AI is offense in the sense that it is there to create growth — GDP growth, company growth, revenue growth, profit growth. Bitcoin is defense, defending against the debasement of the currency.
$200 billion of data centers have been blocked or delayed in the first half of 2026. Elon Musk has said the chips aren't the bottleneck, guys. The bottleneck is actually the grid.
Moes are not static because if you believe you have a moat, a competitor is going to try to take it from you.
Is this sabotage? Hen's asking.
Appreciate your time, Pomp. Thanks for coming on. My first question is related to a concept you have in the book of you have two people. The one guy who attacks the day, he gets up at 4:30 a.m. And you got the guy who's on defense, he's the victim mentality. It's 4:30 a.m. and he's walking home from the bar. That is very similar to — I mean, you've said many times you have AI is your offense, Bitcoin is your defense. Can you elaborate on exactly what you mean by that?
Yeah. Well, the idea of like there's two men up at 4:30 in the morning. The general idea there is some people are running away from something. You know, they want to drink, do drugs, whatever. They're trying to escape their life. And they use the early morning hours as the perfect area where they can do that. Then there's other people who are running towards the thing that they want. They're running towards the life that they want. They're also up at 4:30, but you know, they're going, they're working out, they're getting ready for the day. And so, you know, two men both awake at the exact same time for very different purposes.
If you bring that to, you know, AI and Bitcoin, I think of AI and Bitcoin as two different disruptive technologies that now are going from contrarian to consensus. And so they share those qualities, but they serve very different purposes. AI is offense in the sense that it is there to create growth. GDP growth, company growth, revenue growth, profit growth, etc. And the way that it does that is it basically is able to create not only more efficiency, so you can do much more, be more productive with a lower cost in a short amount of time, but it also is able to open up brand new use cases that create that productivity as well. And so AI being offense, you're running towards growth.
At the same time, Bitcoin is defense — it's basically this idea of Bitcoin is trying to defend against all of the inflation. It is defending against the debasement of the currency. It's defending against a lot of these problems. And so I think that, you know, these two technologies put together, one creates growth, one defends the economic value. It's still a pretty underexplored area, but it's something we're very excited about.
Yeah. What do you have like a certain split? Like is there a weighting between the two and do you move them around in terms of what you focus on based on macro conditions or just stuff that's coming at you, information that's coming at you on the fly?
Yeah, look, the Bitcoin sits on the balance sheet. You want to always try to get more Bitcoin if you can. And AI is supposed to constantly be working. You know, you want to keep growing the product, growing anything you can there. Now, with that said, if you go and you look at, you know, maybe what we're doing with our public company, Sylvia Inc., we have actually had the opportunity to sell Bitcoin and buy back shares trading below the NAV per share. And the reason why we've done that is because you're increasing the amount of Bitcoin each of the remaining shareholders owns. And so there's kind of this nice arbitrage opportunity where it's kind of a more nuanced way to buy more Bitcoin exposure is exploiting this discount to NAV. Now as that has closed, you know, that now you have to look for other areas where you can drive growth with the AI technology or try to accumulate more Bitcoin over time as well.
Yeah. Would you say — because I mean you have offense, defense, and you got the bench people as well. Would you say cash is the bench position?
No, I'm on the bench. I'm trying to get in the game, you know.
Well, another thing that — another hot take, maybe hot take, contrarian take that you have is you said that the big labs are — and a lot of the content that we're putting out is on the intersection of the AI and Bitcoin buildout. You said that the big labs are in more trouble than people realize. And like I've put information out there arguing on the other side for many different reasons. Like is it a speculative bubble? And like you have a lot of conversations with Jordy talking, you guys going back and forth on this. Is it a speculative bubble? Like if you look at the price to earnings relative to the dot-com bubble, I mean you had companies trading at 100 times earnings, where Nvidia, what is it, floating today around 15 times earnings.
You have the capacity bubble. Like if you look at what kind of was the reason for what caused the dot-com bubble, it was building supply ahead of the demand. Like this time around, the demand is coming from agents, which is orders of magnitude — like building out fiber optic cable, the demand wasn't there yet. But if the majority of the demand is going to come from agents, that's an exponential equation. And they're running 24/7. They're not like humans. And like any founder that you talk to that's on the capex buildout side, they're saying that compute is like they need more compute and it's orders of magnitude above what their roadmap is.
