AI 实验室会被国有化吗？

I have some AI friends who are prepared to have a Los Alamos-style lockdown. They expect the US government to fully control them within 3 years. When you think about the most critical national defense project the US has ever had, the nuclear project, we literally put all of our nuclear scientists in the desert and spied on them. We wouldn't let them leave. They couldn't make phone calls.

Right? When we treated it like a military project, AI, in that mythos, is a military project, and the US government may start approaching it that way and say, “Oh, you're a top engineer in machine learning at Meta. You're only using a government-issued cell phone. You're only traveling with permission. You're not allowed to leave the country.” And if the US doesn't do that, we will be leapfrogged by China.

So, all of that affects these AI labs as investments, right? In ways I don't fully understand. Yeah, I'm not saying they're bad investments; I just haven't seen people seriously talk about discounting these factors.

1. From BlockTower To Burnout: Ari Paul's Crypto Story

All right, guys. Welcome back to another really fun episode of 1000x. I think this is going to be a good one. With me today, I have Ari Paul, who actually is my former boss, the founder of BlockTower, which was one of the greatest crypto hedge funds back in the 2017–2023 era. I'm just super excited. I've known Ari to be one of the most thoughtful people that I've ever encountered. He has great takes on a variety of different topics, can go deep, can go wide, and is always a really fun conversation.

Thank you, Ari, for joining me on this call. It's kind of funny to be podcasting with you because we used to work together, but—

Totally. And same compliments to yourself. I'm excited to be chatting with you. We catch up somewhat regularly, and they're always fascinating conversations. It's not a shock to me that, as long as I've known you, Avi, from almost immediately, I had a sense that, similarly, you were a very thoughtful person and we'd likely have interesting interactions regardless of the local professional context. So, it's been interesting following your career path, and I'm excited for what you're working on now.

No, I appreciate it. And I kind of want to start because I think our audience will find it interesting: where you started in crypto, how you got into it, and what really drew you to the industry, because I think you were drawn to it in a way that was both similar and also different from other people. Then we can get into what's happening now in the world and whether you think that vision you originally came into crypto with still pans out today, or if it looks different to you.

Yeah, it's funny. I did so much of the podcast circuit in 2017 and 2018, and then not much last year. Back then, I was always asked, “What got you into crypto?” I would give a pretty similar answer every time. Partly, the angle that was maybe not unique for me but differentiated me from most crypto people at the time was that I was a little bit further along the career path on the finance side. There weren't that many 30-plus-year-old finance people entering crypto prior to March 2017.

That investment angle really drew me in from the angle of an inefficient market as a trader. I didn't hear many people talking about it. When I was raising for BlockTower originally, my pitch to a lot of VCs and investors was, “You don't need to believe in crypto. You also don't need to believe that I'm the best trader or hedge fund manager in the world. This is the most inefficient market in the world.” The really big players—the people who are way better investors than me, more experienced, smarter—they're not in it, and they won't be in it for years because of the regulatory hurdles.

Basically, we got to be a big fish in a small pond as that pond was growing. The inefficient market was a big angle. I did believe in the humanitarian angle very much, as the grandson of Holocaust survivors and someone who cares about people fleeing oppressive regimes globally. The angle of crypto as privacy tech was meaningful to me, in that people like refugees could escape an oppressive regime with just a passphrase in their head and relocate to another country without having to be in a refugee camp and giving up everything.

The last was the technology, which I found very interesting. I'm a nontechnical person, but I really enjoy rabbit-holing down the cryptography and the engineering. So, that was what got me into it. The first few years were really exciting. It felt like being in the garage with Steve Jobs and Bill Gates and these brilliant entrepreneurs and founders.

Then I think my experience was not that dissimilar from most of the people watching, which was a sense of disillusionment over time, a product of a bunch of things. Partly, so much capital came into the industry, and very little of that was productively spent, right? You had projects raising $50 million, $100 million, or $1 billion, and only a few percentage points of that being spent on quality engineering or quality cryptography.

I wasn't really disappointed by the game-theory experiments. I expected those to be hard. For example, building social consensus around a DeFi protocol, where you're trying to use more interesting game-theory alignment mechanisms and utility-token mechanisms, is hard. I actually just saw a tweet from Erik Voorhees about his project with a decentralized LLM and a utility token, and people were talking about maybe he's found a smart, aligned utility token.

At least back in our day, when we were investing, they all failed, right? The utility-token model—it, and my take on that then and now, is that it's just really hard to make it work. It's not that it's a bad idea. It's not that it's impossible. It's that it's going to take a lot of iterating toward success around things like governance, DeFi theory, and all of that.

I think my disillusionment came more from not so much the small missteps or hacks, but the aggregate lack of forward progress, which I think is largely a function of incentives—misaligned incentives in people. What I mean by that is crypto rarely had the true best and brightest. I'm insulting myself in saying that as well.

