat them?
Sam Awrabi :Yeah, exactly.
Margarita Groisman :And then you also just I think it's like you're surrounded by other people just kind of like you, and it's like in any other ecosystem. You're not going to really find that, but it just becomes a normal part of you and it's it's like you are the five closest people in your life in some ways, and so, um, if the five closest people in your life are all founders who are top notch and moving quickly and moving faster than you, then you have to move faster and you have to build quicker and you have to, um, put more time and effort, and it just kind of creates its own self-building ecosystem.
Sam Awrabi :Yep, yeah, it kind of reminds me I played college football in the past. I didn't play that much but it was something I tried hard at. And the NFL combine, you know they get everybody together and it's all the top players from each position and they measure everything and so there's no hiding and you know football is like a team sport but the combine on that, on that specific part of this whole thing, is an individual measurement and I feel like startups Y Combinator, there's like a parallel there of you you go into this and you you really want to do your best and you kind of see other people's progress. That brings out a lot of like drive and motivation and focus. And then you know that I'm guessing like the network part of it's pretty insane.
Sam Awrabi :Just like I'm a solo GP, I came from a non-traditional venture capital background. I'm a non-traditional venture capital background. I have I'm very fortunate now to have very successful, well-known LPs in my fund. But you know I built those relationships, earned that trust. It wasn't like I came from an alumni network and I think that would have been helpful. So it feels like YC really. Has that been something you've experienced Like as you've left YC, that being sort of like a really useful network to tap into. Because the one thing I feel like is different about Aerovolta than maybe the normal or average YC startup is just the enterprise complexity of your software. You know that you're building Like I don't know maybe I'm wrong, you could correct me, but how's that been for you leaving YC and just how complex your software is.
Margarita Groisman :Yeah, I mean. So that's true, Like a lot of we don't sell to other YC startups. That's what I'm saying.
Sam Awrabi :Yeah, like that's pretty much what I was saying. Like, you don't get that part of the whole thing there.
Margarita Groisman :Yeah, so while we don't, I think it's more so like there's a lot of companies that do large enterprise sales and so they kind of give you a playbook. There's a lot of things that you have internally and access to resources. And then a big component is access to your group partners, who've seen a million of these large enterprise deals and can help you if something slows down deals and can help you if something slows down or if you don't know what went wrong at xyz point. They could really really help you. Um, understand kind of what standard.
Margarita Groisman :What do most people do? Like? If I used to I never was like running enterprise deals at my previous job. So, um, this is it's kind of like reading the playbook of a bunch of other founders and a lot of people will talk about, um, there's kind of this internal knowledge base, um, and so a lot of people will share their experiences doing different things and it's kind of a gold mine of information that we have access to. So I think that's one big component. Um, the second is even there's this tool like happenstance um, you can see all your second party, third party connections, um, basically like if I want to talk to this person who in my network, across my LinkedIn.
Sam Awrabi :They need that for venture capital. Yeah, you should try the tool.
Margarita Groisman :Yeah, you get this huge group of people who you're basically one connection away from anyone in the world you almost want to meet, and so you.
Sam Awrabi :That alone is like you sold me there. Yeah, you know.
Margarita Groisman :And so you. That alone is like you sold me there. Yeah, you know. So I think there's a lot of effects from that that you could. You can get access to a lot of people you wouldn't be able to otherwise.
Sam Awrabi :Yeah, and then that just makes your journey to getting to where you need to go so much easier. And warm introductions and tapping, that is just such a more efficient way to build relationships and do business. Basically, on this note too, like I feel like with investors and founders there's this constant sort of relationship where, hey, we, we invested in you, you wanted to get things right. So when things aren't going right, there's like how much do you share versus not share versus go to advisors or your internal C-suite? Is that dynamic slightly different at YC? Because they've already kind of funded you. They might lead rounds at C or A, I'm not sure, but you're not necessarily looking for them to do anything financially, it's just help. Is it a different relationship than maybe if you have a tier one fund who put a check in your seed? You want them to lead your a. You don't want to necessarily go to with a list of problems, like I could be wrong about this, but it seems like it might be a different relationship on that front too yeah, I mean.
Margarita Groisman :So. I think there's a couple things there, which is one. I think investors like I like to look at them as resources, like I probably asked you wait for too many stuff and like can you connect me to XYZ?
Sam Awrabi :I wish more founders did that because I want to play that role. But you have to, like you got to throw me the ball to catch it so I can catch it, or vice versa, and so you can only help someone as much as they want to be helped. And you that's what I was saying earlier like you aren't phenomenal at that. Like and I think that's a great example to other founders Like, if you want to do this and be the best, you have to go in there and like want everybody's help that you can possibly find and be kind of, be shameless, like hey, I'm here to get things done. Like can I get X, y, z? And I wish more founders did that honestly and I do that.
Sam Awrabi :In the VC world, it's just constantly looking for GPs who raised 100, 200, 300 million dollar funds and like, or LPs, like it's just being resourceful, so important. But I guess for you is like, on the whole question, like the YC relationship, is it slightly or did you do just approach it the same way as like investors? Is it all the same to?
Margarita Groisman :you yeah. I mean, I think there's another component which is you have like a weekly office hours or biweekly.
Sam Awrabi :So structured.
Margarita Groisman :Exactly, and not just there's one-on-one with your group partners, but then you also have your whole. You have like group office hours, which is basically a section that you get really close to throughout the batch, and during those group office hours basically everyone shares all their problems and what's not going well.
