NPC Labs CEO: Biggest Shift In AI Is Happening In Front Of Your Eyes (Everything You Need To Know)

Transcript

I guess in a similar way we offer something like that but for compute right and and if so you want to bet on hardware as an asset class and not just Airbnbs as an example you can come to us right put some capital down we'll procure that hardware for you host it and help you monetize it essentially by selling that capacity out to you know everyone out there who wants it at the moment >> just like the streets of Manhattan I'm going to keep this quick over 70% of our viewers are not [music] subscribers yes you. Every week we interview some of the best entrepreneurs and investors [music] in the financial industry. The easiest way to support our channel is to hit the like and smash the subscribe button.

Now, sit back, relax, and enjoy [music] today's show. We're going to be chatting about AI uh compute ownership. Uh and this is our a AI super cycle segment powered by Near. Uh they've got great AI products as well. And so we're going to get into it. Darl, thank you so much for being here. Appreciate you coming on. >> Yeah, absolutely. Thank you so much for having me, Rob. >> Absolutely. Well, uh I realize that you know you're co-founder of NPC Labs, big contributor and builder over at B3 as well. Maybe you could just talk us through NPC Labs B3. What's the relationship and what is your role there uh as you contribute uh to these entities?

>> Yeah, absolutely. I'm going to speedrun this because I don't think anyone is too interested in my background and who I am. think they want to have, you know, more insights on AI and a meaningful conversation and insights from you, Rob. And so, you guys are the best at that at the rollup. So, I'm just going to speed past our background, but in short, I'm one of the co-founders, CEO of MPC Labs. We're a core contributor to the B3 ecosystem. And today we are super excited to talk about B3i IQ which is our newest product that is our take on the compute space and essentially giving anyone the opportunity to own their own AI hardware NVIDIA chips specifically because those are quite hard to get

these days. >> Yeah. And you know, I I saw Jensen give this there was a panel discussion he did on CNBC and I think he followed it up with a Twitter article just today or yesterday talking about this partnership that he now has between Nvidia and some of the largest capital allocators. I saw Goldman Sachs, Black Rockck, Blackstone, KKR, Apollo, you know, were all up there talking about how they will be contributing finance uh financing to this AI data center buildout. Um, and it feels like what you're describing at B3 and this B3 IQ platform is going to fractionalize and democratize access to this asset class. Treating AI compute as an asset class and giving access to this

asset class to the broad swath of investors out there. Is that the right idea? Is that the right way to think about B3 IQ? >> 100% spot on. You nailed it better than I could honestly. But just to give you it in our own words, I think this idea of owning your own intelligence, owning your hardware is just starting to take off and gain more mind share recently. And we are a new platform that enables teams, startups, researchers at leading institutions that are doing amazing research around cancer genome sequencing, embedding LLMs into your phone. But I think part of the problem and the

challenge they have is like very very limited budget and so renting doesn't really make sense. And so our platform gives anyone from larger teams to individuals this ability to purchase, manage and also monetize AI hardware should they need to or when they're not using the machine. And I think it ties back into what you said there Rob around fractionalized ownership and democratizing access. We talked to a lot of startups in the space and the demand for compute right now is extreme on the supply side. There's a huge bottleneck especially for the best and latest generation pieces of hardware like your Blackwells, B200s, B300s and whatnot.

And the demand is so extreme right now that if you go out and rent one of these things, whether it's individual units or a node or a cluster or so on and so forth, you're actually paying premium rates. And so, just to give you a very simple example, if you're renting, let's say, a node of H200s, which is something we offer, you are paying probably 200k a year just in rental costs. And so, at the end of two years, you're paying 400k. But if you actually go and buy the H200 node outright, like you're buying a car or a piece of property, right, and you owning the whole thing, you actually pay 400k upfront. So really, the own versus rent math is super skewed at this point. And ownership makes a lot of

