It’s a boat, it’s a plane, it’s REGENT’s Seafarer | E2205

24:24There are, you know, drones out there right now, lots of drones, which is definitely a focus of DOD. But you can win a battle with the drones. You cannot take and hold land and win a war with drones. You need people to do that at the end of the day. So you still have to move people. When you're putting people into a fight to do that, you have to be able to take them home safely. And so Medivac is also one of our fastest growing applications for this vehicle of actually being able to go out, land on the water, rescue people and bring them home. So to be clear, you're going to be able to serve airlines, net new carriers, various military applications.

25:00And also, I presume there's gonna be some rich people who just buy these for themselves. Big market. Big market. Which brings me to my last question for you, Billy, which is just, you're doing it in Rhode Island. And like, look, Rhode Island's a tiny state. No one knows who we are outside of like the Northeast. But I find it really exciting that not just Regent are here. I mean, we were talking about Vatten Systems before the call. And you mentioned Havoc and Enderal has the prisons here. What's the Rhode Island pitch? Like what types of companies and startups should be thinking about putting down some boots here?

25:29Rhode Island is sort of in the midst of this maritime defense renaissance, which is very well aligned with, you know, national priorities. And that whole defense scheme I laid out of, you know, national defense for the next decade plus is going to be water based. And so you have a lot of sort of overlapping things in Rhode Island that make it compelling. Obviously, it's the Ocean State. You have Narragansett Bay. It's an incredible testing facility. There's a great military presence here. Naval Service Warfare Center and the War College are here. So you get the top brass, the generals, the admirals, the decision makers coming into Rhode Island.

26:07And you can call them up and say, hey, we're a half hour away. Come check it out because everything in Rhode Island is a half hour away. So you can just come on over and check it out. So that's been great for, you know, engagement at the highest levels of defense leadership. One of our senators, Senator Reid, is one of the ranking members on both Senate appropriations and Senate Armed Services. So he's an incredible support to companies in the state. And being in a small state in Rhode Island, it's like he comes and visits, you know, quarterly, personally. You get so much more hands on, like compared to like California, for example.

26:40I mean, I've run into like the mayor of Providence was just like eating in my local diner. And I was like, this is this does not happen in San Francisco where I used to live. You know, like you're not as abstracted from government. It's right there. Exactly. And then I'd say the last thing is we're in what's called Quonset Business Park here. So like General Dynamics builds their their nuclear subs here. And there's just this huge growth going on because of the compelling financing that the state offers. And we're just building factories left and right. Like Quonset is the factory factory right now.

27:09Is Governor McKee sufficiently engaged? For sure. I mean, I was with him last week. super engaged in this, growing this out. The state is investing real capital into growing the manufacturing base here, all focused around maritime and defense. But for example, we are building our 255 ,000 square foot manufacturing facility that we were able to build. I mean, this is a$50 million project. We were able to build this as a series A startup that had raised a total of 60 million to date. I know. Crazy leaning in from the state. It's really incredible. So that facility will open in the spring. We'll start cranking out sea gliders for both those commercial and defense customers.

27:50We'll deliver our first sea gliders in 2027. But it's not just us. Anduril's building their facility across the street. And we have all these other companies coming in. Saab is bringing their undersea warfare center. There's a good presence from Raytheon and RTX. It is a very cool state to be in right now. I mean, Narragansan Bay is essentially one large subpen. Let's be honest. It's great. It's awesome. I love to be enthusiastic about a company and my adopted home state. So Billy, thank you so much. I will come down at some point and say hi and just see what y 'all are building. But 2027, last thing, early seven, later 27, when do these first touch customer hands?

28:28I would say second half of 27 is where we're focused right now. First half is going to be a lot of sea trials. So they'll be out on Narragansett Bay. They'll be working.

28:41They'll working on those missions as well. Second half is those 27 deliveries. Can I come on a sea trial? You got to come out. Yeah, come on over. Just walk, just walk here. I think I will get in my EV and drive down, but I would love to see this before it's fully baked and such because I'm a huge dweeb for both airplanes and boats and getting places faster than Amtrak. So a pleasure. Billy, it's regentcrafted.com and any job you're looking to fill, you want to shout out while we have you. We actually have a ton on regentcrafted.com slash careers. We're growing right now. We're looking for, you know, technical support on building the prototype.

29:16We're going to start to build our manufacturing team very soon as that building's soon to be open. So definitely go to that website there and check out careers. And we have a bunch of cool videos too. Billy, you're the best. Thank you, man. We'll talk to you in late 26, early 27. Hopefully before then. We'll get you on the C. Goddard before then. You can be a brilliant founder with a great product and you can have an amazing team. But it means nothing if your sales process is disorganized and if customers are slipping through the cracks. Trust me. If your pipeline's a mess, you're going to spend way more time getting organized than talking to customers and closing deals.

29:49If this sounds like you, write down this name, PipeDrive. They're the number one CRM for small to mid-sized businesses like me. I use PipeDrive and have for almost a decade. It brings your entire sales process into one simple, elegant, and central space. That's why the sales team here at This Week in Startups uses it day in and day out. You just drag deals along your pipeline and you're good to go. Plus, they've got some amazing new AI integrations, generate sales reports with advanced metrics and personalized suggestions based on where your company is struggling, and automate your sales conversations with AI-generated prompts.