Margarita Groisman :What's working, what's not working Exactly and it just feels like a very like intimate and open place that you could just share, and so there's no feeling of, oh like, let's pretend everything's perfect, because obviously it's not. You're going to run into tons of different issues and um like yeah.
Margarita Groisman :So I think it feels very open, both with the group partners and that's something I definitely want to um start doing with all my investors basically, um, feeling like the investment is there, um, it's, it's more so like how can you help um like bring us to kind of the next level? And yeah, I don't just see that with YC as being the only investors where I could be open with. I think we were really diligent in who we wanted as investors and kind of the type of people that we wanted to work with, just because we knew you'd be along the journey for the long ride.
Sam Awrabi :Yeah, it's going to be a long time to be working together and so like.
Sam Awrabi :with that, who are the people that we could kind of be it joking, but seriously and he's like you know, sam, most lpgp relationships last longer than the average marriage in america. I'm like, wow, one, you're right. And two, it's the same when you invest in a founder, you know it's like this is a 10-year journey. Sometimes it's like longer and yeah, so I feel like the the core at the core to that is like trust and respect and like transparency. Yeah, and like I love how you're like thinking about that and and I think it's like just good, good advice, and it's actually similar to being a gp with your lps. It's about, hey, what's working, what's not. What decisions were made? Why were the decisions? Like trust is, I think in this AI world where so many things can be made easily, faked, easily whatever, like trust, is the currency going? It always has been, but like going forward, I think it's even more and more important because people are able to just produce so much more content or insights or whatever it is. Do more like trust is like really big.
Sam Awrabi :So, on that note, I won't say who your investors are. If you want to share that, I mean I'm an investor, but besides that, I'll leave that to you. But how did you pick your investors? Like, and I mean it's a really impressive group. So whatever you want to share there, feel free to and like. What was that process like for you and jack, and maybe examples where you decline people you don't have to say their names, but what was that like? I mean, I don't feel like there's a lot of online talk about that and I feel like you've done a really good job of that, and I think other founders and listeners even lps or funds of funds or investors would want to hear how to get into what I think was the best YC company of your batch, maybe of this whole year. You know I is that's hard to do, so I think you could kind of share some tips there for the listeners.
Margarita Groisman :Yeah, I guess there's a couple of different things, which is like, for example, our first meeting, you were like you had run sales at most small companies. You had been through enterprise sales. You had closed really big contracts, like you knew what it was and like how to do it. And then also a lot of your companies were extremely related to us and useful for us and potential customers, and so for that, that was like okay, this is like an obvious person we would love to bring into the cap table, and I think every investor we decided to bring in had something different to offer Some people. We just really like them as humans, honestly, and we were like okay, that whole trust me right?
Sam Awrabi :Yeah, exactly Like, just being a good person and having a good reputation goes a long way. Yeah, inlikable.
Margarita Groisman :Another set of our advisors. They're super, super connected in like the whole mining space, which is another side of data centers that we are looking at and think are really interesting, and also our investors into data centers themselves. So the actual operations and so every person we kind of brought onto the cap table where we kind of had a vision of how they would play into the company overall and that's kind of um the decision making we had and obviously we had so much inbound and, like it was, it was pretty difficult to just navigate and have so many meetings like beyond how much was it like like inbound wise, like I and I think that I think most investors know this but when a company is desirable, what it takes to actually get in that investment, and then you know to lead around in those investments, what that looks like, you know, I feel fortunate.
Sam Awrabi :I don't have to get anyone's approval. I can meet you and say on the spot, yeah, here, I'll write a check today, and so for me it's like speed and that expertise, like that is what I bring. And it's the network effect of all the other founders. I've backed you getting access to all them. I know what they're doing, what they're not doing and can give good guidance. And then my own experience is being an operator scaling a company from C to series B and all those things are great. But for the listeners, like what was that? Like the inbound in, like how? Yeah, I mean, I just I, I kind of know it, but I don't think most people actually understand what that looks like from an investor or founder point of view yeah, I would say we got like probably three to four emails and like three to four more, like maybe two to three LinkedIn messages a day for months.
Margarita Groisman :And that was like even like when I had really early on like I'm quitting Microsoft, starting a company, reach out if, like, you're working in infrastructure, like that's when it started. And then it just kind of increased in scale once we got into YCN. So, yeah, like even in terms of like who are we actually going to meet with? Do we want to just do like 30 minute meetings with everyone or do we kind of want to be more selective here?
Sam Awrabi :Which one did you pick? I'm guessing you would. You just said no, we're going to filter out so much based on our credit. We filter out a lot yeah, and then go from there.
Margarita Groisman :Yeah, like we didn't want to talk to, for example, like an analyst who reached out from like, where we would have to have like four meetings at the firm to like, talk to like the decision maker. So, like a lot of that, we filtered out and then so that kind of started that was helpful and then filtering based on we could look at different kind of like reviews of VCs.
Sam Awrabi :How do you look at that? I'm curious. Yc has a way to do that. Yeah, Is there anything about being in in there?
Margarita Groisman :Um, we can talk about it off the podcast, but we're leaving a five-star review.
Sam Awrabi :Okay, there we go, all right.
Margarita Groisman :Um, yeah, so then, yeah, we could look at reviews of investors, and that was a big filter.
Sam Awrabi :Learning so much today.