That made sense in the very beginning. Something I said constantly was, “The very best investors are not in crypto. That's why I can win. That's why BlockTower can win.” By the time they're in crypto, because crypto is big enough and institutional enough, if we're going to compete, we're going to have to be competing with the best. But we've got some years to build up that capability and skill. Maybe I'm hiring those people. Maybe I'm poaching people from the sell-side, or whatever.

That didn't end up really happening. From my perspective, the quality of people in crypto didn't really level up in aggregate. Of course, there are some great people and some brilliant people, but in aggregate—and I think that was in large part due to structural reasons—the quality didn't level up.

The ambiguous regulation under the Biden administration and prior administrations almost guaranteed that criminals would lead the industry. If you don't enforce against the worst criminals and you do enforce against people who are mostly playing by the rules—they're registered, they're regulated, they're filing—what happens? You basically guarantee the industry leaders will be Binance over Coinbase, let's say.

Coinbase is kind of the one exception because they played the institutional game and lobbied very aggressively. But for the most part, the industry was led by the people willing to act least ethically, least legally, and willing to take the most risk, and that pushed out the best people. The people who didn't want to take jail risks or steal from people, a lot of them went into AI or stayed in TradFi.

I think AI has soaked up a lot of the absolute talent. You look at some of the people leading AI efforts, like DeepSeek. It's people like—I forget the name of the DeepSeek founder—but he was being compared to Terence Tao as this super math genius when he was 10 years old. He was winning math Olympiads. You look at someone like that. Those are the people leading AI.

Not everyone leading AI, but just these super geniuses. We don't have that many of them in crypto. Those that do exist are usually relegated to engineering positions, and a lot of their work is ignored or politicized, or they tend not to really be leading projects in a meaningful, productive way.

So, let me take a pause there. I could expound on this a lot more, but—

No, I think it's good. There's one thing that you hit on in the beginning that I want to dive into because I think it's really important: the concept of table selection and why crypto. Maybe talk about what you saw. What was so inefficient about crypto at the time? Maybe some examples were, “Oh, wow, we can really take advantage of this.”

2. Why Crypto Was The Easiest Table In Finance

We could come in and take advantage of that, whether you thought people were just massively mispricing some of the future value of Bitcoin, some of the future value of these technologies, or whether it was just amateur hour and you could clean up. Maybe talk about your process of—

Sure.

How you decided that crypto was an easy table to play at.

This is bringing me back. I haven't thought too much about this in the last year. That's fun.

There were a lot. What first got me into Bitcoin was that I had a very high-conviction mental model, loosely based on George Soros's reflexivity and a model around the adoption of assets, bubbles, commodities, and currencies. Basically, I felt in 2016 like I had this almost crystal ball, from a trader's lens, around cycles of adoption.

People were learning about Bitcoin. I was watching people in endowments learn about Bitcoin, buy it for their own account, start exploring it, and think about buying it institutionally. I felt like I saw a really clear roadmap where, if nothing changed over the next 2 years, a lot of people were going to be buying Bitcoin. By “nothing changes,” I mean that the stuff already in progress continues being built, right?

People were building better exchanges. They were building better infrastructure. There was better regulation coming down the pipeline. So if nothing new happened, we were going to have 10 times the number of market participants who were potential Bitcoin buyers because they now had accessibility. They were socialized to it. They were educated. And if even a small fraction of those people bought, the price had to go up. That was the most basic thesis I had prior to 2017.

That thesis kind of ended by the end of 2021, certainly by 2023. By the time Trump was elected to a second term—the Bitcoin president who keynoted a Bitcoin conference and talked about a Bitcoin strategic reserve—it wasn't early. It might be early in terms of global adoption, but it's certainly not early in a brand-recognition sense, right?

Everyone in the world has heard about Bitcoin. Most of the potential buyers have either bought it or tried to buy it, or you even look at countries like El Salvador adopting and then largely abandoning Bitcoin. So the thesis went from, “Wow, there are all these people who might want to use this thing and own it. They just can't yet, or they haven't heard of it yet, or they haven't tried it yet,” to now, kind of, everyone who can use it or might want to buy it has either tried it, thought about it, or made a decision, for better or worse.

At this point, you need something new to happen for a bullish thesis. It's not just more people hearing about it, right? It's pretty accessible. You can buy it anywhere legally, right? That thesis was a secular tailwind. That's not to say that it won't continue going higher, but it's no longer free money, and it hasn't been for a while, in my view.

The other piece is market microstructure. Prior to 2017, you had 30% arbitrages between South Korea and the United States, for example, and that persisted into late 2017. You had a 30% arbitrage at one point. That existed largely because of regulation, but also because of a lack of arbitrage capital and money allocated to taking advantage of it.