30:23And when you use my link, you're going to get a 30-day free trial. No credit card, no payment needed. Go to pipedrive.com slash twist to get started. If you listen to the show on a regular basis, you know we talk about AI probably too much. The one thing we talk about in the realm of AI is agents, little things that can go out there and do stuff for you. This usually comes up in the context of SaaS or enterprise software or something incredibly boring like expense reports or, I don't know, refactoring your spreadsheets. But there's a startup out there who wants to use AI agents to find old, unused, bio IP, test drugs out, and maybe bring them to market.

30:58It's a really interesting idea. I love the company. And I wanted to learn more about both how it works and how fast it's going to change our world. So please join me in welcoming to the show. It's Ayan Parikh from Convexia. Hey, how you doing? Hey, doing good. Thank you so much for having me, Alex. Looking forward to chatting with you. My pleasure. I just love to talk about AI in a context that isn't like trading or something involving enterprise SaaS. It's such a nice stage. So Convexia, as far as I understand it, essentially wants to use a series of AI agents to find unused IP or just leftover ideas, vet those ideas, run business queries about how popular or commercially viable those drugs could be, and then help bring them to market.

31:39Is that fair? Yeah, pretty much. So we're looking at all types of drugs. They could be really early stage drugs that are just sitting in labs. They could be shelved assets that are sitting in the shelves of these small, mid or big pharma companies, or they could just be things that have been overlooked. So we evaluate the science viability, the clinical viability, and the commercial viability to score which ones have the highest chance of succeeding and which ones can save the most number of lives. And why are those the most two important metrics? I can think like speed to market or number of people you could impact maybe on a non-lifesaving basis that would also be good vectors to judge against.

32:15Yeah. So I mean, when you consider like what will bring a drug to market and what will make it succeed there's generally or the way that we think of it is like what will cause a drug to fail essentially and the three reasons drugs will not come to market are because either it's not safe and effective in human bodies either there's too high of a clinical risk meaning like there's some manufacturing problem or just running a clinical trial is really really hard or it could be that there's no commercial viability that there will be that there's already something out there that's way better or the market size just isn't there or perhaps that like an insurance company will never recovered because the standard of care is already so good.

32:48So those are the three key reasons why drugs will fail. And those are the, that's why we look at those and consider why they'll either succeed or not succeed in the market. Okay. No, that, that helps a lot. Cause I was really curious about like, to me, there's a lot of ways you can approach this, but you have to narrow it down. The thing that I was really kind of hung up on prepping for our job today was just how much leftover IP or ideas on the shelf or I don't know, preclinical trials out there. Like how much stuff is there for you to go out and look at and run through the agentic stock that you're building?

33:19An incredible amount. So just to put some statistics out there, out of all the drugs that enter into phase one, only about five to 10 % ever even make it to market. And the percent of drugs that even make it to phase one is a really, really small proportion as well. So the point being that there's a whole bunch of preclinical drugs that are never making it to clinical trials. And that even when they are entering clinical trials, many of them are shelved for reasons beyond just the molecule not being good. For example, the clinical trial might not have been run the right way. Or a company might just have a change in strategic incentives or in strategic alignment.

34:01And now they just have this really cool drug that's just sitting on their shelves. So there's many, many examples I can list off as to when a potentially life-saving drug was just sitting pre-clinically in a lab for really, really long. or from where it's just sitting on a shelf even after entering clinical trials. There's a lot of examples like that. And I think as AI drug discovery continues to increase, the number of drugs being produced will simply continue to increase. Creating novel drug molecules will almost to some degree become commodified. But what will become really critical then is actually trying to evaluate which one of these will actually be successful.

34:37Out of all these drugs we're making, which ones of them will actually be successful in clinical trials? And I think that's kind of where the future of the space is going to. The future of pharma is an M &A. I can talk a bit more about why that is. But the larger point here being that we want to be on that forefront. Okay. So down the road, you know, five, 10 years from now, we're going to have AI models of sufficient quality to kind of come up with their own ideas. And then from there, we can vet them and then eventually get them to market if they work. Today, though, AI models, maybe not quite there yet, but there's tons of stuff out there that you can go out and mine.

35:08Okay. That makes sense to me. Glad that I understand that. The question is, how do you get access to the information? Because I presume that if I was Merck or some other large pharma company, if I shelved something, the public may not know about it. So how do you get the ability to have access to the information you need to do the work to vet as these drugs as possible candidates for convexity? So I think the last mile of this is always somewhat relationship-based. In order to actually strike that deal to some degree, you need to know the right people. However, in order to even start those conversations, without already being in that circle, there's still a lot of predictive work that you can do to understand what is shelved and what was being pursued at some point.

35:46For example, you can mine company reports. You can mine how much a company is manufacturing certain molecules. You can mine patents. There's a lot of information that you can look at to predict what is being pursued and what is not being pursued. And we've built our models to be able to do that and predict which drugs are up for market or are up to be sold or out licensed, which ones are just sitting there, and which ones are actually being pursued right now. And this is the first layer of your agentic stack, as I call it, which is your sourcing agent, the thing that goes out there and pokes around and finds stuff.