sense because you're just paying AWS 400K and at the end of two years, you're not you're left with nothing, right? You don't even have the piece of equipment. >> Yeah, I I think everyone wants exposure to this asset class. I mean typically when you look at the largest capital allocators it's not a bad idea to copy trade them and so everyone sees that there's an insatiable demand for tokens right and obviously compute is the most necessary raw material in producing tokens and so everyone wants exposure to the asset class of compute because it is just in such high demand and you're right typically the way to get compute has been to rent but now you're offering the opportunity for firms to buy compute hardware and infrastructure. And so give

us an idea of what happens if I were to, you know, if I was a user and I came to the platform B3 IQ and I wanted to go ahead and buy so that I could own, you know, some more of this stack. Do you guys own Blackwells, you know, in a in a vault or a data center somewhere? How how does this work if someone comes to your platform with, you know, some dollars ready to spend on compute? >> Yeah, absolutely. We try to make the process as easy as possible. Whether you're an AI researcher or you're a individual that's just looking to have a local inference setup, right? And you want to try some of the latest and greatest open models. You come to the site, you essentially have a self-s served browsing experience initially

where you can check out all the different types of setups that we offer from H100s, H200s to your Blackwells. we have access to the suppliers and distributors to procure this inventory. And so it really depends on your workload. I think for some of the AI startups that just raise a lot in capital and want to split between renting and owning, they'll go for the higherend pieces of hardware. Let's say your Blackwells as an example. So you come to the site, you select the equipment that you want and then you book a call with our team. I think this is a pretty you know high budget offering and also it's pretty technical in the sense that we need to really deep

dive and understand what your workload is, what your setup is, lead times and build a custom configuration here that works for your training needs. A lot of folks are training using our platform now and after we do the consultation, we purchase and put in the order should things work out and then we act as a one-stop shop there. So, uh, we go out, procure the hardware, we host it at our 27,000 foot Oregon facility that has all the power, all the cooling, all the requirements that you need, and you access the hardware once it's live through either a dashboard if you're less techsavvy or if you just need to connect to it directly through some orchestration or your in-house platform

and piece of software, then you can do that as well. We offer basically physical single tenant bare metal machines that you can connect via SSH or you can do some other you know access method whatever works best for you. So you can use these machines, connect to them via SSH, use them for compute, use them to produce tokens. And it what I'm getting at this isn't just a tokenized model of gaining uh access to an investment opportunity, right? You know, we've seen crazy things. We've seen tokenized cows and, you know, all all sorts of, you know, weird tokenized RWAs, right? And I I I'd imagine that, you know, there isn't uh a class of investors that are out there

that want exposure to the upside of the price of this hardware, but don't really have a need for actually using it for tokens. But no, what you're saying is that you can both own it for an investment and you can also use the technology while you have financial exposure to it. >> Absolutely. Absolutely. So, it depends on what you're looking for. We think of our audience, at least our initial audience for this launch as two primary segments. One is if you're an AI startup or you're a researcher at a leading institution and you are bought into the idea of owning compute, but it's difficult right now because you don't have the technical expertise to set things up or you don't have that much capital upfront because these machines

are expensive. We offer this one-stop shop platform. We'll help you buy it. We'll finance it. will set it up and you can monetize it if you wish when you're not using it. The second person I think Rob is what you're getting at where there are a class of investors especially I think in the US that have portfolios that are just cash flow generating right I have Airbnbs on the side they generate income every so often there's a property manager that looks after everything I actually just put in the capital they manage the bookings they do the setup and the cleanup in a uh I guess in a similar way we offer something like that but for compute right and and if so you want to bet on

hardware as an asset class and not just Airbnbs as an example you can come to us right put some capital down we'll procure that hardware for you host it and help you monetize it essentially by selling that capacity out to you know everyone out there who wants it at the moment >> I I want to get into some more of the nuts and bolts about you know the rent verse own uh you know equation and balance there and then also some of these these different customer groups Before we get too much deeper though, I I just want to catch up a little bit on B3 because we did have Victoria on the show. I want to say I mean maybe it was a year ago at this point, but but you know quite a few months ago and it feels like uh a a shift. Um if I remember

correctly, there was like this NFT gaming component. And so maybe you could just describe the evolution of B3 and NPC Labs. How did you go from NFT gaming to B3 IQ AI compute ownership platform? >> Yeah, absolutely. Great question. Great question. Also, before the call, Victoria and I had a team meeting and she was saying that she loved working with you all. She loved the time on the podcast. She was supposed to be the one to come on the podcast, but for some reason the team decided, you know, the next best option was me. So, you guys are stuck with me for the moment. But just to just to give you the background and context. So I think it starts from back our I guess our time back at