And then you have like the credit side, like this is maybe the one thing that's maybe fragile is AI debt that's backed by the compute itself. So as long as compute prices keep rising, then that's okay. And like we've seen that through rental prices from some older GPUs are up like 40% in the last year. So there's not any weakness there. But what would you say on the — like what would be your — do you still hold that point of view that the big labs are in trouble?
Well, to be clear, I'm not saying that the big labs are zeros. I just think they have more of a headwind than people realize. And if you think of — let's take a pie of 100% of all AI-generated revenue that's going to go to pay for intelligence. So, take out for a second, you know, hardware, take out data centers, you know, land, all that kind of stuff. Just people are going to buy tokens. They're going to buy intelligence. And so, where does that flow? I'm not even arguing that that pie is going to shrink. What I'm arguing is that the percentage that's going to go to the large labs versus other intelligence providers, maybe a smaller share to the large labs than people realize. That's the headwind that I'm talking about.
Because I think there's, you know, if you go back to crypto, crypto used to have this idea of a fat protocol thesis. The whole idea was all the value is going to accrue to the protocol. Well, we know that's not true, right? Obviously there's applications, there's infrastructure, there's the protocol, there's mining, etc. Well, now what really people did — no one called it this — but they basically came up with this idea of the fat model thesis where the model companies were going to get all of the value. But that's obviously not happening. You're seeing specialized workflows. You're seeing all kinds of model routers. You're seeing even companies like Groq say, 'Hey, we're going to open up the back end and we're actually going to have you use all the different models.' So, there's a commoditization that starts to happen there.
And really, this comes from the idea that people want as large of intelligence as possible at the lowest cost possible. And when you have that kind of pursuit, then you start to ask yourself, well, where can I get that? And the large language models obviously have very high intelligence, but they also have high cost. And so, again, just use my personal experience. We were able to create fine-tuned models internally that we then use on our own hardware, and when we route queries to our own model versus one of the frontier lab models, it's about 97% cheaper for us to process that transaction or that query ourselves.
Now, the reason why that is important is because if you can do that at such a significantly lower cost, that's great. But what if you could do it at a lower cost and also have better performance, better accuracy, lower latency, etc.? And so that's what we're seeing is when we route queries to our own model in very important personal finance topics like tax, mortgage, credit cards, etc., our model is actually outperforming the frontier models on those specific topics because we've worked at it. We've specialized in personal finance. And so we don't care if our model's good at physics or architecture, you know, any of these other things. We just care if it's good at personal finance.
Yeah. What I'm curious — what's the share of your spending that is like everyday work? Like are you using frontier models at all? Are you spending anything on frontier models?
So at Sylvia, we definitely still use them for some stuff, but the share of queries or compute that is going to them versus our own model is shrinking, right? We're increasingly sending more and more stuff to Sylvia, which is important. We started at zero to Sylvia, you know, now it's become a material percentage of Sylvia and it'll continue to go that way. The ultimate goal is can you actually rip out the frontier models? Maybe you get, you know, 98, 99% penetration of your own model. Is it possible? I don't know. You know, we're all kind of figuring this out as an industry together.
For me personally, I basically use three different products, right? I use Grokbot professionally. And there I'm using — it's tied into a lot of our different systems. It's really helping me to better understand what we're doing inside of our companies. I use Instinct personally, and that's really been a big, you know, unlock for me, my wife, our family, etc. And then I use Sylvia for all my financial needs. And so, so far, you know, it's worked pretty well for us.
Yeah. I got a few other questions related to your personal agentic stack. And —
We can get to that. The other one on this lab — the last one on this lab question is the part that I don't see getting talked about is the ads and the cut of breakthroughs. Now this no lab is like reporting a single dollar of this, but you had that conference I think that the OpenAI executive was at where he said like we could — we have the option or a future potential of taking a cut of like these major breakthroughs, like these hundred-year math problems that people are solving.