36:16and brings it back to the company. Is that part of the company's technology in production today? Is that going out there and doing this? Or is that something that you're working on building so far? We've done this. So we're not only using this in-house internally, but we've also sold this technology out to some customers as well. As we kind of talked about, it's not only does it mine both structured and unstructured data, but it's also kind of hypothesis generating and thesis aware, which is how we compare to a lot of these other, you know, typical competitive intelligence tools. So that's really the alpha that we generate compared to those.

36:47Essentially, you better understand not only biology, but also the pharma industry. And so with that information, you can just make much better guesses. Exactly. Then comes the next thing, according to your rundown, which is the scientific agent. And this uses a number of, as far as you can tell, just like AI bio models. I'm not as familiar with how fast those are advancing. Can you give me just, I don't know, a state of the industry, if you will, for that part of AI? I guess the gold standard for a lot of these in silico tools has for the past years been physics-based models. And that's been pretty true for a lot of the entire world of computational biology and computational chemistry.

37:24But now we're kind of seeing a bigger and bigger push towards ML-based models. Like just today, some cool models were released. Yesterday or a couple of days ago, Bolt's Gen was released from MIT. So a lot of really, really cutting-edge tools are being released in this ML space. And I think like when we look to the next 5, 10, 15, 20 years, I think these ML models will become increasingly better and become the gold standard. Whereas right now and in the past, it's been more so the physics based models. So I think like like in different worlds, there'll be different. I mean, in different kind of sectors of the science side of things, change will be seen differently.

38:00Like right now, if you look at like binding, the ML based models are actually pretty good. But if you look at like toxicity, there's not a lot of good ML based models or even models at all. So I think there's a lot of room for growth in this space, but we are getting there. So physics-based models for the work that we're describing to my non-biologist brain seem like the way to go. Why are ML-based models, in your view, going to supplant and surpass them? Well, I mean, I think the pace of ML innovation is faster than anything we've ever seen before. Yep, okay. So they're growing faster than ever before.

38:33And even physics-based models, they have a lot of limitations right now as well. A lot of physics-based models can actually predict everything that an ML-based model can even touch on. So I think the capability as far as compute goes, as far as speed goes, as far as raw capability goes, will soon be, I mean, the physics-based model will be far exceeded by the ML-based tools. That makes a lot of sense to me. Now, the next thing is your commercial agent. That one makes great sense to me. Then we get to the clinical agent side of things. This is when I get even more excited because you're talking about using digital twin simulations.

39:03And I think digital twin is a brilliant approach from, I mean, a lot of places, everything from, you know, factory design on down. But in practice, how good is digital twin simulation for individual drugs? Because to me, they could be a cool technology and maybe a tough use case. So I'm curious how well they wed, if you will. Running digital twins is a very, very difficult problem. And I think, like, we're still exploring what that looks like to its fullest capability. I think like kind of where we're starting is by not trying to boil the ocean and like, you know, try to create these digital twin simulations, but more so starting with like, okay, how can you understand the manufacturing?

39:37How can you understand the patient stratification? How can you understand even the trial design, right? These are all small components of clinical trials that might lead to a clinical trial failing, but aren't like the whole, you know, digital twin in itself. So I think like the way that you eventually build to like being able to run a digital twin is by starting off with smaller components. So that's what we're focusing a lot of our time on right now. So if we go from sourcing through science, through commerce, through the clinical agent, and we get all the way to a phase one trial, and you said earlier that what, 5 % of drugs make it through phase one, more or less, and you get out on the other side?

40:09No, so 5 % make it to market. So 5 % make it through phase one, two, and three. When you guys have this full pipeline in place, how much better do you think you'll be able to do on that 5 % from start of phase one through production? I think significantly better. We've been doing a lot of internal backtesting and we're going to be publishing some really cool results. We're actually publishing a paper pretty soon that'll talk about specifically some of these benchmarks. So I think that's coming up pretty soon and we'll be excited to kind of publish what those exact results look like soon. But I mean, based on what we've been seeing, there's a lot of promising results that we've been seeing by just doing simple backtesting of our agents.

40:46Are we talking about like 2 % better from 5 % to 7 % which would be 40 % quite a lot or are we talking like 5 % to 25 % like a 5x? Yeah, something like the latter. It'll be significantly better. And that's just because, like, not only are we now able to predict, like, just commercial success, like, that wasn't too difficult before, but, like, also being able to look at, like, the actual science of this, being able to understand, like, out of all the history of clinical trials, how much better is it to, you know, use this contract research organization or this manufacturing organization? These are all very, very small details that we can actually fully understand and understand how that affects the probability of success of a certain drug.

41:25And I think that's where we get really excited because of how granular we can get with some of these details. Well, I mean, that explains why you've decided as a company to sell bits of your technology to other people and you're seeing uptake from it. What does your customer base look like today? Are you selling to the former companies that I can name or are you selling your technology more right now to companies that are more in the startup realm? It's not big pharma. Most of our customers are small to mid cap pharma companies, and then quite a few like investment firms as well. So things like biotech VCs, hedge funds, we're starting to work with some investment banking, like biotech IV firms as well.

42:02So that's what the primary customer base looks like right now. Is the future of Convexia building technology that you sell to other companies that then use it to bring more drugs to market? Or are you eventually going to have the full stack put in place and you're going to run the process yourself? Because I can actually see a great argument for either direction for the company. I think the goal is to be, as the tagline says, an AI maximalist pharma company. That's kind of the long-term goal. The way that we think through our company is that in order to actually bring the drug to market, you need a lot of information, you need a lot of experience, and you need the right team.

