Austin Barack: My AI Bull Thesis (...And What I'm Holding)

Transcript

All righty, guys. Welcome back to AI Supercycle episode number 34. Our weekly premier AI show powered by our friends at Near. Visit near.ai to learn more about their product suite. Today, we've got Austin Barack on his Venice Grass Near and AI thesis. Austin is the founder and managing partner of Relayer Capital. Guys been absolutely dialed in on the timeline as of late. Good to have you Austin. >> Yeah, really glad to to be here. A lot a lot of interesting things happening in in cryptoland these days. >> There is, man. There is um man, let's just start, man. I I used Venice for the first time a little while ago. We had Eric on the show in February at Nearcon and um I typed in like, "Hey, is this really private?" And it was like, "Yes,

the inference is private, blah blah blah." And then I wrote back, I was like, "That's [ __ ] based." And then it responded back and it was like, "Yes, it's pretty cool, isn't it? We with Venice, you have this." And so like there's kind of like this really cool first moment when you use Venice and you're like, "Oh wow, all of my chats that I've been using um with the you know typical providers are such are so not public but the data is going to the large providers and all the deepest darkest diary rights or freaking you know whatever it is business secrets plans etc. And so give us your private AI Venice kind of high level view here, market structure, thesis, founding team,

and let's start here. >> Yeah. So, Venice is really interesting because they've gone through so many different iterations over time and I first came across the project January of last year because I I was really big into virtuals and AIXBT and a lot of the initial airdrop went to holders of of tokens in those ecosystem. So, that that's when I first came across it. It was really, you know, interesting product. Um, but AI and and it's wild that it's only 16 months or so, but AI wasn't as ubiquitous then in terms of just in, you know, a necessary part of everyone's daily life. And over that period of time, we've gone to the point where

whether it's claude or chat GPT or or another service, it it's not only it started off as supplanting Google search like, oh, I'm going to just search whatever query I want on, you know, an AI platform and using an LLM instead of Google, but now it's around creation. It's around solving tasks. It's, you know, it's having a team of people around you and using agents. And what I think people are realizing over time is that in AI more so than anything, privacy is incredibly important because you're sharing your health data, you're sharing your financial data, you're connecting all of your files, you're you're sharing your entire life in a way that you haven't before. Because in the past, people are like, "Oh, how do I think

about privacy in the context of social media? Do I have a public or private profile? Does Facebook have too much information on me?" And this is is not like 10 times more data using AI. It's a hundred times more data. And what's really really cool about Venice is it's not about just having AI usage in a private context. It's about not sacrificing user experience at all and actually improving the user experience because instead of being wedded to one model where it's like oh I'm using chat GBT and you know I'll use their models as they upgrade or I'm using anthropic and you know the different models as they improve over time or Gemini or otherwise or the different open source

models. You can use the right model for every task and you can select which models you want so you can be as custom as you want. So what they've built is first of all a consumer product that's just really really really good and most of their users don't know anything about the token and then the token has this added element on top that's really cool but I think that's the and I'll stop talking in a second but you can tell I'm very bullish on on what they're doing. It's you have this situation where and this is where I think consumer products are headed in crypto where the token can be an a really important part and like supercharge the experience but for the majority of users they don't need to know about the token to find it useful.

>> Yeah. I mean that's like the breakthrough of consumer product crypto underneath but it adds this kind of like interesting element. people are actually comparing it to like Luna um and this like staking to get this DM token and then you have like this debt with with your uh inference that you get you know um that you get to buy into by staking the token and people are like people are having issues with the token dynamic right now. I'm curious, Austin, what is the right way to think about the Venice VDV token and DM token as they exist in the current Venice flywheel and then kind of

take us through some of like the revenue side as well of Venice because they are doing some buybacks but not a ton. How do how do the how do these two tokens work and um why is it not like a Luna? Yeah. So, they have three million users. They just announced that and and they're growing incredibly quickly. So, they got a million users in the last um three months or so, a million seven months before that. So, they're constantly accelerating. They have two tokens. So, they have the VVV token which protocol revenue is used to burn VVV, but also you can also stake VVV to get a free membership. But the most interesting part is you can stake and lock VVV and