Coinbase, right? So we we started at Coinbase. That's where I met Victoria. I met Sean and some of the other team members that are now core to V3 and we were always working on NFTTS infrastructure problems at the company knew bats and when we left we decided to take essentially that expertise and apply it to gaming that just felt like the natural fit. We were all gaming nerds. We all love the idea of maybe playing a game that you enjoy and also monetizing at the same time. But the gaming thesis, I think while the fundamentals make sense, I think it's going to take more time to bake on integrating NFTTS and onchain components into games. While we were in gaming, we

actually saw uh firsthand how GPUs and you know this new asset class was rising. Even early 2025, we saw that you know GPU prices were skyrocketing. Actually, a lot of gamers were quite pissed. They were extremely angry at the fact that like this GPU that they bought for their, you know, gaming PC used to cost 3,000, now it's like double the price. So, they were actually up in arms about it. But we saw that as a data point and a signal that AI is something we should be looking closer at. And so, we took a step back and evaluated what can we build and deliver in the AI space that's value ad. And we landed on two things. One is software. So, that's B3S.

We built a agent harness that enables um you know secure and reliable uh onchain transactions for uh agents and humans and then B3 IQ which is uh coming out of stealth today on on the hardware side enabling anyone to uh to own compute. So I think in a lot of ways you know there's companies that start somewhere and then they end up you know in in kind of a different place. Um Nvidia did that. Uh Slack did that. Right. I was going to mention it sounds very similar to the Nvidia story of starting in you know hardware for gaming uh which was what GPUs right uh were were predominantly used for and now obviously we've got this very very big use case for GPUs in AI.

>> Yeah absolutely. I think the thing that sucks, I mean, just talking off the cuff about gamers or being a gamer, you're kind of always, you're like at the forefront of something, usually at the front of all these tech trends, but you're kind of like the guinea pig where like, you know, once it works for you, then all the other, you know, builders and all the other use cases pile in and then gamers always seem to get the short end of the stick, right? like they tried experimented experimented on but like you know they get a worse user experience at the end of whatever tech trend is going on.

>> It's like the uh it's like the bands who get famous and then they change their music style. It's like you want the guys that are underground. That's where all the culture is. >> Yes. Yes. Yes. Absolutely. >> Well, it makes sense and I mean you know similar similar trajectory to Nvidia. I'm hoping you know it reaches similar scale here for B3. Um, and so let's talk a little bit more about, you know, what you anticipate from some of these customer segments. Um, and walk us through why, you know, we might choose to maybe maybe just some of the metrics and the math here about, you know, kind of the the you mentioned $200,000 in rental costs, $400,000 to ultimately, you know, rent to own or or upfront costs here. walk us through some of the

math in terms of a user that might be coming in, you know, they're looking at I mean, you know, 20 bucks a month or 200 bucks a month for some of the pro models. In those cases, does it make sense for them to go ahead on B3 IQ and, you know, own are you looking for more of a broad distribution play or is this primarily a corporate offering that you're looking at for, you know, some of the the companies that are uh, you know, token maxing through these uh these uh model APIs? >> Yeah, I I think you bring up such a great point there, Rob. think there are large enterprises and folks that are on you know cloud max or codeex max or whatever chat GPT calls their plan today