42:33And the way that we eventually get to that long-term goal is by first working hand-by-hand with our current set of customers and partners to understand how they do it. And then we would leverage those insights to eventually run this process for ourselves. So the long-term goal is to be fully vertically integrated and to be a pharma company. How far away is that? Is that five years, 10 years, one year? I just don't have a good guess. Yeah. So when you look at like a lot of the other companies that have done something similar for them, like Schrodinger is a good example. Recursion is a good example.

43:01These are all companies that it took them like anywhere from like nine to maybe 12 years. I think our pace can be much, much faster. We're already helping a lot of companies make decisions about drugs to in-license. So I think like, I mean, we're keeping our eyes out if we spot something that really piques our interest and that makes sense for our portfolio. There's no reason why we can't pursue it if we have the confidence that our models are actually that good and that it's worked in the past. So I think it could be as near as a year from now. It could take a couple of years as well. It's pretty cool how quickly we can move in this space.

43:35And given the right circumstances, we can actually move on some of these drugs pretty quickly as well. Is your ability to move quickly predicated on simply the advancements in AML we were talking about earlier? Or is there something else going on at the regulatory level that's providing the tailwind just to your pace? I think it's a bit of both. I think we certainly understand that these models will continue to get better, and we want to be on that forefront. I think Sam often talks about this. As the model gets better, our company will also get better, so we're not really competing against models, but it's very synergetic.

44:01Yeah, sure. Yeah. I think there's also some interesting regulatory tailwinds, specifically with rare diseases that we're very, very excited by. We can get into the specifics of them, but CNPVs and PRVs, these are all vouchers that can essentially make it very lucrative or very like monetarily exciting to pursue a rare disease drug if you actually achieve some of these vouchers. I think this is like a really, really compelling space that, I mean, that's why we've been looking a lot into rare diseases, severely underserved market. And our model, like we're a really, really small lean team. So whereas like a Pfizer, they're looking for a$10 billion outcome.

44:38We can actually like an under a billion dollar outcome is incredible for our company, which is typically what the ultra rare or rare disease space might look like. Now, what is a good example of a rare or ultra rare disease that people listening might be familiar with? So I feel like we should ground that in something that can do an example. What's defined as an ultra rare is under 200 ,000 people get affected by it each year. So I mean, there's a lot of examples. K-Sachs is an ultra rare. And there are also some more commonly rare ones as well. But anything that kind of falls under that perspective is what would be defined as a rare disease?

45:09200 ,000 per year per around the world, not just here in the US. In the US. Oh, okay. So 200 ,000 in the US per year. That doesn't seem that ultra rare to me. That seems like my hometown times four. So that's what rare disease is. Ultra rare would be under 50 ,000. Still, that's my hometown. I grew up in a town of 50 ,000 people. That's a non-insubstant. I thought it was like five. I had this entirely wrong. Okay, but essentially, because convexia is so lean and because you don't run wet labs, you have lower operating costs, et cetera, you can afford to go after these drugs that don't match the cost structure of one of the pharma companies that we can make.

45:46Exactly. How many of these rare diseases are there out there that people need help with? Like, is it, I don't know, a couple thousand? I mean, how long will it take to chew through them? An incredible amount. So I guess for some perspective, out of all the rares and ultra rares that are known right now, only 5 % have cures. So 95 % of the market is still up for grabs. So there's an incredible like thousands upon thousands of ultra rares and rare diseases that are like that are that there's no solution for right now. And we've been spending a lot of time thinking about not only developing new cures, but also just repurposing existing drug composition.

46:21We've been working with a select number of partners to understand, can we build out repurposing models using some version of our agents to help them find how some approved drug can be repurposed for some rare, ultra-rare disease? So a bit like how GLP-1s are great for reducing your appetite, for example. They're also great for diabetes, I think. And they're also good at addiction control. They do a lot of stuff. So single compounds can have multivariate impacts. And so you want to see if existing compounds can help these super... How do you do that? What does the AI model or agent do to try to form fit existing compounds into known but unrelated problems?

47:07So there's a couple of ways. I think the kind of the standard way of doing it right now is just using like a graph neural network. I guess this is the standard way is in terms of like how we would develop it or how we have developed it. I'm essentially trying to understand like what mechanisms is a certain drug hitting and can those mechanisms, like whether it's through off target impacts or on target impacts, can any of those be reapplied to fix diseases or to fix some of these ultra rare diseases? So just like cross applying like, oh, this drug hits this mechanism. Can that be used to help solve this drug?

47:39And then there's a lot of other steps that kind of go into it. You have to look at the pharmacokinetic properties. You have to look at like the dozability, the toxicity at certain different levels. So there's a lot of kind of sequential steps that go into it. But at a high level, it's like it's a version of like making this puzzle based on all these different pieces of information that you have available to you. Okay, but if we have improving models, if we have agents that can do more and more, we're able to collect the data and do a lot of work outside of the lab to see what might be functional and we can go after lower commercially viable drugs.