that allows you to mint a token called deem. You can also buy deem in the open market but you can stake VVV and and mint dem. And what deem is is for every deem token that you own, you get $1 per day of free inference compute on Venice's platform. So you can think about it like a perpetuity where you know over the course of a year you would get $365 of compute but it's use it or lose it. So it doesn't acrue over time. If you use 50 cents in one day the next day you restart with a dollar and the next day you restart with the dollar. Um, and it's created this really interesting almost like loss leader customer acquisition tool is the way that I think

about it because unlike Luna where you got to this insane amount of um tokens outstanding and like billions and billions of dollars of stable coins, they are very very explicit about boxing in what the potential cost of this is. So right now the the the amount of deem that you can mint per Venice token declines as more is in circulation and creates more or less a hard cap around 38,000 deem. So at the current point if all deem was locked and used for inference because if it's freely circulating if it's in an LP pool I if it's you know not in that system then you're not earning the dollar per day.

You just have that asset that could earn the dollar per day. But if all of it was being utilized, that's $38,000 per day maximum of cost to Venice. And that's directionally $10 million a year. And it can't be more than that. Right now about 10,000 deem is being used per day for inference. So that's about like a three.5 million annual cost. And that's offset by the fact that they have uh probe subscriptions, premium subscriptions that range from $18 a month to $68 a month to more. And then as people use the platform, they purchase tokens and and you know purchase additional credits to use the models. Um they've grown from a couple billion daily token usage to you know as

of recent count 70 billion. So done like a 50 15x in the last couple months. So I would say the the way it's different here is there's a maximum potential cost to the company but by people using DM they're also using uh subscriptions they also purchasing other credits if they need more than the dollar per day per token um so it's very easily covered by the revenues of the business um and well exceeded by the revenues of the business but what what is really cool about it on the other side like you know how is it potentially lunar and and I think it's it's clearly not. But the really cool part is you can guarantee access to compute in the future. And right now

it's valued at like a 20% discount rate at around $1,800. I think something like this should be rout valued more or less along the lines of like corporate bond. So say like 8 to 12% which would put its price a lot higher. just, you know, as an example, 10% would put it at um $ 36.50. Uh when I started uh looking at it, it was valued in the $200 range. >> $200. >> Yeah. And I was thinking about it. >> Yeah. And I'm like, there's no way that something that's producing $365 a year should be worth >> $200 unless people think that Venice is just not going to be able to support it.

U so at that point, it was a no-brainer. I I still I still think there's room to run. Um, but on the economic side of Venice, outside of DEM, the the numbers are staggering and and unlike anything that that we're even seeing in crypto, it's kind of like growth rates that you're only seeing in AI, which is why it's so compelling. And >> and so you're pretty a firm believer that I I think then it's tapped almost $20 today, the actual VV asset. >> You think this one a half to2 billion dollar range is firmly undervalued for VVV?

>> Yeah. So when I first started buying in January, it was like, you know, about two and a half bucks. And >> that's when we had Eric on Feb. >> Yeah. Yeah. It was, you know, uh Eric has consistently been ahead of the curve in everything that that he builds. And, you know, I've been following him for for a long time. So that made this investment a lot easier because, um, he's he's proven himself time and time again. Um, but at that point, they were doing a couple billion of tokens a day. Now they're doing 15x that. They had about, you know, maybe like a million and a half um total users. Now it's 2x that.

They are by my estimates probably doing 3x or more the amount of revenue that they were doing before. and likely trading in the like 20 to 30x um revenue range for a company that's growing I don't know like 20% month over month uh- which I think is still incredibly cheap and you compare it to an open router which is worth the same valuation doing similar amounts of revenue maybe less um growing perhaps not even as quickly uh and the difference here is they own the customer it's not backend and infrastructure that's plugging in.

People that use Venice are using Venice every day. I mean, that's the only way that I use AI is is through Venice. Um, and yeah, I I I think there there's a lot of room to run here. >> Yeah. >> And of course, none of this is financial advice. Just my personal opinion. >> So, I haven't been up to speed on grass at all. You've been harping on this one for a while. It looks also ready to fly. I'm sure it's pulled back a bit today. Whatever. $50 million or more of annualized revenue accelerating triple digit growth. What is the premise of grass? Why, like how does it make money and why is it compelling?