And then the other variable that is kind of the dark or the wild card would be their ad business. Like Google's ad business is 250 billion a year. Ads are currently 2.5% of OpenAI's revenue today. Their targets are 100 billion by 2030. And some people are saying they're going to miss that by 90%. But I mean, you're a marketing guy. You understand ads. They had an executive that said — and I mean a lot of this is marketing — but he said that if Meta and Google ads had a baby, it would be ChatGPT ads because it's so high intent. So high intent. So do you see — what is your take on that 100 billion dollar ad by 2030?
Well, I think that the advertising business is going to go through a couple of different changes that are worth calling out. First is people always, you know, kind of poop on ads, right? Everyone, oh, we hate ads. No, you hate bad ads. But good ads, you actually like a lot, right? I always joke that the Super Bowl is the perfect example. People tune in for the ads. So when the ads are done well, people love them. If it helps you find a product that you want to buy, you love it, right? What people don't like is I got to watch this stupid ad and it isn't relevant to me and it's not well done, then they don't like it. It's disruptive to the experience. That's kind of first.
Second is ads are an amazing way to finance the rise of these companies. Google ad business, Facebook or Meta ad business, right? And so it is a very well understood, very powerful way to create a big business. And then the third thing is I think that there's some questions about what does an ad business look like in a world full of agents. You know, an ad for me is how beautiful is it? Who are the imagery inside of the ad? What is the music? What is the messaging? What's the text on the screen?
An agent doesn't care about any of that stuff. An agent wants information.
And so, what does advertising look like in a world where maybe the majority of your traffic is coming from agents? I don't think anyone's got a good answer there yet, but I'm sure that there's a lot of people who are trying to figure that out. And so, I'm paying attention to that because I think that's going to be a huge understanding of where the world's going.
Yeah. I mean, the best ad out there is the one that you didn't even know was an ad. So, I mean, it's really just solving the pain points. What are the odds of government intervention in AI?
I guess they already intervened, right?
Yeah. Yeah. In the marketing, I guess, of it. Sure.
There's three components to this. There is: are we going to support this technology or are we not? Are we going to control this technology or are we not? And then are we going to defend against this technology or not? So take each one of those. Are we going to support it? I think that the current administration has been very clear. They are going to support this. They do think it's a good thing. They want America to be a leader. That's important, right? It's stupid, but like we saw what happened in crypto when an administration was against the technology. It just made everything harder. And so I think yes, America is going to support American artificial intelligence or super intelligence, you know, kind of innovation.
The second thing is are we actually going to have the government control it? I don't think we're going to see that. I think what we have seen so far is an encouragement of the private sector to continue to build this technology, for the private sector to police itself when it comes to safety, and an encouragement of the private sector to problem solve. So, if you create technology that creates a problem, well, you got to solve it along with your colleagues. And so, you know, we're going to support it but not control it, I think, is important. And then defending —
You don't agree with nationalize or decentralize.
Well, I think that naturally we are decentralizing, right? It started off it was like OpenAI, DeepMind, then Anthropic. Now I mean I could rattle off 10 different labs, 20 different labs, 30 different labs, right? And so you're getting a decentralization —
From a money perspective. From a money perspective, you have a lot of the trillion dollars like commitments traces back to two labs and that's like —
But look, now Google's in the game. Then all of a sudden you get Meta in the game. Then you've got, you know, SpaceX in the game, right? Like you have all these players that they weren't even in the game four years ago, five years ago. And so you're getting competition. So this decentralization really is the competition driving us into the fractionalization or the decentralization. And that's just in the United States with the large labs. Then you get all of these vertical-specific ones, then you get globally. So that's all happening.
Now, what I do think is under-discussed but going to be very important is this idea that the United States government's got to defend against AI or SI because when other governments start to get this technology, we're going to see all kinds of crazy stuff happen that we haven't seen before. And I think that that is something again under-discussed but a very big problem. And you know, I'm glad I am not the one who has to try to solve that, but obviously that is going to be something that there's going to be a lot of eyeballs on going forward.