48:08It sounds like if convexia ends up winning, we could tackle quite a lot of these rare and ultra rare diseases in the next five to 10 years, which would revolutionize quality of life for millions of people here in the US. Agreed. I think that's the goal. I think there's a couple of other players working in this space as well. I think the general vision of being able to find and bring drugs to market faster and use fine existing drugs and seeing is there a way to repurpose these for existing companies or for sorry, for existing rare diseases. I think it's a really, really exciting space. And I think given, as you said, that the models are getting so much better so quickly, I think there's certainly a path to seeing a lot of value come out of this space in the coming five, 10 years.

48:50One question I have, though, is just about capital, because right now, if you look at where venture dollars are flowing, I don't think it's biotech's best year ever. And so I'm kind of curious, are you guys classified as a biotech company that uses AI or as an AI company? Woo, easy fundraising that's working in bio. A mix of both, I think. Like, I think for our seed round, we just closed up our seed round. Hey, congrats. Thank you. So that was, I guess, more so the latter, like an AI company that's being used for bio. I think like as we transition over towards becoming a pharma company and actually like having an asset pipeline, I think we'll transition more towards like the kind of the bio side of things, the bio side of the spectrum.

49:27But I think right And now it's more so oriented as like being an AI company that is being applied for the bio slash pharma space. I love this. You're giving me actual hope for the future that when I get old and need help, there'll be much better things there. So thank you for that. Two things before I let you go. One, what is the place people can find the company online? And also in case you want to shout it out, is there a job you're looking to hire for and having a hard time finding the right candidate? They might be listening to the show today. Of course. Yeah, convexia.bio is our website. So we have a lot of information about the company there.

50:00And of course, all the socials are linked as well. We're looking to work with AI ML people. I mean, as is everyone that are very interested. Good luck. That are interested in the bio and health space. So if you're interested or working in that space, we would love to chat. We love making great conversations. And if you're interested in hiring or interested in getting hired, even better. All right. Well, thank you so much, my man. And when you have your next milestone, whatever that is, Come back and tell me about it because I have an eye on you guys. I think you're absolutely big. Thank you so much, Alex.

50:31I really appreciate it. I freaking love data companies. I love companies that store it. I loved Box. I loved Dropbox. I've known Databricks for so long. I think I actually grew up with that company, but that does not mean that we have solved the problem of information in the cloud and using it. And that brings me to the startup we're talking to today. They are called Archil, A-R-C-H-I-L. And they have a very interesting, may I say innovative way to ensure that you can access your cloud data very quickly as if it was local. How does that work? How is it going to make money? Let's find out. Please welcome to the show.

51:05It's Hunter Leith from Arkill. Hunter, hey, how you doing? Great. Thanks for having me, Alex. Dude, my absolute pleasure. So let's start with, did I get that right? At a very, very high level, Arkill wants to make it possible for me to interact with my cloud data, say on Amazon AWS, as if it was local to me. Therefore, it's faster to access and just easier to work with. Fair? That's right. Traditionally, you would spin up like a server in a place like Amazon or Google. And then you would have to transfer that cloud storage to your server in order to be able to do anything with it, which is slow and expensive.

51:39And so we've cut out that step completely and just made cloud storage appear local. So for folks out there who are not deep into the storage weeds, instead of having to drive to a place and put something in my car, drive my car home, and then walk back and forth to my car, it's essentially already in my living room. And you guys make that possible via some tech that we'll get to in a second. But fair? That's exactly right. Now we can go kind of one level deeper. You've created a caching system. And this is where I begin to have questions. Because I get caching, for sure. And I get the idea of having a cached copy of something that you can access more quickly.

52:12But where does the cache live? And you still need to get all the original data from, for example, S3. That's right. So we, in the general sense, like to put everything in the same. We offer like a SaaS kind of model where you sign up with us, you get access to our mega shared cache, and we only bill you for the amount of data that you're actually using. So you don't have to worry about any infrastructure to take advantage of the performance benefits and the cost savings. How does the caching element of this work? Because that seems to be kind of the magical thing you guys are bringing to existing S3 customers, for example, to make their data more accessible.

52:49So maybe give me the idiot version and then the actual version in sequence. We effectively, if you think about S3 or cloud storage in general as this low cost, slow storage, there is also high cost, high performance storage you can buy, which are traditionally like the SSDs that you might have in your laptop. We run a fleet of servers with those drives attached to them so that we can serve the data that's in the cache really, really fast. and then we have a bunch of magic on the backend to effectively predict what kinds of data you're going to need so that you're never waiting for us to go to the cloud to download that data for you.

53:29How effective is that intelligent pre-caching? Because my concern would be if I'm paying for this as an active gigabyte per month and we'll get to pricing in a minute, I would not like you to pre-cache more stuff than I need and therefore possibly incur either a higher bill on my side or unnecessary costs on your side. Yeah, I think that's right. And so what we aim to do is effectively make our service as simple as possible for anyone to use. This strategy is used by every company out there, Databricks, Snowflake, Netflix. Everyone is effectively trying to cache data out of places like S3 to make it more accessible.

54:09And so we want to make that available to people. But then there is this next level of configuration around what if you know what you want? What if you don't want us to pre-cache? What if you're cost sensitive that we can expose as knobs to our customers who care about that level of detail so that they get exactly the experience they want? So essentially, I might be able to tell our kill in the future that, hey, you know, don't pre-cache too much. Let me just kind of walk around and tell you what I need. And then you guys talk a lot about data being available effectively instantaneously. I presume that's after you've taken stuff out of S3 and then put it into the Alex Incorporated cache.