>> Yeah. So for grass, they collect data sets and sell those to Frontier AI labs uh that need that data to train new models. And these labs are generating new models at an incredibly rapid pace. But in order to generate those new models, they need more data. And it's not just like random scraping of the internet. It needs to be very specialized. It's very specific data sets and it needs to be high quality. So that's the the role that grass is filling. And because there's such an incredible amount of spend around building these models, grass is a beneficiary of that because then there's just a larger and

larger demand for data. So the team there h has been you know building for years. They in you know I think they did like $3 million in a re revenue in a quarter at one point last year. By the end of the year they did 12 million or almost 13 million in the quarter. um based on my estimates, they're they're growing even faster and they're going to have their token holder call in the next like month, month and a half. Um and you know, we'll we'll get more information, but this is something that is showing tripledigit growth. It's 50 million AR or annualized revenue based on the last numbers. I expect they're probably closer to 80 right now and it's valued around 400 million. So, for something

that's growing so quickly to be valued at 5x just makes no sense in my mind and and is um a really good candidate to rerate. >> And does Grass have a any working relationship or anything to do with Venice at all? >> Uh they they they don't at the moment. Um Venice is is generally not building their their own models. Um, right. >> So, yeah, there there's no relationship at the moment. You know, who knows down the line. Um, but I look at it as like two interesting >> different sides of the equation where it's one is how do you use AI and how do you use it in a private way and the

other is how do you build the models in the first place and Grass and Venice are tackling both of those sides. And so grass at 5x of revenue is I mean some of these things in crypto trade 20 30 40 50x revenue. So you think like the $400 million range is is uh bit of a no-brainer. >> Yeah. And and I think a big part of it is like there's other things in crypto that that are trading at like relatively low multiples but they're not growing.

And people come to crypto because they want to invest in growth. So, I think a lot of those examples don't necessarily hold water because there there's just not the flows. But for something like grass that's one of the best examples of something growing so quickly, I I think it it it merits attention just on that uh basis, let alone the fact that that it's, you know, in my opinion, fairly cheap. >> Yeah. Okay. Um do you have a thesis on near at all? Have you been paying attention to Near? >> Sorry, say that again. You Do you have a thesis on near? Have you been paying attention to Near?

>> Yeah, I've been following Near. I I think Near is an interesting one even outside of the the AI component because they are the infrastructure that's powering so much of crosschain um swaps and we saw that you know in October and November last year gain a lot of attention with um swaps for for Zcash right as people were going in and out of Zcash and Near intents while useful and probably the one of the best crosschain experiences for swapping it also plays a really powerful role role for agents because it it's in my opinion one of the best uh pieces of infrastructure to plug into to do crosschain swaps where you don't have to you have all these other

dependencies. Uh and they're seeing that grow really quickly. I I think right now it's like if you're an L1 you somebody tweeted this out earlier today. Um, so I'm like I thought it was so well uh said, but I'm I'm repackaging it and I I don't want to like not attribute credit. Uh, but it's essentially like you need to be a vertically integrated app experience. You need to be 10 times better at something or you need to um you need to be like really really really good at one type of application. And I think Near is just absolutely crushing it on the intent side and they're doing a lot of other things on like confidential intents, other elements

around like AI usage, but they're one of the L1s that's really found their place. It's like an NBA player and all these new L1s and L2s are coming to to market and they're the really interesting rookies and you know over time some become the superstars and some just kind of like fade away. But then you have the the role player that's just really really good in their role like I don't know like Lou Dort on OKC or Alex Caruso or whatever. Um and Near feels like one of those players where you know it's not LeBron James but they're really important because they're just so good at what they do.

>> Yeah. And it's also like sitting in the middle of like this privacy AI thesis which I think is like really obviously gaining traction. Another one that's been slept on that Robbie has been harpening to me about, unfortunately, he's not here today, is um Akos and how they were so early to this kind of like distributed inference, distributed models, this kind of like decentralized training, right? This whole that was like this that was at the first iteration of crypto. Then we had like these like fugazi agents with like the memecoins and now we're on to like the next iteration of the decentralized inference models training, but like with like far superior products. Have you looked into AOS and what they're doing?