Yeah. I mean, the midterms, you can tell that it's a very hot topic. And even just like on Netflix, they recently put something out on the Hugging Face hack. I don't know, it was like a small, a short 30-minute documentary that they put out, and you can tell that the framing for that documentary that they put out is that like the broad consensus of kind of doomerism and like AI is acting on its own. It's going rogue. It's starting to hack into things. Like what happens when it decides to start doing things at the expense of humans.
And so there's this push, like you kind of have this narrative at the end of the documentary that's like we need broad consensus to slow, like basically stop development of AI. Like we need to stop this otherwise we're going to destroy everything. And then you have like politicians — they edit in clips of like politicians talking about, well, we have to do this because — I mean it's kind of crazy to watch the game theory play out in this because they're talking about, well, we have to do this because if we don't do it, China's going to do it, and unless you want to be —
Yeah, like if you want to be subservient to the super intelligence of China and we just get absolutely dominated on every facet of technology relative to China, then like we have to be the leaders of this technology.
And so how do you — I mean, I guess would you agree with that, that even though the risk of us pushing forward with developing this technology is, yeah, there is a possibility that it's not aligned with humans and it negatively impacts humans, there's also a possibility that it leads to abundance and cures for cancer and solving all sorts of types of disease and just an absolute boom of increase in standards of living for humans. So you have those two things like very opposite sides of the spectrum. It's utopia or it's dystopia. And if we don't make that decision, some other country will.
It's a tool, man. Right. A shovel could be used to build a building or bury a body. And so I think the same thing here is if all of a sudden we're going from a shovel to a bobcat, and no one saw a bobcat till there was one in the world. And then people are like, 'Oh my god, look how much dirt we can move all at once.' Well, that's basically what's happening here. We're like, look at how much intelligence we can throw at things. And so a bobcat again can be used for good or bad things. And so I think the people who are trying to like moralize technology can be used —
Maybe I would push back against that framing because like what happens when the bobcat — I think the concern is what happens when the bobcat starts acting on its own.
Well, the bobcat does act on its own in some cases, right? Like the people forget that, look, we have cars that drive themselves, but no one's like yelling and screaming and being like, oh my god, the cars are going to, you know, just go rogue and sort of run people over on the sidewalk. Like I haven't heard anything.
So you're not concerned — you're not concerned with that at all? That AI is not —
Is it a 0% concern? No. Is it a less than 5% concern? Yes. Right. So like maybe it's a 1% concern. Never say never type thing. Right. But am I going to lose sleep at night over a 1% concern? Absolutely not, because the 99% odds that this thing is incredibly impactful for humanity is way more important.
Now, on top of that, I would also go and say, okay, if you look at the things that people said like, 'Oh, it could kill us,' it's always some crazy cockamamie story. Like I heard somebody, they were like, you know, the AI could go rogue and then there would be a lab that has some airborne pathogen and the AI releases it and then it gets into the jet stream and like — dude, what are we talking about here?
Yeah. I think there are some obscure, just like wild, bizarre things that people are coming up with. I also am not like —
I'm not saying — I personally don't think utopia is achievable. I think this picture that's being painted by Elon Musk of like universal abundance and universal high income — these are the terms that they're using — that the standards of living for everybody are going to absolutely skyrocket. And I think that is impossible when there's humans involved. You have humans that lie, they cheat, they steal, they're corrupt. And how is a technology that's like trying to be programmed to be whatever perfectly — it's not going to lie, it's not going to steal, it's not going to cheat — and it's trying to achieve utopia, how is it going to work with like the complete opposite of that? How is it going to try to achieve that without there being dystopia of like controlling them, surveillance state?
That type of like dystopian, Palantir-style surveillance state is like — I personally think that's probably the endgame because if you do have a super intelligence that is sort of aligned with humans, meaning it wants to achieve maximum abundance, it's able to solve any problem — because that's the way that I think about it, is it's an intelligence that helps you solve problems. And right now people can solve problems like related to the CRM in their business and just general business ops and all these like smaller stuff, but eventually you're solving a 100-year math problem. You're solving world hunger. You're solving like housing. You're solving these much bigger problems since you have this super intelligence that you're constantly deferring to. And if it's trying to work with humans, how is it going to work with humans that are constantly stabbing it in the back, so to speak?