54:47It is, yes. Okay. But there's a different model of accessing the data in our system than before. Whereas if you were using, say, a 10 terabyte folder that was in S3 before, generally you would have to download this entire 10 terabyte file in order to use any of it, which backed everything up in your system. And what we've done is we make it possible for your applications to go ahead and use the data, even if it's not in the cache, and they will wait as they request individual pieces so they can get started earlier, even though we might be backfilling it asynchronously. I don't need to wait for 100 % of a folder to make it over into the Arkill cache.

55:34You guys will intelligently help me know what has been cached, and then I can interact with that part of the data set or part of the files in question. Okay, that makes good sense to me. Silly question, but as someone who hasn't really written code since high school, how hard was that to do? Oh, it's incredibly difficult. Because you're making it sound pretty easy. I have a question after this that I'm getting to, but I'm just curious. It's really interesting. So I built this company after spending eight years at Amazon working on a service called the Elastic File system, which does something similar in shape, which is offer this high performance storage in a fully elastic, expandable way.

56:13And it's really interesting if you look at the cloud companies and their offerings, because Amazon is the only company that has actually built a product in this space because it is so tremendously difficult to do. So it's easy to explain. Everyone understands caches, but it's extremely hard on the back end. That kind of answers my next question, which is why hasn't someone built this before? Because going and looking at your kind of marketing speak, you know, 90 % lower cost than using EBS and 30x lower latencies than using S3 directly. It sounds almost like a magic potion for the problems of moving data around online and interacting with it at the corporate level.

56:52Like it seems like magic. That's good. That's exactly how we want you to feel. Well, everything feels like magic to me because I mostly bang rocks together and rub sticks together. Why were you able to do this when Amazon and Google Cloud and Azure haven't been able to? Is it partially because you're no longer inside the beast, as it were? And so because you're outside, you don't have to follow, I don't know, normal Amazon development patterns? Or what was the unlock that made you, Hunter and our guild, able to do something that, you know,$10 trillion worth of market cap cannot? Well, I think if you look at the clouds as a market offering, you can effectively draw a through line from how people used compute and storage in the 1970s to actually the products that they offer in the cloud today.

57:37It's like servers, multi-tenancy, hard disks, object storage, done. And I think what the unlock was for our product is really understanding that the cloud has actually changed the way that we can offer these primitives to customers. And so it's very non-obvious that this product is like, A, worthwhile to build, and B, is possible to build. But you guys talk about how everyone has to kind of kludge together a similar caching system if they want to have a similar thing today. So it seems like the need for this is pretty obvious. And I'm curious if you guys have managed to convey what you're building to the market effectively.

58:22Like, are people reacting strongly to what RKL has cooked up? Oh, yes. I think what is, I feel very lucky to be a part of is there is tremendous demand for what we're building. And we get to collaborate with some of the largest companies on earth, as well as some of the most innovative startups, maybe working in AI that need this kind of technology and need it yesterday. So I think so. Obviously, we can always improve. And because we're such a technical product, simplifying and making it easier for people to understand is a really important part of how we think about marketing. I did complain a little bit internally.

58:59I'm like, why do I always fall in love with companies that are very complicated that I have to explain to an audience? It's just like, why can't you guys just do SaaS for bunnies? You know, that'd be so much easier. A question about where you guys actually store the cached data on the SSDs or whatever. Is that also a bit of cloud technology that you're renting from one of the major cloud companies? That's right. Yeah. So we actually rent servers from the cloud that have these SSDs attached to them. We then pool all of that storage together and then we rent that storage space out to our customers at a smaller granularity.

59:35But even being on top of the clouds, we're able to effectively turn a profit on a service like this. We'll get to that. But I was curious if Amazon hates you because I feel like you've kind of shown up to their party and been like, these people can't dance. Let me show you how. But it sounds like because you're helping in many cases, I presume, Amazon customers better access their store data on AWS. and you're also a customer of AWS, you're probably like their favorite person because you're doing the hard technical work for them and they get to have two customers instead of zero. I think that's exactly right.

1:00:09People will frequently ask me about what it is like competing with Amazon. And I have to kind of tell them that we're not in competition with Amazon. Amazon is in this great position where they win no matter what happens because they own the underlying infrastructure. And so if we're able to unlock AWS for more enterprises that couldn't otherwise use it, or if we're able to bring these startups that are doing the new things in AI to Amazon's infrastructure, that's a net win for them. Do you guys plan to have a similar setup with Google Cloud, Azure, OCI, and, you know, et cetera, et cetera, et cetera?

1:00:45There's a lot of cloud providers now that are doing quite well. It's not 10 years ago when AWS was kind of the only game in town. So one, do you guys support other clouds today? Or two, is that something that's coming next year? Yeah. And this was one of the shocking things that I found out after leaving AWS was that there were other clouds that other people used. But we are today in Google Cloud as well. We intend to be in all of the major hyperscalers as we're just following customer demand, as well as the GPU clouds that are actually becoming popular now. So your core waves, et cetera, et cetera, et cetera.