that are essentially getting great deals right they're getting fantastic deals to prompt claude or prompt codeex you know a million times a week right because these companies ahead of their IPO are subsidizing the crap out of these programs so in in a way that makes a lot of sense. I think what people often underlook or underappreciate is there's still a cost to using these centralized API providers if you're just a user and a consumer of inference. And that cost is your proprietary data, right? And so if you're, you know, a personal user like myself, I'm giving Claude, you know, my deepest, darkest thoughts. I'm

also giving Claude, you know, my personal health data, right, to track my risk and my family's health data. And so, I think from a consumer perspective, like I'm fine making that trade-off, like paying 200 bucks a month for Claude to be my therapist plus my like physician, fantastic. I think when you're a company, small or large, the opportunity or I guess the potential cost of leaking your proprietary data information that I think is pretty significant and it's something that people are becoming more and more aware of over time. And so the viable alternative is using open way models that we all know are getting better if not you know exceeding on certain dimensions and workloads. And the way to

use those, you can rent from, you know, an Opra router or you could just install those on machines that you own, right? And going back to the calculus of owning versus renting. If you can combine a machine that you own and it's cheaper than something you would rent on a long-term five, seven, eight, nine year basis, then really it's kind of the best of both worlds. And and that's what we offer to someone who's looking to, you know, uh reduce, I guess, token maxing and also push forward what we're calling like, you know, private inference and owning your own intelligence. >> And so there the the value proposition of this is twofold for the customer segment that is looking to actually use this hardware. There's the one side which is the investment proposition.

Owning rather than renting actually leaves you with some principle at the end of the day. And there's also this uh private inference or local inference it although it's not local but private in the terms of your data not getting leaked. Um and so am I correct in saying that uh you know one of these frontier model companies if I were to use claude or chache BT I I have to use their stack I can't run those models locally on my own hardware. So, Frontier Labs, no, they don't publish their weights and they don't allow you to install them on your MacBook or local machines. Open

weight models you can, although it doesn't go beyond that. Only the weights are published. So, there is still some risk there inherently. I would say in the future I expect open AI andcloud to have a mix of maybe closed source models for >> frontier tech right frontier workloads and then open source models for everything else because it's going to be really hard to compete with these Chinese models and even thinking machine right that came out with something that's just as good if not better than fable on certain dimensions I think that's a really good point >> uh thinking [clears throat] machine being the US openweight uh alternative.

Um and so yeah, and and just one more thing I'd throw in there is that uh you know, we may see these frontier model companies with openw weight, closed weight models that are available, but then also potentially even internal models that they are choosing not to release open or closed for public use. Um you know, whether reasons uh various for various reasons. >> To your point there, Rod, they might release their own hardware, right? They could be releasing laptops, local inference boxes, phones. I know OpenAI has still been exploring that. So there's different ways I think for them to get their models in front of the right audience.

>> Absolutely. Absolutely. And so there's this private inference uh that that you can run on your own stack uh that you can pair along with these open weight models uh potentially save on compute as well if you're token maxing on these frontier models. And then also uh you know the investment uh upside that could come from owning the stack. You know that's kind of on the on the segment that is going to be using the hardware as if if we were to jump over to maybe you know the investors that are looking at this purely as an asset class and potentially even you know a passively uh generating yield passive income asset class. Can you just walk us through the math? Like what is the current you know yield that someone can expect from

owning a a Blackwell uh or you know kind of one unit of compute? >> Yeah, absolutely. So I'll try to simplify this as much as possible and people have detailed questions. Ping me, call me, text me. I'm available anytime. But let's just take a very simple example. Is there actually a way for me to share my screen or >> Yes. Um, you can. It's going to go ahead and share your screen here in the studio. Uh, we we uh sometimes we do this on the fly. So, all right. I see your screen there. Just give our producer a moment to pull it up. I don't know. Aon, do we have some uh elevator music or something that we can play [laughter] in the background?

>> We we might have we might have something for these little transition uh transition things that happen once in a while. put you guys on the spot. So, I should have prepared this earlier, but I think it's a great question. I want to make sure I answer it. >> Okay, we're uh I'm going to give you a heads up when uh when we do have uh Oh, I hear a little bit of music playing in the back. It doesn't sound like elevator music. >> There we go. >> Okay, >> there we go. Now we're we're we're getting somewhere. All right. I see our producer.