Do you have a thesis there? >> Yeah, I've definitely looked into AOS. They started off with with um you know decentralized CPU marketplace and then GPU marketplace but you can actually check on the amount of data that's running through open router. A lot of that is going through AOS which is uh the Akos ML which is very cool to see and that's just you know public data. Um, but yeah, I I admittedly am not like it's not one of the names that I'm following as closely, but it's for a team that's been around for a really long time, it's constantly iterating.

It's very cool to see them find uh real product market fit right now. Uh that that seems to be accelerating. >> Yeah. And I'm just going down the just going down the list here. Um there was a company called GitLob. I don't know if you've seen them. They're like a very low cap on base. Um, and they're crushing the amount of tokens in production per day as well. Um, and there's this kind of new cohort of like highly speculative AI tokens on base right now. And there's like so many little niches to understand in the puzzle here. And so I guess like not to ask about kind of like low caps or you know what you're seeing there, but more

like broadly is there a specific part of this stack that you think is most ripe for massive growth when it when you can apply like the blockchain rails to it? Clearly we've seen private inference and uncensorable chatbt with Venice with near it's kind of like this agent marketplace with Akos it's this AOS ML grass the data sets what is the is there a couple like key sectors or parts of the AI stack that you think you know investors should look at for being meaningfully replaced or usable when it comes to like applying them to the blockchain context.

>> Yeah. So I I think you have the privacy content uh context around private usage of LLMs as well as uncensored usage. Uh you have the data collection side to trade models which is grass. Um you have inference and compute um which you know you mentioned AOS we're seeing other inference marketplaces pop up. There's one that's actually building around deem as as well as you know other services that that allow you to um sell your spare compute and that's called an seed. That's been an interesting one that that I've been following. There's um there there's no live token yet or anything.

Um but but that's one that that I think has been doing some some cool things um particularly um aligned with with Venice and Deem. Um, I think there's decentralized model training as well, uh, in terms of how can you build open-source models that also have the ability to retain ownership and monetization by using private weights. And there there's a couple that are building in that sector. Plurales is is one of the ones that that I think is most interesting um that that's building there. There's news research that that's doing some really um interesting things with with Hermes. Um and you have um prime intellect a couple others. Uh so I

think like the decentralized training, the inference and compute marketplaces, the infrastructure around agents, the data, and then just like the consumer application of model use is is the main areas that that I'm looking at. And then, you know, Austin, I think like there's this AI thesis and, you know, covered a lot of like the big names here, but there's also this other thesis that you've been, you know, just pounding on, which is there's this new way to think about token models and economics. You've been a big proponent of, you know, like Aerome or Hyperlquid.

I'm just curious how do you think more broadly outside of the AI context as we kind of come to a close about this net token value flows and thinking about like credit and debits and how how like the industry is evolving in their thought process specifically around analyzing token economics and the right mental model for assessing the plus and minus or the credit and debit of a given crypto asset. What is your current framework for this? And like do you have a like do you agree that investors should be thinking about this like a credit and a debit like a plus and minus sheet? How do you kind of think about uh net token value flows?

>> Yeah. So, I I think there's a couple ways and and one it's not one sizefits-all because let's start, you know, at a high level with uh buy and burn. And that's something that Hyperlid has made um really popular and everyone's like, well, look how good Hyperlid is doing and they have buy and burn. But there for every one Hyperlid, there's nine other tokens that have the same buy and burn and their token price has been horrible. And what's the lesson there? It's hyperlquid is a great business, hence people like the token and buy and burn is how they deliver value to token holders. But if it's not a great business that's growing really well, your token's not going to just go up

because you do buy and burn. So like that that's the first thing that I think people conflate. The second is are you driving value to token holders? So whether you're doing buy and burn, buy and distribute, um whether you're do, you know, even just reinvesting in the business, having money in a bank account to give yourself more bank balance sheet flexibility, do the token holders maximally acrue the value of what's being built? And that's the case with Hyperlid, that's the case with a drum.