1:01:19That's right. That's interesting, though, because when I think about caching lots of information, I can see how a company that wanted to use its information for, let's say, RAG in an AI context would want to have access to that data. But how does it interact with a neocloud directly? They seem like different related, but different conversations to me. I presume I'm missing something. No, I think you're right. And so we actually like we segment the market a little bit internally, which is that there's all of these existing players, like maybe geospatial companies that are doing data analysis on imaging and bioinformatics and things like that, where they have applications that are in the cloud, but maybe need better performance or are just otherwise underserved by the storage offerings that exist today.

1:02:05And then there's all of this new stuff that's happening with AI that we're excited to be a part of. And that spans everything from pre-training, where people need to ingest a tremendous amount of data into the GPUs very quickly, of which there are only a few storage solutions that can support this at very high cost. and even all the way down into the kind of agent space that people are talking about today, where you need these agents to be able to crawl something like a file system in order to find out what data is available and use it effectively. Hence RAG. Okay, so it sounds like for training use cases, you want to be able to move a lot of data into a NeoCloud for training.

1:02:45So you'd want to have it cached for quick transfer. Fair enough. And if you're doing anything in the agentic realm, you want to have access to specific information, lower quantities, but very quickly. and that's where you guys fit in as well. Oh, interesting. So you're kind of like, you help facilitate both inference and training for AI. Ah, well, that's a huge market. I like to tell people that our goal over like the next three years is to become the universal access layer for data, which is that our product, because we offer this kind of rare combination of performance benefits and cost benefits, should be the easiest way for anyone building any kind of cloud application to actually interface with their data, no matter where it lives.

1:03:26So how long until I don't need to go to Amazon to start my storage journey and then layer Arkil on top? How long until you use white label S3 for me entirely? And I come to you guys and I'm just like, I don't really need to care about the cloud. I just want to have my data pre-cached and available around the world. I absolutely think that we will get there in time, but today for us, it's a matter of focus. And we are so focused on building this high performance layer on top of things that that's probably where we'll be in the short and medium term. That helps me with some timelines. Now, on the customer side, you know, I can absolutely see why companies that are deep into the AI weeds would like to use this.

1:04:06But I'm curious what other company sizes, industries, sectors, whatever the right question is, are taking up our kill for their own uses. Because storage is so broad, we see our customer use cases also be incredibly broad. We see tremendous pull from enterprises that are trying to spin up research environments for their ML and AI engineers who need to collaborate and share data very quickly and effectively. We see, like I mentioned earlier, kind of geospatial and bioinformatics companies, which traditionally don't have applications that connect to cloud storage natively. It needs some way to bridge that gap, which we make possible for them.

1:04:46And let's pause you there because this is actually, I think is pretty important. You guys use something called POSIX, P-O-S-I-X, to ensure consistency and also to make it possible to treat storage served via Rkill essentially as if it was local. So apps don't need to be cloud-nated to access it. They can treat it like it's on machine, which probably makes it more flexible. That's right. Right. It unlocks an entire segment of applications, which is to say that traditionally S3 has this API that applications need to be written against in order to take advantage of. And while a lot of applications over the past 10, 15 years have changed to actually use S3 as their backend, most applications don't.

1:05:29If you think about databases or like FFMPEG, kind of the video transcoding stuff, all of these things are things that you could imagine running on your laptop against files and folders that you could see in Finder. And so we take that cloud storage and turn it into this POSIX is the right word, compatible format that can run with any application that works on a machine today. It's freaking awesome. What stops Amazon from just buying you now? Because I feel like you're going to become bigger and bigger and more expensive, but I can't see any major cloud not wanting to have this offering as part of their quiver, if you will.

1:06:07I agree. And, you know, if that becomes relevant for us, it will certainly be interesting. But I think we're in a really good spot to execute on this quickly. And because of kind of the incentive structure that the clouds have around the products that they're selling today, I think it's unlikely that in the short term, anyone builds this themselves. So that's a very interesting statement. Unpack that for me. What are the incentives that are pushing back against them building this? You can be rude. You don't work there anymore. I have tremendous respect for my people who are working at AWS and Oracle and all of these places.

1:06:39Here's the boilerplate, kids, in case the startup fails. Andy, take me back. But because these clouds in the storage realm today generally sell object storage like S3, which is something that is billed per byte used, it is very low margin for the cloud. And then in addition to that product, all of the hyperscalers also sell effectively a hard disk kind of product where when you launch a server, you tell the cloud how much storage space you need on that server and they provision something that looks like a hard disk. And you pay, in this instance, for the amount of space that you asked for, not the amount of space you've used.

1:07:21This is one of the core things you guys put out as a developer sticking point, as something you're trying to solve by making it essentially elastic. That's exactly right. And so if all of these clouds are making billions of dollars on effectively the unused space that's on those disks, it's very unlikely that they would want to spend a tremendous amount of R &D to get rid of that extremely high margin income stream. That makes a lot of sense, especially right now when technology companies are at once their most profitable and their poorest because they can't afford anything. Let's talk about costs.

1:07:54So for folks out there who have not looked up the S3 price list recently, if you're curious, for S3 Express One Zone, it's about 11 cents per gigabyte. You guys are currently charging about 26 per quote active gigabyte per month. It does seem like you're not charging that much because if you have to pay for S3 yourself to a degree or compute, there's only so much margin there, Hunter. So tell me about the economics of this and how it works. I think that's right. And this is another reason why a lot of companies are not super interested in building storage solutions is because it's not your high margin SaaS play where costs and revenue are completely unrelated kind of a thing.