Margarita Groisman :Yeah, Um and then yeah lots lots of, lots of different things there. So I think that was the main thing we filtered on, based on what other people said, um, and then yeah, basically can we get to the decision maker quickly because, like I, precede.
Sam Awrabi :Like you're just running on such limited time and I and it's like I don't know, can't comment on other funds structure, um, but like that was one of my frustrations and why I wanted to start a fund was I just saw, like the founders I had worked with when I was ahead of sales, like how much time they wasted educating a general partner so the general partner can then go educate yeah, the five other dudes who don't even know anything about ai, and then then they just come back with more. It's like this never ending time suck. And that's the majority of funds. They're not. They're not born in the ai native world. You know the most funds today. They do build some build like andreessen does a great job. They have an expert team doing ai, infra and ai and I think they they have a really proven model to build like a 40 50 billion dollar aum with expert teams in each kind of defense crypto. But a lot of funds don't do that. They're like trying to do defense crypto, green tech, and I'm like how the fuck, like that's such a bad founder experience.
Sam Awrabi :Like to me there should be this like really clear what is needed to make the decision, how long will it take? And like precede, you're already just running on such limited time and that was part of the reason I started banyan and a huge kind of value prop to the LPs and to founder. I'll make a decision quickly. I've seen AI applied at Uber, at Tesla, at Cruise. All these different people I've worked with have real relationships with that I can contact and then I've literally helped them establish their experiment and model management kind of software stack internally. So like that's the insights I'm relying on and you just got to like make that decision.
Sam Awrabi :But do you feel like in the future, say in like 10 or 20 years, how do you think venture capital will be rearranged? Do you think we'll see a lot more like specialist funds or do you think the mega funds will like continue to thrive like the Sequoias and Dreesons, like I don't know? I'm just curious from a yc founder point of view and just what you think about that um, I think maybe like the top of the line, like the sequoia and dreesons they survive yeah, they kind of have a big um advantage there.
Margarita Groisman :But then those like mid-tier where you still have to talk to a lot of people and then are using the older model, I think it's these generalist investors.
Sam Awrabi :They're're just somehow doing it all, yeah, at once.
Margarita Groisman :And it's hard when you're a founder, because if you're spending all your time raising like, when are you actually building a company? And so it's like same being, by the way a GP.
Sam Awrabi :Like every hour I spend fundraising. I can't spend helping you and I can't spend trying to find the next Aerovolta in a different space.
Margarita Groisman :Yeah, yeah, that's like exactly, and so you really need a time box, or that was the decision we made, like, ok, well, let's just time box we're going to do like funding in this period of time, even then, like let's still have a portion of the day to like work on the company. Let's time box like all the talks within a week and then kind of like the closing and negotiation, all that stuff can like happen in the next couple of weeks, and I think that way you really have control over the whole process.
Sam Awrabi :Otherwise you're just in these constant loops of yeah, Like whatever the person's mood is at the time. Exactly Like if they're feeling optimistic, it moves quick If they're whatever skeptical. So you just time box it and pick the investors who can match that, and then people who have a network and judgment that can help you, and everything else just gets put to the side yeah, exactly how does that change for series a?
Sam Awrabi :because you know series a is more money, probably a board role, probably like pretty custom terms that that a maybe not, maybe it could be a safe. I mean it doesn't have to be a price round. But how do you think that changes for you? And it's okay if you want to say, yeah, we're, we're not sure, we don't talk about that too much, but not to put you on the spot but like, how does that change, you think as you go to raise bigger rounds, sort of that dynamic from a founder fundraising point of view?
Margarita Groisman :yeah, I. I mean, I think it's still in some ways the same thing, which is, even though you're raising a series A, like you don't want that to be a two month process, like the company still needs your time and lots and lots of attention, and so if you the company is suffering because of your series A, you've made a mistake in my opinion. So that's like. So I think that kind of same logic applies.
Sam Awrabi :Obviously, at that point there's a lot more legal complexity, especially a price round, and there's a lot more terms that could really change your future yeah run those by me when you do that, because, like I've seen some, wow, there's a lot out there in those terms and I mean usually it's founder friendly, but they, you know, it could sometimes unfortunately not be founder friendly when it comes to price round terms yeah, and I think a big thing is you don't want to be in a position where you're you're kind of uh like you want to be in the position where you can easily say no to anything you want to be zuck?
Sam Awrabi :exactly um zuck is running like uh, I don't know their market cap 40. No, it is. Do you know it? Meta's market cap is like 400 billion, trillion, I don't know. It's in the hundreds of billions, and he still runs it like a three-person startup.
Margarita Groisman :That's bootstrapped yeah, that's who you want to be zuck you don't want to be.
Sam Awrabi :Hey, I need your approval to hire this uh vp of engineering. Hey, I need your approval to uh fire this person.
Sam Awrabi :I mean, I've seen stuff like that, I mean painful, like it just slows things down and right, and I mean in the US we have safes and we have a culture of let the founder be the founder. I think that you know, as my firm transitions to fund two and leading more rounds, you know that will always be my thing Let the founder be the founder. The founders should run companies all the way until the company doesn't exist. I think that nobody knows how to get things done like the founder be the founder. The founders should run companies all the way until the company doesn't exist. I think that nobody knows how to get things done Like the founder. Nobody understands the lineage of decisions like the founder and, like you know, peter Thiel talks about that.