>> It's kind of It's probably what uh Ray Dalia was playing in Aiza last night. >> That was hilarious. seems like 7 77 years old. >> I know. How did the man not have a heart attack? Jesus Christ. >> That was wild. What do you think, Ray Dalio? You think uh he's got a the right idea here? The world's coming to an end and civil US civilization is uh at a turning point. >> He's going to say that when you're ready, Dalio, you know, it just it's even if the world, I guess, goes into a turning point, he'll be fine, right? But what about the rest of us? I mean, >> it just it's such doomerism. I don't know. I don't know. What do you think?

[snorts] >> Oh man. I mean, you know, you can see the the infinite money printing writing on the wall, right? So, we've kind of chosen to opt out into, you know, the Bitcoin and gold economies. They do. I mean, they say they being just people say that, you know, that hyperinflation sort of thing tends to occur at, you know, the end of regimes or empires. So, I don't know. Correlation's not causation. So, we'll we'll see if this time is different, but uh you know, people say that for I think maybe that does happen. Uh it's just does it happen now? Does it happen over the next few years or does it happen in another 250 years for America?

>> Yeah, well said. Well said. >> We're ready. Okay, that was uh that was great. Little little uh sideway uh segue there. Um okay, I see it. Let's pull it up. get the the screen on the uh uh Daryl's screen here. Let's go ahead cut to it. And uh Darl, maybe you could just tell us what we're looking at here. I see we've got $292,000 for a machine, a 8X H200. Uh give us a little bit more. This looks like the B3 IQ platform. >> Yeah, absolutely. So we built a earning calculator on the platform so anyone can just come in and just better understand the earning potential of some of these machines, right? And none of this is

financial advice. This is just numbers here that are based on real data, but we're giving folks an opportunity to evaluate, you know, the investment for themselves. So to your point, let's take the ADX H200. The way to think about this machine, it's used quite frequently for training runs. It's one generation older than your black well, but because black wells are in such short supply, the prices of H200s have actually risen pretty dramatically. Even H100s, they've been up 40% since the beginning of the year, I think H200's up even more. When you buy 8x of these things, it's essentially a node. And usually when you train a model, you load them on multiple nodes in a cluster. So then you're able

to load and train the model more effectively. The cost of these things, they're not cheap. So it's almost like a Ferrari or a high-end vehicle or a property, right? And so it's 200 between >> this and a Ferrari is this one's going to appreciate as soon as you drive that Ferrari off the lot. It's gonna Oh, yeah. It's going to cost you. >> Yeah. Oh, it's going to cost you a pretty penny. Absolutely. So this one is actually, you're right, it's appreciating these days because there's so much demand for them. But let's say on the lot, you know, you're getting it for 292K and the costs there are like pretty high a sticker cost even for a startup that raised a significant amount of money, right? 200 300k is something not you know uh nominal. So uh what we do with

our financing model is uh we actually give everyone the opportunity just to put 30% down on uh the machine itself. So, let's say you're looking at this purely from an investment perspective. Instead of buying this Airbnb, you're [snorts] buying this machine, right? That's 300K just like a mortgage. You only have to put, you know, 30% down. So, you're putting about 87K down. The cash that's coming from the machine that you're getting in your pocket day one is about three grand. And that's, you know, this is how that breaks down. Essentially um the machine uh you know this node we're able to sell the capacity to AI startups in the space to other folks that have demand for compute and inference that you know topline

gross earnings is around 14K. Um you minus 10K here you subtract 10K uh because you know there's a there's a share of um you know that revenue that goes back to pay the balance that you owe. So the difference between that 87k and the 300k, right? And then the remaining 30% rest share is about 4k. There's a hosting fee that we charge for, you know, assembling, hosting, power, cooling, all of that. And so at the end of the day, you're getting about 3k in cash per month. And the nice thing is because there's so much demand for these machines, you get to own it in about two years, 25 months. And if you're looking at it strictly from

investment metrics, the IRRa, the four-year IR is really high. It's 63%. Right? Because you're only putting 30% down. So, it's essentially buying these machines on leverage. And [clears throat] I think from an investment perspective, we've had really good conversations, really good appetite from folks that don't want to put, let's say, you know, money in Airbnbs and properties because interest rates are quite high, but want a different way to express a viewpoint and an investment on compute, right? And this is one of the purest ways to do that because you're actually owning the hardware and making money on it.