1:08:39We are selling much like the AI companies, effectively something that has costs of goods sold that we have to deliver actual margin on top of. Cogs plus margin is how it goes. So that's right. But when we talk to engineers who are at these enterprises that are interested in using our product, they're very indexed on wanting to use cloud storage like S3 because for one of many reasons, it's very low cost. And so they want in the default case to only pay the costs that they would have for storing that data in S3 if nothing else is happening on the system. And that would make it a good design. So when we think about charging, our pricing is directly tied to what our customers are doing.

1:09:29If they're actively using that data and we're caching it and we're facilitating access for them, we're charging them for it. If they stop using their data, it's no longer in our cache. We charge them nothing. And the data is just on S3 for them to use as if they only used S3 in the beginning. So there is enough margin there for you. And it's positive for customers because they're just not paying for unused capacity, unlike the current object-oriented storage system set up by the major clouds. So you're more developer-friendly and you're speedier. That's right. Pretty good combination of things.

1:10:01It does seem though that companies over time will sometimes try to take a bit of margin back, a bit of a little customer surplus. How are you guys going to avoid becoming this sort of company that puts ads in consumers' video subscriptions a la Amazon? It's funny. It's kind of hard to put ads on someone's file system, though we have thought about what that might look like. Don't do that. Please, God. I think that for us, the story is the same as it is for something like any AWS or Google Cloud or things like that, which is that storage is a place where we start. And then we want to provide services on top of that that might be more valuable to our customers.

1:10:42So rather than offsetting these prices with things like ads, we would rather offer things like compute on top of the storage that provide a better experience because it's tightly integrated with what we do to generate additional revenue. So if cloud computing and storage is abstracted compute for companies, it sounds like what you want to build is like a synthetic layer on top of that that takes kind of the raw grunt of data centers that other companies have built and makes them much more modern, if that makes sense, and offer them to people in a manner that's much easier and better to consume and interact with.

1:11:20I think that's right, which is that over time, the way that we see people interact with systems continues to become higher and higher level, which is like maybe 15 years ago, developers really wanted to use S3 directly, use disks directly, put all these blocks together themselves. But, you know, now people expect higher level primitives. They want to just use Vercel and send us some JavaScript and have that work. And so I think that you'll see the same arc on this backend data intensive work, like even Databricks, where people just want to ship a problem to us and have us solve it for them in whatever way makes the most sense.

1:12:01Well, being pretty darn asset light because you're not making your own data centers. You're not buying land. You're not trying to get power generation sorted out. You kind of get to lean on what has been built and take it a step further. I really like that. Last question for me is you guys raised some capital. I think it was in June or July of this year. I think Felicis led that. Who was your lead investor? So Stasia at Felicis. I love Stasia. She's fantastic. One of the smartest VCs that I've ever met, actually. She was at Redpoint before, right? That's correct. So how much capital do you guys need, though, to take us as far as you want to in the next couple of years?

1:12:35I ask because I've talked to a lot of companies that are raising amounts of money that don't seem to track with their needs. And so I'm just kind of curious, what will it take in dollar terms to get you to the next, I don't know, three or four years? I think it's hard to say. And I think like there is obviously a spectrum of companies that you see today from all the way on the left, like light capital requirements, your traditional SaaS, where there's not really any costs outside of R &D to all the way on the right, where you're building a foundational model and you need to, you know, go buy a billion dollars of NVIDIA chips.

1:13:07We're neither of those things. We're somewhere in the middle, which is because we're offering storage as a service and storage is something that, um, both the capacity and the performance and the safety actually scale with the number of servers you have. We have a fixed cost to actually appear in any region, which is why we're only in a couple of regions in AWS and GC. I see you have to, okay, right. You have to provision or kind of take, take onto yourself a certain amount of, let's call it flash storage or whatever. And so if you replicate that in every single AWS instance, that's very expensive.

1:13:44Or region, I mean, sorry. Yes. And so we do have capital requirements. They are low compared to obviously model training or data centers or energy. But we hope to raise more capital over time to be in more places and then fund more R &D into the higher level services that we chatted about. I'm even more excited about the company now. Where can people find you on the internet? And what is a role you are hiring for? You would like to shout out to the audience in case there's the right person listening. So you can find us at archil.com, A-R-C-H-I-L.com. If you've heard this and decide you already need the product, we also have disk.new, which you can go to and just immediately get logged in and set up with something that gives you high-performance storage.

1:14:27We're also on Twitter and LinkedIn, of course. As far as open roles go, everything for us is engineering. as may or may not be a surprise to you. I'm blown away by that revelation. No marketing people? You're not hiring on the GTM side yet? Not yet. I think that if you look at companies like Modal, for example, which are this kind of high-performance compute platform, one of the things that they did well early on is just focus exclusively on engineering. And if you get the best engineers, you can grow the product and then fill GTM in later. And I hope we take the same playbook. I think that's the right way to go in your case because you have to sell to developers who are famously relaxed, casual, and not picky whatsoever.

1:15:07That's right. Hunter, an absolute pleasure. Thank you for teaching me so much. Good luck. And I'd love to have you back in six or nine months. Thanks so much, Alex.