Sam Awrabi :Non founder on companies often perform worse. They don't innovate quick enough and at any given time, there's like 10 things that could kill your company. The founder will always be the one who could make that and it's like I think VCs get that wrong and Vinod Khoshla talks about that like 90% of VCs add negative value. They literally like mess up your decision making because they're like getting in your ear about things and asking for insights or data and it's just slowing you down. So I definitely think you know these are all the things that I've tried to use as first principles when working with folks like you and I, you know, want to keep building on. But that makes sense. About the series A, just you're going to time box it same kind of philosophy there and put the company first, you know, so you don't slow things down too much.
Margarita Groisman :Yeah, yeah, and I think the biggest thing is making sure you're raising when it's like a good time for the company and then when there's a lot of headwinds kind of, and not when you need suddenly you're desperate for cash, right, so that's. That's kind of, I think, how how we want to go into it.
Sam Awrabi :Yeah, I mean so much of raising money is like psychological and leverage based and I don't. This whole like fear of missing out is kind of like talked, yeah, talked about a lot. I'm curious your thoughts. I think it's like it's it's just about value, like if you do this, there is an inherent upside value. Yeah, exactly like. But how much do you think about macro versus that whole equation?
Margarita Groisman :Well, any like business deal or sale or anything. It's all just humans making decisions and humans kind of have different like emotions and needs that you have to work with. So I think that's one component. But then the other component is just an equation of OK, if we close X, Y, Z and then we get this customers and we were the company, this this much, how much can we make in profit and does that make sense for you as an investor?
Margarita Groisman :um in terms of, like, where the cap is and different things. Um. So yeah, I think it's just kind of balancing, kind of yeah, the, I don't know the fomo, the, but also, like, is the company really really strong? And I think that core principle of just focusing on the company, it kind of just brings what you need not focusing on, like, how do we create FOMO, or how do? We do XYZ.
Sam Awrabi :It's just focusing on the company yeah, I mean it makes me think about Adam Neumann and uh, some say he's like the best. Uh, he's Israeli. I think he's Jewish. I love him personally.
Sam Awrabi :I lived in Israel. I'm a big fan of the Israeli culture and everything, so. But I mean some people say he's like the king of fundraising and I think that there's a lot to learn there, Like how do you get into a fundraise and get out and have options and do it on your terms? I mean that is a part of like if you went on a piece of paper and wrote the six most important things of being a CEO, I'd argue like fundraising is like very high up there, because if you run out of money, you're done, and most startups can't just do what they need to do without fundraising. It's still like the world we live in.
Margarita Groisman :Yeah, yeah. So I guess, in terms of the question there mean in like fundraising was, like I said, it kind of just was brought on by focusing on the company, like we started with building a product and we started by trying to get our first sale and pick the right problem exactly 95 percent told you we need this, we're open to it.
Sam Awrabi :I mean that alone is do you feel like that's like the most important thing when you're getting going is like picking the right thing? Because you've been through YC I mean I've heard stories about, I've met a lot of YC funders like the and it's not just a YC thing, just in general the amount of pivots that can happen. Even investments I've made that didn't go the right way how I had planned. Luckily it's been very few but it's happened. Like, what do you think is the key there? Is it picking the right thing? Is it picking something you love? Is it solving a big enough problem? Is it market timing? I mean it's an arbitrary question, but what do you think? You could get right or wrong there pretty easily.
Margarita Groisman :Yeah, I mean I definitely don't have the magic equation for how to build the perfect.
Sam Awrabi :I mean, you got it right for AeroBolt at least, yeah.
Margarita Groisman :I think it's um, like there's a lot of uh, you have to be really humble in terms of like, oh, like I know this is going to be the best thing in the world, like I have to build it. And then, even when you get like initial traction or anything initial it's, it's like can you maintain and continue to have make actual, like physical money, like revenue? I think is kind of just, it speaks for itself. And I think, yeah, you could just pivot a million times, but to build something people want, it takes time. Like you have to iterate and you have to deploy something and have someone say, oh, this is almost there, but we need X, y, z.
Margarita Groisman :So I think it's one really just listening not speaking too much, and like listening to what people are really telling you and kind of what's below what they're telling you. Okay, we're saying you know we can't really change products, but we really dislike our existing software system. We honestly haven't even been able to get 30% of our hardware onto that system. Okay, what they're telling you is, if there was something in the market that could do that, they would switch in an instant. They just don't believe that there's something in the market.
Margarita Groisman :Um, so I think one thing is listening to what people are saying, and then two is just iterating on your initial thesis, um and, and getting to kind of a point where people are saying hell yes to whatever you're selling. And I think, like I said, you could pivot before you get there, but a lot of the time you just needed to iterate more to get to that product market fit. So I think there's a lot of components there. So we've tried to do the best we can in terms of listening to our customers, iterating on the product, continuing to build and just focusing on that component.
Sam Awrabi :Got it. That makes sense and it reminds me of just touching on a few things you said. It's in the back in my operating days, an individual I worked with. He had built some pretty large teams, like I think at the one point he had managed people in the thousands and he he talked about to the organization.
Sam Awrabi :Like this concept of you need this extreme belief in what you're doing but at the same time, just be willing to change your mind and change your thinking. And it's like I think that can be hard Cause you're, you're so not you, but like, and just as you're a founder builder, you're, so you're believing in what you're doing, but you have to be willing to like, listen and change your mind. But at the same time I'm not an expert on Steve Jobs, but I've heard him say things around people not knowing what they want. So how do you strike that balance of I believe in what I'm doing, I'm willing to change my mind, I'm going to listen, but people might not know what they want. Is it because they don't know what exists to your point, or do you think there's something else going on there when you're building product?