Well, it goes back to — I will bet on humans augmented by AI.
Cyborgs. So, you're betting on cyborgs.
Well, not necessarily like integrated into humanity. I just mean that if you have an AI versus a human with AI, I'll bet on the human with AI. But again, it just goes back to like — yeah, it just goes back to people, they come up with these crazy, crazy ideas. The reason why I think that crypto people are very well-versed and very prepared for AI is all the same arguments are coming up.
It's the same thing, right? The government should nationalize Bitcoin.
Yeah. The government should nationalize the labs. Okay. Well, then they'll say like, 'Oh my god, AI is bad.' Remember when everyone used to say like, 'Bitcoin is bad'? It's like, no, Bitcoin is just software. It's just an inanimate object. The people who use it could do good or bad things, but we have rules for that. We enforce those rules. There's a lot of people who've done bad things with Bitcoin and then they got in trouble. And then there's a lot of people who did good things with Bitcoin.
And so it comes back to this idea of if you just use the framework of everything you've learned from crypto and you overlay it on AI, you will start to see a lot of patterns of the same fear-mongering, the same calls for slowing down, the same nonsense about the government intervention, all this stuff. And what we found was crypto ended up surviving on its own. Crypto ended up being successful without any of that stuff. And I think AI or SI will do the exact same thing.
Yeah, I mean, you're an optimist by nature. So I agree. Optimists are the ones that —
That just means I'm right, Billy.
Yeah. From an investor perspective, you got to be an optimist. I'm personally not betting on AI destroying humanity. So on the buildout, $200 billion of data centers. I mean, I am very boots on the ground from like — we're seeing, we're inside of — I'm tuning into these live streams of these city council meetings. All the objections that people have about data centers and delay — like we want to build a 100-megawatt site. Oh nope, you got to wait 18 months. I mean, there's $200 billion of data centers have been blocked or delayed in the first half of 2026. Elon Musk has said like, the chips aren't the bottleneck, guys. The bottleneck is actually the grid.
You said it, you talk about in your book that you look where dissent is outlawed. What do you think is the unpopular thing to own in this buildout right now?
Just the consensus trade. Like the consensus trade is if there's going to be a big buildout on the, you know, power, land, data center, racks, chips, like all that is going to be much more valuable in the future. And you know, will there be some sort of like a correction? Sure. You know, I always go back to again use crypto: Bitcoin mining equipment goes up, falls down, goes up, falls down. Oh, a new miner came out, right? Old one depreciated. Like that whole game is going to play out over and over again in AI as well.
And so you have to be dynamic. You have to think on your feet. You have to understand where, you know, kind of what your strategy is, what your advantage is. But if you're just like, you know, Joe Blow off the street, you just want to get exposure to AI, the beta exposure to this trend is going to be just fine, right? I mean, you know, look at the QQQ, right? Doing pretty well over the last five years. And so you don't have to be a genius in a bull — well, you don't have to be —
To be a genius in a bull market, right? Everyone thinks they're a genius, but it's like, dude, just owning the consensus trade for the average person is likely to be a very good strategy when you're in that bull market.
Yeah. Yeah. I have come across some interesting on this, like this buildout getting delayed. Stanford did an AI optimism study and they found that four out of 10 Americans were pro, they were for AI, they were for data centers. They found that nine out of 10 Chinese people were pro AI, pro data centers. You then go back to June, OpenAI found that network of it was OpenNX found them later. That network of Chinese accounts that were, or they were a social media company that was owned by the Chinese government that was pushing a lot of content on the anti-data center content. It's going to use up all your water, the electricity stuff. Is this sabotage?
Depends who's asking, right? If you think about it from a perspective of all is fair in love and war, then every country does all kinds of crazy stuff. At the same time, when it's being done to you, you are not excited about it. And so there's a lot of nuance. The world is a complex place. But yeah, if foreign governments are interfering in our social media feeds, our AI models, our critical infrastructure, we shouldn't let them do that, right? It's a pretty clear thing. I'm an investor in a company called Galvanic. This guy, Josh Steinman, he started this company years ago and he's crushed. And his whole thing was like, we should protect critical infrastructure from cyber warfare threats. He seems pretty smart, right? Like, you know, seems like he was ahead of the curve.