>> Fascinating. Appreciate the walkthrough and the live demo. Um, couple questions here. Um, oh, we lost your screen. >> Okay, let's go shared. [laughter] >> Um, >> okay. So, uh, what a couple questions here. One is who is the lender here? I'm putting 30% down. Someone else is covering the 70%. Is that you guys or do you guys have a capital financers that come in and provide the other 70% of this loan? Yeah. So there's very few capital financers out in the space and most of them support institutional organizations, institutional setups like your hyperscalers, right? And your data

centers of the world. So core, weave, mbis, etc., right? And not many folks do it on like eight machines. So actually we step in. So we take money off of our balance sheet. We buy the machine in its entirety and essentially we provide the financing support that you need to get the machine. We assemble it. So in that way we act as that like one-stop shop where you don't have to worry about too much capital hosting assembly or monetizing it. >> And I take it you know this uh managed hosting you mentioned all the infrastructure buildout connecting the machines setting them up putting them where they need to be. There's also the aspect of going out and sourcing demand for the compute itself, right? You got

to go out, you got to talk to an AI startup, you got to get someone to come in and utilize the machine. I realize that everyone's token maxing these days. There's tons of demand out there, but you guys are out there forming relationships and getting the customers to ultimately consume the compute that comes from these machines as well. >> Absolutely. And I think right now we are in the golden age of compute, if you will. There's a lot of demand for compute and there's literally not enough machines on the market. It's a complete sellout when it comes to black wells.

You're lucky if you find, you know, more than 500 GPUs of Blackwells or, you know, a H200 node that's a thousand GPU count that's available right now. So there's a severe supply shortage and so what we do is we we try to focus on what we do best which is procuring these pieces of hardware, assembling them, hosting them and in terms of the monetization and pairing it with demand. There's folks that text me almost like once a day now asking if we have extra GPUs for capacity. But we also uh in order to scale this uh partner with different folks um both the marketplaces and also some of the brokers out there

uh who have you know demand that's kind of their job all day and they pair that demand with you know the supply that we have on behalf of our you know investors and also you know the AI startups that aren't using it full-time. >> Fascinating. I'm I'm hopping around a little bit, but I'm just kind of throwing questions at uh your way as as I think of them. Uh next one is um as you know, Nvidia continues to build this the you know the new and invent new uh uh machines.

What has been the typical trajectory for these machines in terms of their value? And and the reason I ask is because obviously you know 63% as we see here is a tremendous yield. Typically in the yield farming DeFi summer days on chain >> 63% didn't last very long you know like [laughter] it was >> it was driven down by token emissions and so even if you were getting 63% you know the the principle would depreciate. We just so happen to be in this AI super cycle. Uh shout out AI super cycle name name of the segment powered by near. Um and so we actually have this converging confluence of you know the hardware becoming you know it's appreciating uh

at the same time that we have this incredible yield but I take it once new machines come out right the old machines might depreciate. Can do you have any data or or you know historical examples about uh what the prices of these old machines tend to do after a new machine is released? >> Absolutely. I think that's a great question. One I get a lot around essentially the residual value of these machines, right? And I think what's crazy about the moment in time we're in for the first time in PC parts, computer industry, dynamics, like older generation pieces of hardware are actually appreciating or

holding steady in terms of price. So the floor has risen on older generation pieces of hardware like the A100, the H100, even some of your 3090s, consumer grade level hardware have also either held the floor pretty steadily or appreciated in price. I don't actually think that's going to last, right? As Nvidia comes out with Reubins, which is supposed to be sometime right in the next 12 to 18 months, as Nvidia comes out with more generation or I guess next generation pieces of hardware, it absolutely will have an impact on the prices of older generation pieces of hardware. I think the numbers that you see here, what we've baked in is a pretty steep decline in the hardware rates per year. So essentially you see a