Margarita Groisman :Yeah, I mean, people just know what they've been exposed to. So if, to them, a standard deployment is, like you know, a one-year timeline, with having to bring in an engineering team on your end, like that's just how it runs and that's how in the past, that's what they?
Sam Awrabi :know Status quo Exactly.
Margarita Groisman :And so if you could change things to like not just like a, you know, a 10% improvement, a 15% improvement when you see a 90% improvement, that's when people didn't really picture what that would look like and they didn't really like know to ask for that um. So I think on that end, you just have to um really like, just have the belief that you will bring that that's like 10x improvement, not 10%.
Sam Awrabi :10x like yeah, 1,000 times better than the existing way of the world. At that time, you have to be able to see that and then build it and like, adjust it, and then have the conviction to do that, even though status quo might say it's literally impossible. You're wasting your time, yeah, and like a lot of people will say, um, yeah, yeah and.
Margarita Groisman :But also I think there's a lot of noise. Um, especially when you're building a company, there's like so many opportunities and so many people have different opinions on so much stuff around the product sales. Yeah, there's just like a lot of noise and, I think, being really structured in your thesis. And then what noise matters? Because some does. Some customer feedback matters, um, certain people who have been building data centers for 30 years. Their thoughts matter, um, but really cutting through everything else, it doesn't matter.
Sam Awrabi :Yeah, it's like, when you get feedback that challenges your understanding or threatens your understanding or your mission, do you take it in and adjust or do you ignore, like that's the key, that's like to me, the judgment that's required to advance to that 10x multiple?
Margarita Groisman :of value creation.
Sam Awrabi :Yeah, I wanted to ask you. I mean, there's two things we've sort of touched on. I'll touch on the first one. So this concept of in the AI native world and I've definitely seen this I mean at some point I'm going to release a data report on it. It's probably in the future. It's not now on it, it's probably in the future. It's not now, but in the AI native world revenue per employee is astronomically higher than the software world of the past. The growth rate, the top line revenue growth rate, is also astronomically higher. I think Olivia Moore and Andreessen Horowitz released a report, something like for B2B enterprise six months into selling and this is selling. So it's not like into your company's history, it's like from when you said, okay, the product's ready, I'm going to sell it six months in the top quartile. And they don't define top quartile, they don't say if it's top 1% or 5% or 10%. But the point is like they say top quartile.
Sam Awrabi :Now in the AI native enterprise world which you are in six months in, you're around at like 2.3 million in revenue and 12 months in you're at, I think, like top quartiles, five million in revenue and on average it takes like six or seven months into selling to raise a Series A and you're only raising on average top quartile two or three million dollars before the series a so just on, like unpacking this paradigm shift in the ai native world just for enterprise, because you're an enterprise founder is like me as a vc, and andreessen and, and I think a lot of others who have adjusted to this is like we're looking for companies six months into selling at two million in revenue, one year into selling at five million in revenue and, to your point, exactly what you've done in Aerovolta. You're like three people, so you've raised maybe a couple million before you're gonna go gear because you're not raising anymore. I don't believe you will. You don't need to. Yeah, before your series a. So you, you, my view check every box and I won't disclose any confidential information.
Sam Awrabi :But I mean, how do you see that kind of changing? Like, do you think like if you're not in that bucket now, you can't fundraise? Like do you think for those median or lower tier enterprise AI native companies, like they just kind of miss out because the data is showing uh, also, like in total there's less rounds, um, so less people are getting a shot on goal at pre-seed and the bar is so much higher. And I've been trying to explain this to founders don't go raise a pre-seed until you have traction. But curious to your view like was that talked about a lot at yc? Um, beyond that initial yc check, you get like get the traction first, or how did you see that? How do you see it evolving?
Margarita Groisman :Yeah, I mean, I think the round dynamics and like how people raise is really different for every company and it's kind of like making companies a journey. So in terms of like where we were even like six months ago or one year ago, it looks really really different. Um, even like on a month by month basis. And I think, um, yeah, because of the ai native world and how companies they have access to markets or people who want to deploy capital to make things significantly more efficient, um, and there's kind of like grabbing that market share and I think people who are able to do that build and expand really quickly.
Sam Awrabi :Like founders selling to the like you're saying the appetite is so big, that's why. That is why this growth is. In one year you got 5 million in ARR, like I come from, like AI infrastructure, ml ops before the AI Native Fund and Banyan. But I mean if you could get to a million in revenue in two years of selling, you were like the talk of the town back then. Now that's like kind of like that would be like you're mediocre. So like when you're saying that's really coming from the appetite of the customer, it has to be right. They're the ones paying for this stuff.
Margarita Groisman :Yeah, but I think I mean there's another component which is like Figma didn't even like start making money until five years in or.
Sam Awrabi :UiPath, there's so many examples. But do we live in that world? That's what I was asking. Like can there still be startups that took five years of R&D? Obviously, investments we have, like I think the top one like in terms of patent number top, that's the metric is 50 ish patents. That company needed a lot of money and time to get to that level of R&D and physical infrastructure historically is hundreds of millions of dollars. To do an ASIC chip tape out is at least six years and 40 to 100 employees, maybe more. That being said, companies like Positron are changing that entire narrative like they're getting. Some are skipping ASIC chip design or they're at least getting it done in record time with even less financing and they're getting to significant revenue. So even there that paradigm is changing. But I guess my question to you is do you think we'll see generational companies that took multiple years to get going and selling, or is that gain all that knowledge so quickly In the past it would take months knowledge that employees have and are able to work.