you know, for all the, you know, wish that a lot of people have that grass did more buybacks, it's very clear all the tokens, uh, sorry, all of the contracts are signed with the foundation. All of the revenue goes to foundation bank accounts. All of that is controlled by token holders. So, I think there's a lot of different ways of thinking about it. And then it's a matter of token flows. So, you know, Hyperliquid has a theoretical maximum of how much could be unlocked every month, but really it's like two or 300,000 tokens and the amount that's being purchased from the ETFs, from the DATs, from the assistance fund is much much higher than that. So, naturally, you have more buyers than

sellers. Something like Aer Drrome, if you lock Arrow into V Arrow, soon to be called S Arrow when they do their ETH mainet expansion in July, you're earning all of the revenue from the platform and you're directing emissions to the relevant pools where you where you think it's most important for there to be liquidity and where you're going to generate the highest revenue. And they're like, well, if in a given period the value of emissions exceeds the value of revenue, then you know it's net negative or it's net earnings. Sorry, net negative earnings. And I think that's really the wrong way to look at it. The way to think about it is how much is being gener earned as revenue

versus how much is freely circulating afterwards and not being sold because you have the you know the um they they recently changed the name to the momentum fund but you have kind of essentially the foundation that's doing buybacks all the time. You have people that are earning arrow that are locking it every period to stake it to earn the revenue. And then you also have people that are just bullish on the token or not selling. So when you think about it on that basis, every single epoch, so every single week, there's meaningfully less tokens that are going to the open market than the amount of revenue that's being generated. And then when you

factor in recent launches like you know there was the atlas aura and others at this point aeradrome is earning meaningfully more revenue and when I say air I mean token holders than is actually coming out as emissions. So I think for every example you have to look at it in a nuanced way but at the core is do token holders earn the value of what's being generated and then you can start to iterate from that lens. I think the entire industry is shifting into a similar mental model. As it's very nuanced um but it's like we have two emerging like things happening right now. It's like these like revenue fundamental companies and then these like more narrative niche very like useful technologies

Zcash uh Venice Near these kind of like very niche AI privacy things and then just pure businesses on chain and like in the middle right now there's just not a lot going on. >> Yeah. No, I think that's right. And one of the interesting things about this market is there's a smaller set of interesting tokens because people are a little bit more realistic around like what's finding traction, what's like real in the moment and not just like hype and not when I say hype, I don't mean hyper, but like I mean like like what what's like euphoric uh dream like excitement and you know there's maybe like 10, 15, 20 tokens that have really really strong

fundamentals. So I think for that reason we're seeing those tokens meaningfully outperform the market because for the first time in a while you are able to pick from a much smaller subset. So all of the flows are going to the things like Venice, like Hype, like Grass, like Arrow, like Near, uh like Zcash, which is another one on the privacy side. you know, people are, I think, a little concerned that Bitcoin is becoming too impacted by Sailor and that that's like a whole another conversation. Uh, and that Zcash is the original ethos of of Bitcoin with the similar structure. And Zcash, while it's, you know, doesn't have revenue in that context, it's um it's an interesting asset because it

becomes more useful the higher the price goes because then it has the higher higher likelihood of being entrenched in having uh consensus community value around it. Um so yeah, we're we're in this really interesting place where it it's getting a lot easier to bid the right coins. You just need to like you know, be be a little bit more um you know, like channel your your research into what's real and what's not. And you know, for people looking for the 5 to 10x or the 3x, you're much more likely to get it than trying to get the 100x. And maybe you get the 100x

anyways. Um but but I think like there's a there's a set of businesses that that are doing really interesting things and those are the ones that I'm anchoring to. Yep. No, I mean it's like as it's easier if you are disciplined and just stay in the lens of like what's working and what's going up. And I mean today's been a bit weird, but you know, you zoom out here. A lot of these things are I think they're kind of like buy hold and forget. Trust that you're in quality. Trust that you're going to do well. Austin, have a great Memorial Day weekend, man. Thank you for joining us in this episode of AI Supercycle. Been loving your content. You've been absolutely spot on last six months. So, keep it up, man.

>> Thanks, man. Glad to be here. Talk soon. >> All right, buddy.