20% decline in the prices of hardware year and I think that's a very conservative and aggressive I guess conservative but also aggressive in the sense that like most folks in the space don't think the hardware will depreciate as much because they see so much demand for compute coming right now the demand is you know about like 55% for training the rest for inference but talking to folks in the space the inference demand is just growing month-on-month because of how good these openweight models are getting, right? And so to me, how that translates into hardware prices, it's hard to say a specific number, but I think the floor will be higher than expected because these older generation pieces of hardware will still have a

useful life beyond like three four years primarily if you're using it for inference or you're using it for your own needs. And I think the numbers here are interesting because you can own it right in in two years and after that it's kind of like a free call option to the upside. Makes sense. Yeah. And is that the same case for the yield on these machines? Do do you anticipate, you know, even if there's a new machine, uh I'd imagine that there's still demand. Maybe it's not for the hardware, but actually still using the hardware, you know, do you do you anticipate uh even after a new machine the yield uh compressing or or staying relatively stable?

>> Yeah. So, one difference on our platform is the offtake, the demand that we match these machines to. We aim for longer term contracts. So in the space right now, let's say Rob, you buy a machine, you host it, you assemble it yourself in your house, right? Which I'm sure if you have a significant other, they'd be kind of pissed about because these machines are really loud. And I have like a bunch in my bedroom right now. My wife is super pissed at me every single [laughter] night. But if you if you have one, right, you're hosting it. Um you could go on vast or you can go on open router even if you install the model in it and uh use it like serve inference monetize it yourself. Um the difference

between that and what we do is uh we try to go direct right so we get larger purchase orders of demand and these are multi-quarter or multi-year contracts. So essentially when we sign the contract we're locking in that rate for multiple years and that's why we're able to underwrite you know some some of the kind of projections that you see on here because these are not let's say on demand rates that will like fluctuate or be volatile over time. >> Got it. Okay. Make makes sense. And um another another one here is uh we see on the left side of your screen this utilization assumption. Um you got that set at 80%. Could you just explain that metric a little bit more in detail and

what causes that to fluctuate? >> Yeah, absolutely. So, high level utilization is how often the machine is being used, right? And so, right now we set it at 80% because in these multi-arter multi-year contracts, we are essentially assuming 95 to 99% utilization from, you know, the purchaser, right? So if you're AI startup that's training the model, you just pay for a year's worth of compute, right? You're you're paying based on the number of hours in that year. Whether or not they use it the entire time, that's another question, but they've effectively paid for one year, right?

And so translate that to utilization, it's basically 100%. But, you know, that's a let's say two-year contract, right? And to sign another two-year contract, there's likely some gap in between, maybe like a month or a couple months. And so over the life of four years, let's say, you know, to be conservative, we want to say like, okay, this machine will be used 80% of the time. So that's that's one way to think about it. The other way is when you go, let's say Rob, you're using your own personal machine and you go on the on demand marketplaces, the term lengths are usually much shorter. So they're like less than a year, sometimes they're a week, sometimes they're a couple months. And so, you know, the utilization rate varies dramatically.