Margarita Groisman :Like basically, I think all of our employees are like 20x because they use AI in their day to day, 20x yeah Than what they would have been pre-GPT release. Exactly In terms of just their efficiency and so you can move and operate a lot faster. But I do think, in order to build a product that people really want and to get product market fit and to scale quickly, it does take time it takes time.
Sam Awrabi :It takes especially for, like what you build, like we talked about it, right, these low level integrations. Nothing's the same on a even on an account basis. Each data center is different.
Margarita Groisman :There's so many generations of hardware, exactly so it take, that takes time exactly, and it takes time to um kind of build a product that doesn't. You know.
Margarita Groisman :You don't have to manually onboard every customer at a time right exactly so I think certain things take time and certain things are 20x faster. Um, and it's it's kind of like putting your artists in touch on a product. It's like really caring about those small details, the performance and making sure everything's fast, but also the actual like cutting down all the stuff you don't need in the product, because XYZ said it would be really nice to have this thing and this thing, and really cutting it down into what is that core problem statement that your customer has and how do I fix it. And so cutting it down into what is that core problem statement that your customer has and how do I fix it. And so cutting your product into a set of features that work, really, really really well.
Sam Awrabi :The cloud MVP yeah, exactly.
Margarita Groisman :And having those be deployable in those easy to use, low time to value. Yeah, and then, especially with data centers, you have a lot of yeah, and especially with data centers, you have a lot of you know like, basically you need to be SOC 2 compliant and get a lot of compliance and security and and those type of things, and that also takes time, um to make sure, like Vanta I think can only do SOC 2, I don't know I last I had done it on like been involved.
Margarita Groisman :It's like still six months, usually soup to nuts well, delve, you can get it, I think, in like a couple months.
Sam Awrabi :But then you're type two. Oh yeah, like how about sock two? Type two, is that still a couple?
Margarita Groisman :you still have an observation period so that takes a couple months, but the actual like uh type one is is much faster now okay, well, that's good to know.
Sam Awrabi :so, um, so, in like, in summary though you you're it sounds like you're leaning towards, it still takes time. There's an artisan touch. There could be companies that took longer than this top quartile six months, two million, one year, five million in enterprise AI, native software. There still could be decacorns that didn't meet that mold. But you would argue, on the aggregate you're around 20x more kind of efficient and it allows you to get to an MVP faster.
Margarita Groisman :Exactly.
Sam Awrabi :And so that should significantly tilt the skills. And on top of all that the appetite is so large if you're building something in the right arena, which you, that really helps too.
Margarita Groisman :So it's like all these variables you'd almost say in most scenarios that that data from a16z looks correct to you yeah, and I think another component is like yeah, you can scale up your revenue really fast, but does it turn like we've seen?
Margarita Groisman :that's true right from application companies yeah especially consumer and like these ai sdrs, like it sounds amazing to have invested in that like right, yeah, but you know, if it doesn't work right and the customer doesn't love it, they're gone and that revenue just quickly dissipates. So it's how do you have something that's actually sticky, staying power exactly like.
Sam Awrabi :So for me, what I look at um you. I have five or six investments in AI native vertical software. I look at how deep is the workflow ownership of this AI native startup. Like I have an investment in Finalis, the fastest growing investment bank in history. I mean they're keeping the investment bank compliant, so they're not going anywhere.
Sam Awrabi :The investment bank compliant, so they're not going anywhere once. Or, or Lex room AI Uh, I think it's probably the fastest growing software company right now out of Italy, but they're, I believe, at past a thousand paying customers. Um, they've grown revenue like at the top 1% does sell. Since we invested in um, they'd provide a legal research platform that's completely custom to Italian case law, which is complex and hard to understand. So if you're a lawyer, that's 30% of your time Once you start using that and it's very high accuracy.
Sam Awrabi :The churn's been very low there and you know WandaEye is another example. They're working with like Precite out of the Middle East on a really they have like on a lot of large deals. It's like a co-pilot. So, for you know, franklin Templeton or customers like this are going in there to do their knowledge worker work, but it remembers everything and it's compliant because, you know, gpt has, like data sensitivity issues and they've had memory installed in their offering for the last two years.
Sam Awrabi :So basically, I look at how serious is that workflow ownership. Is there a network effect? As you, as you get more data, do you offer more accurate offering? And I think the shift we're going to see is what starts as vertical, becomes horizontal and becomes a system of record, a system of working. In your case you're the system of intelligence and operations for the data center. Maybe your case is a bit different than like the traditional well now, traditional vertical software application companies, but you're still moving to like that operating system. You kind of wear that from the jump. But that's kind of what I've seen to avoid this whole churn thing of like the ai str is like a great example. It's easy come, easy go, yeah, yeah, exactly, um, another question for you too is like a great example.
Margarita Groisman :It's easy come easy go.
Sam Awrabi :Yeah, yeah, exactly. Um. Another question for you, too, is like, um, I guess, when you're, when you're, when you go out and you fundraise and I'm going to pull up my uh notes on this when, when you're talking to investors, like, is there any kind of like features or future facing visions that you guys, you and Jack, have that you're really excited about? I mean, I think I kind of asked it earlier, but is there anything bigger than what you're doing now that is like really exciting to you? Like, maybe the technology is not there yet, the AI capabilities aren't there yet. Is there anything like that that you've been thinking about?