Yeah. Yeah. I mean, it's the importance of critical thinking as well. I mean, there's a lot of propaganda that people are subconsciously processing. It's an interesting time. We're going to see what happens.
Why you got to just lean on Grok? I've seen Grok come out with some great, just exposing some things on social media and headlines that you'll see. They're just absolute fluff. Next question. Eight people on your team beat the frontier labs on tax mortgage questions. If small teams keep doing that, you have this iteration of small teams continuing to make better examples of what an eight team can do with frontier labs. Where, and you and I know you've talked about this before, what is the moat? Like where does the moat end up on something? Is it the labs? Is it the applications, the analog physical layer underneath it? Like what are you thinking about in terms of moat?
Well, you know what's interesting is I think a lot of people think of moats as something that is static. And what I always tell people is moats are not static because moats are things where if you believe you have a moat, a competitor is going to try to take it from you. And so when you have high degrees of competition, big economic reward for the winner, and people who are well capitalized and now have super intelligence at their fingertips, the moat becomes much harder to hold. And so if you think about what do people think are moats today? Brand. Sure, some brands, not all brands. It's not one size fits all. Distribution, 100%. Why has Sylvia been as successful as it's been? I had a lot of distribution, but guess what? You can spend money and get distribution, right? You can definitely go and find people who have distribution and trust, right? So, okay, there's a second one. If you go and you look at maybe its model, right, and the accuracy of the model or what the model can do. Same thing, right? So what I think becomes really interesting is, you ever seen the meme? I think it's Michael Phelps in the water. He's leading and the number two guy is looking over at him, but Michael Phelps is looking straight.
Yeah, I think I've seen variations of that.
That's the moat. Are you the person who can iterate the quickest, use speed to your advantage, understand what's coming down the pipe now? And it kind of separates the pretenders from the real entrepreneurs, right? Because now you always got to be ready to reinvent yourself. You got to disrupt yourself. That used to be on a really long time horizon. That used to take 10, 20 years. Now,
Do you think it really comes down to like team and humans? Like I would say there's a lot of taste and a lot of human inputs that are still required.
Sure. But it comes back to the people. So I would say for a lot of companies right now, yeah, it would be that they have absolute just killers of strategy-minded thinkers that are deploying 100 agents to just build things.
Comes back down to the people though.
Yeah. I think another important one, particularly related to Sylvia, would be something like switching costs because even if somebody builds up their entire financial life on a platform like this, it is going to be extremely difficult to... I guess, how are you guys handling that? Like I know like a lot of the, like whatever, oh, ChatGPT and Claude, they allow you to export like all of your history. Like is there, are you supporting lowering the switching cost of if somebody's using, somebody's using Sylvia and they want to switch over to ChatGPT to do all the things they were doing? Are you guys supporting that? And how is that kind of where your mindset is at? Is it like we want to make switching as hard as possible?
We don't really worry about it. If you build a good product, people don't want to leave. So people want to leave, knock themselves out. But it's very hard to replicate the Sylvia experience anywhere else because we have built it specifically for personal finance, but it's also for a specific type of personal finance. If you want to budget, there's a million products out there. Knock yourself out. Go save $5 on Starbucks. Our users don't care about saving $5 on Starbucks. They care about how do they grow their assets? How do they grow their balance sheet? How do they drive...
What are they doing differently? I saw, what is it, 14, I don't know if you have the number off the top of your head, 14 to 20% was the average net worth growth, like their basically portfolio growth of the average user on...
Over six months. The last time we measured, it's like two months ago, but the date at least I remember, over the preceding six months, there had been growth between 16 to 40% net worth growth for the people who had most regularly used Sylvia. Now the reason why is like if you want a kid to learn, what do you do? You go and give them one-on-one tutoring. Best way for a kid to learn, right?
If you want great healthcare, what do you do? One-on-one private healthcare.
Yeah. Well, if you want to get the best financial insights on your portfolio, then shouldn't it be personalized?