So, you know, if you have the opportunity and you have these machines, ideally you would sign multi-year B2B contracts. Like that's where like you can lock in the rates and that's the best risk adjusted, let's say, you know, um contract that you can sign in um Inc. >> Okay, got it. Um do you guys amortize these contracts that you have with uh the direct consumers of the compute for instance like uh you and I both you know own a own a you know node of H200's someone is you know renting mine uh someone's renting yours and then you know my contract ends but yours continues. Do you get the yield or do we

share it? >> That's a great question. So right now it's paired on a direct onetoone basis. >> Okay. >> I think the next product that we're considering and this is you know been a couple quarters in the making. The next item on the road map is this is a great starting point for someone who is really eager to express a different view on compute by investing directly in hardware. But I've always thought what crypto has done a great job of there's been many misses, right? But I think what crypto has done a great job of is abstracting, I think, complicated, you know, investment instruments into something that's simple for the average person to understand, especially on the retail side. I'm pretty dumb when it

comes to like how I like to invest and ape in certain things. And so, how can we turn this into that kind of product? And there's some ideation going around. One idea is essentially to turn each of these machines, each of these offerings, the squares that you see here into RWA vaults, right? So an ADX H200 instead of need to understand IRR payback period, all of this, right? >> Minimum investment size of maybe like $1,000, you put a thousand in and you get a certain amount of APY per year, right? And if you want to redeem it for some ownership in the machine, that's fine. There's probably a legal structure we could set up to do that, but you don't have to understand like whether

this is, you know, a 2026 Ferrari or 2024. You just look at the APY and you get, you know, you make your choice, right, based on that decision. So, we're trying to simplify it. >> Yeah. Yeah. I think that would certainly simplify it in the eyes of, uh, mercenary capital allocators that are simply looking for a yield. Uh, and that is a customer segment that this industry has appealed to uh, in the past. So, >> absolutely. since the days of total Bitcoin mining back in 1314. >> Yeah. Yep. Totally up to you guys if you guys want to attack that customer segment. But, uh, if you do, uh, wish you guys well. Darl, I think I might be out of questions here. This has been fascinating. Um, I I really appreciate

it. Oh, the other thing I was going to mention is if uh, you know, there is a direct to consumer onetoone pair with these machines. I want to know who's renting out my hardware. like I want to know, you know, what is their company name, how much longer they have on their contract, these sorts of things, so I can keep up uh with them, make sure that they uh, you know, they're they're making good on their contract. >> Um, >> Darl, I mean, I'll just pass it to you, I guess. And any closing thoughts here. I I really appreciate the walkthrough, the live demo. It's not something that we get every day, so kudos. It takes a a brave man to do a live demo on a live show. So, kudos and uh congrats on bringing this out of stealth. Thank you to you and your production team for

handling that last minute request. Beautiful uh setup and also beautiful music in the background. Um I think Rob to your point you you do want to know who is using your machine. I think the way we do that is through the dashboard so you can log in and you can see exactly who's using it, how much they're consuming and the rates that you're receiving. And also most importantly there's export controls right now. I have to emphasize this again. And I think my lawyers make sure I say this time and time again. You know, AI startups from China, mainland China, and other countries on that list cannot use, right, these workloads and machines per the US export controls requirements. So, it's just something to

call out why we have that interface in the dashboard. So, you know your customer, you know who's touching the machines. But, yeah, I think if uh just to end things really quickly, like this was awesome, Rob. Thank you so much for the time in the space. Um, if folks are curious about B3 IQ, uh, they can come to the website. They could also follow us at, you know, B3 Labs on Twitter and X. And, you know, if you have a need for compute or just want to check compute anytime, uh, it really feels like crypto back in, you know, like 201 uh, 145. I started really digging into crypto in 2015 and the space feels like that where everyone is just trying to uh understand the space more deeply from a technical

level, from a financial level and also really help each other out versus it being so pee. Everyone's like, you know, working together to build new offerings like this and trying to make compute more accessible to the average person. Yeah, these uh these market conditions create a lot of pressure, but uh there are some, you know, rockstar people and companies that are born out of that uh and those market conditions. Uh Darl, thank you so much, man. Absolute pleasure. Would love to have you on again soon. Uh I do think that this is certainly uh more I think it's just cooler uh and uh a very a very uh you know, interesting development beyond what we saw in the first rendition of crypto AI. I think this is a step in the

right direction. So, thanks for fighting the good fight. Congrats on the launch out of Stealth and would love to have you on again soon. >> Amazing. Amazing. Thank you so much, Rob.