Margarita Groisman :Well, a huge component is kind of that data model. Like I said, every customer now has, on a per second basis, data from every single device Coming from their systems.
Sam Awrabi :Yeah, exactly.
Margarita Groisman :And they have years of that data.
Sam Awrabi :It's stored somewhere Exactly Okay.
Margarita Groisman :It's all stored and so that's kind of a gold mine for these companies and for us, um, and because they have so much data. Eventually, our goal is to get to the autonomous data center, or where you don't say, oh operator like do xyz, and you'll save this much, or the so it is kind of like agi for the data center exactly.
Sam Awrabi :Your co-pilot would complete actions autonomously 24 7, based on a fine-tuned agent that's deployed on that data center instance using historical data. It's just not all the way there yet to start making decisions on that scale.
Margarita Groisman :Yeah, right now it still is kind of there's a human in the loop. The human says, okay, does this make sense? Like they double-check, triple-check everything and then they deploy the recommendation. Eventually, you potentially don't even need that component, um, so I think that's. It starts with the data layer. You can't just jump or skip all that.
Sam Awrabi :Yeah, like the model can't make decisions until it's been trained on enough scenarios exactly to have. I mean, I don't even know if there is room for error in this. If you, you're a Tom Simic you'd almost need 100%. No, there is no room for error yeah, and I don't even know if any AI is at that level yet. Even Waymo back to that, like maybe it is. I don't know, I'd have to check, but it seems pretty low likelihood that something's at 100% accuracy in AI.
Margarita Groisman :Exactly that's AGI. To me, like 90% isn't good enough. Basically, it's like you need all those edge cases to be accounted for and that takes a lot of data input and basically seeing what those edge cases could be.
Sam Awrabi :Yeah, do you think you're like two years away from that? Three years, is it hard to know?
Margarita Groisman :It's hard to know. I also think the future of the data center itself looks really different. Like Stargate, they're doing like what? 4.5 gigawatts of capacity, like we haven't seen anything like that before. But then there could also be completely different data centers that are much smaller than what we seem to employ and work on completely different energy types, and then separated grid, like separated from the grid, fully autonomous data centers. Like there's so much coming to the market and I think we're just going to be moving with our customers and seeing what's out there and just building the foundational layer of those operations. That applies to any type of data center in the future.
Sam Awrabi :Yeah, Like listening and building, being agile and you know, building on this already very strong foundation.
Margarita Groisman :Yeah.
Sam Awrabi :Question two for you is you know I've seen all different perspectives on this already very strong foundation. Yeah, question two for you is you know I've seen different perspectives on this and I have my own, I guess, hypothesis about it. But do you think we're in a over build out Like? Do you think it was like a Zuck said, hey, let's buy as many GPUs as we can and then we have Elon down in Texas just going absolutely bonkers?
Sam Awrabi :yeah I mean, I saw this graph like um, he built us, like this latest data center faster than the average american homes built. So you know, you have bigger deployments being built faster than ever and at larger scale. And then a lot of critics, like on wall Street or other arenas that maybe aren't at the front lines of this. Well, they aren't, um, they're maybe missing the context. Um, I mean, what's your view? Do you think there could be a overbuild of infrastructure or do you think it's still like an underbuilt uh situation due to, like, supply chain constraints, like Nvidia can only produce so many chips at a time? So I mean, I mean, I'm curious your view on that.
Margarita Groisman :Yeah, I mean I have a lot of thoughts here, but in some ways, like the US used to be the manufacturing hub of the world and then that shifted. But I think there's a lot of you know like, can we bring manufacturing back to the US? And I think manufacturing intelligence is actually much more interesting than manufacturing clothes or toys or other things, and we literally have the capability here to manufacture intelligence, which is something that can be applied to all the basic work that's done, all the documentation and writing things and kind of putting together different data points and writing Excel sheets and making PowerPoints. All of this kind of intelligence layer can be built out and they're trained in these centralized facilities, at least right now, and if that continues to be the case, then the demand for making more and more intelligent things will only increase, because there's no reason to say oh we like, we're good where, that's good enough. So I think if data centers remain in the kind of similar model, I think it's honestly probably we're just going to continue to build up.
Sam Awrabi :So you're, you're, you're thinking, hey, intelligence is such as AI driven intelligence is such a supreme value add to the world. At the minimum we keep the pace of the infrastructure build out. And now we have policy actually looking to increase the appetite, increase the R&D, increase the at home US built infra. So all these things are kind of pushing forward, but the main driver is just the value add of the intelligence being created. You don't view it as an overbuild whatsoever.
Margarita Groisman :You and at the minimum, think we stay at this pace of build out I mean, how much value has open, ai or anthropic added to your life on the day-to-day? For me at least, it's like there's a huge amount of value in terms of how much I can produce from the world because of these foundational companies. Like you said, a 20x increase. Exactly In every employee, in your people's work.
Sam Awrabi :Yeah, in yours. Minimum yeah 20x, I mean not. But then there's okay, devil's advocate. There's, like critics that say like okay, yeah, like they're trying to do this, like I don't even know the numbers, like I think it was 30 billion dollars a year. They're going to pay Oracle, but they only produce two or three billion. So, like, what do you say to those critics that are like hey, we're just looking at this like this company is going to go bankrupt at?
Sam Awrabi :their current spend, but like it's kind of a that is what it takes to get to AGI, like or supreme intelligence, or be the intelligence leader. It just takes such a such a capital intensive effort. And that was like another thing I wanted to ask you is like this talent war, um, like how are you competing to hire talent? I know you're building out engineering with like 100 million dollar offers now, but I mean it's kind of two questions in one is like do you think the critics are right about the overbuild, overspend and then talent wars, and then they could probably wrap up, but like how are you kind of thinking about both those things?
Margarita Groisman :yeah, I mean in turn. We're obviously not giving out 100 million dollar offers not yet. Um, yeah, if you do, I'll join yeah um, but I think, I think people just want to be part of something um, like and part of mission? Yeah, exactly, um, I think money, you know like, once you get to a certain amount, like it's, it's all the same after that in terms of your happiness level. I think people want to be part of something, um, something that's growing, something that's exciting, a great environment.
Sam Awrabi :It's like flat autonomy to build and make an impact and drive the decisions of this company. I mean at meta. I mean, yeah, zuck's great, I love, I'm pro meta, I own a lot of meta shares, but like I still think you're not going to get that at a huge company. Yeah, so for you that and like I was talking to a friend of mine she's a reporter at forbes last week and that was a similar answer I gave her was just like the autonomy and the flatness of the decision making and the impact and for you it's like mission. When you're, is that like kind of the big? I mean, are you seeing it in candidates like want way more money now, or is it like not kind of like that, unless you're like the author of original authors of GPT or something is it? Is it isolated? Because even in the media and a lot of VCs a lot of people are wondering. I'm curious what you're seeing on the front lines.
Margarita Groisman :Yeah, I mean like I think it's also very much dependent. If you're hiring AI researchers, it looks very different than hiring, like forward deployed engineers that are customer facing. I think, in terms of like how we hire, it's really just like this is a team that we're interacting on the day-to-day and this is what you're going to own um, and engineers love really, really hard interesting challenges harder the better.
Margarita Groisman :Yeah exactly where they could own. Basically a completely new way of doing things, um, and so if you could offer them to that, them, uh, that that that gives a big advantage I think, um. So I think having a really interesting product stack and product to work on and something completely different and a really difficult challenge, I think is always exciting for engineers.
Sam Awrabi :Yeah, that makes sense. So it's kind of on the front lines outside of the researcher role. It's not really changing the discussions too much and maybe even in the researcher roles there's so few candidates that could demand that premium. It's just kind of not overhyped, like it's a real thing, but it's not really changing too much on the recruiting side.
Margarita Groisman :Yeah, I mean like recruiting is a huge challenge and yeah, we're hiring.
Sam Awrabi :Right.
Margarita Groisman :So if anyone's watching this, but yeah, I think the challenge is more so finding that right fit for the company, the person who is going to go, you know, put everything into a really hard challenge and is willing to go really, really deep into a specific problem. That's kind of. More of the challenge is finding those right fits.
Sam Awrabi :Yeah, makes sense and on this whole, and then we can wrap up the people saying, hey, we're overbuilding the revenue, numbers don't make sense, open AI spending way too much versus what they're producing in revenue. Like Anthropolic is now raising from the Middle East, which was something they said they never would do, but they're saying now we have to to sort of be able to build the business we want to. So you're seeing, even they're like them changing their own stance that they self inflicted to have a chance at being number one. So do you think those critics have warranted claims? Or mean, how do you view that like over spending versus revenue ratio at such a big scale? Like how does that all shake out? Because even, um, I think, uh, soft banks investments, like faltering a bit, and uh, into open ai. So I'm just curious, like your lens on that?
Margarita Groisman :and you know, happy to wrap up yeah, um, in terms of the actual build, like the answer is I don't know in terms of like, what the future looks like five years from now, honestly, even one or two years from now, um, but I do think if you could uh bring a certain amount of value into the world, you, once you actually deploy, or uh like basically, release the intelligence value right.
Sam Awrabi :That is going to be you know a magnitude bigger than that 20 billion you've burned on GPUs that year.
Margarita Groisman :Yeah, exactly. So if that actually works out and we can build something that provides that level of value to the world, then I think some they're going to be very, very happy and I personally wouldn't be betting against open AI. No, having like seen talks from Sam and kind of what, they're gonna be very, very happy and I, I personally wouldn't be betting against open ai no um, having like seen talks from sam and and kind of what they're working on, I I wouldn't be betting against them yeah, and I mean we could wrap up.
Sam Awrabi :But like on my side, like the, I feel like the biggest regret I'll have in 10 years, uh, could be twofold one that I it's not about the entry price on pre-seed versus seed uh, I'm flexible on terms because to me it's like when I look back, in the next 10 years did I at least get into some of the most important companies of our generation? So that could be a regret if I don't happen to do that, uh. And then the other regret I think is not having deployed enough capital into the winners we already have back, because, I've looked back, our investments have done, I think, almost half a billion dollars in bookings from pre-seed and seed, and there's already ones where I'm like I wish even my own capital had put more in there. So it's like to the open eye point, yeah, I wouldn't bet against them and I completely agree like that capital is almost peanuts compared to the value for society coming yeah, exactly yeah, sweet.
Sam Awrabi :Well, margarita, thank you so much. This has been fantastic thank you.
Margarita Groisman :Thank you so much.