There were so many examples where the same asset was priced differently on 2 exchanges. You had really large arbitrages and crazy exchange movements due to liquidations. I remember, in 2017, at one point I was click-trading on Coinbase. I think this would have been before you joined us, and there was a point where the Coinbase API went down completely, so the market makers couldn't make markets. The website was updating once every 3 seconds, so it was almost like a stalled-out website.

Then it would update, and you would see a price and a massive amount of activity that had gone through. For about an hour, Bitcoin was swinging back and forth by 20% on Coinbase. I think I traded it 5 times. I don't remember quite where the price was, but it was something like: sell at 15K, buy at 11K; sell at 15K, buy at 11K—all in an hour.

Why did that opportunity exist? Why was I able to flip Bitcoin for 20% a few times? The answer is because you didn't know if you were getting filled. No one could trade electronically at all. It was terrifying, right? I was placing an order on Coinbase without knowing where the market really was, without knowing if I was getting filled. There was a lot of guesswork that not that many people were comfortable doing, and most institutional traders would have been uncomfortable with that environment.

There are a million examples like that, and almost all of that stuff diminished over time, as you'd expect: more professional market makers, more professional traders, and less dumb money. DeFi is a great case study of this. When DeFi summer was first taking off, you had incredible high real yields. Real meaning, yes, there's risk, but you had a 100% headline yield. Maybe the risk-adjusted yield was still 70% a year, right?

The default rate on Uniswap or Aave wasn't more than 30% annualized. You fast-forward a couple of years later, and now everyone's piling into Luna at 12%, when I would argue that, at that point, its default was already somewhat foreseeable, or at least a very high risk. Basically, you had worsening risk and far lower return. That happened because you had more people chasing the same set of deals, the same set of opportunities. It gets crowded. The real yield is diminishing, and the demand for it is rising.

You fast-forward to today, and obviously there are people making good money. There are market makers and arbitrageurs, but it's increasingly the smartest, most professional, lowest-cost-of-capital, most scalable money. Guys like Jane Street continue to mostly crush it, I think, in crypto. But it's not easy to compete against Jane Street. They're very good at what they do. They're very professional. They have a low cost of capital, the lowest latency, the best algorithms, and extreme diversification, where they can take massive swings in any one name.

Basically, the market is starting to look more like a traditional market in terms of tight spreads and tight arbitrages. The exceptions usually have a good reason. Whenever I look, there's always a good reason. You can say, “Okay, there's an arbitrage on 2 scammy exchanges.” But that exists because the smart people don't want to give their capital to the scammers. They don't want to short on FTX, or they don't want to short on Binance, because we've seen what happened before and what can happen again.

So I think, at this point, the market inefficiencies and opportunities are just thinner. I think it's a similar approach with this kind of holistic, smart risk management necessary to capture those gains. Either you're an arbitrageur and a good quantitative risk manager, like Jane Street, or you're someone like yourself, Avi, who's applying a holistic, common-sense approach and asking, “How risky is this exchange really? How risky is this DeFi protocol?”

You're approaching it using more qualitative data to form a quantitative conclusion, versus a quant who can't answer that. A quant might say, “Uniswap's never been hacked, so what is the hack risk?” Well, it's not zero, but we don't have any data to go on. We can only put together a smart guesstimate based on qualitative data points.

So today, if someone said to me, “Hey, I'm looking to get into trading. I have a little bit of mentorship and experience. It's credible for me to be a professional trader. What should I trade, or where should I look? What table should I sit at?” I would still say crypto is a great place to look as a small trader with limited capital. It's still definitely a more inefficient market than most. It's just that the risk-reward ratio is probably never going to be as good again as it was when we were first getting in.

3. Crypto Is Over: The Best People Left For AI

Yeah, I think that's probably a big problem that I've run into a lot. When you look at crypto, and you go back to 2017 and 2018, it really was the most exciting thing on the block, right? There weren't that many other industries that were, A, that small and had such large potential for growth, and B, had a lot of excitement and energy pouring into them.

I think it was very easy for crypto to capture people's attention because they didn't really know where else to look. Then you fast-forward to today, and there's so much happening, right? You look at what's happening with the hyperscalers, what's happening with AI stocks and memory, and the rebirth of American infrastructure.

Everyone's pouring in. Caterpillar is going through the roof because they're selling power generators for these data centers. People are talking about Trump funding new uranium startups or new nuclear startups, and people are getting really excited about uranium.

It almost seems like there's more excitement today. From your perspective, is that true, or is that just recency bias? When you were getting into crypto in 2017, were you looking around—you were sitting in Chicago at the time, helping manage the endowment as a portfolio manager—and saying, “There's just nothing else. Crypto is the most exciting thing”?

Yeah, absolutely. That was part of it. At the Chicago endowment and other endowments, we were all actively talking about finance. Twelve years ago, it felt and looked pretty boring. We were all talking about where to look for higher yields, right?

Especially if you're an institutional investor and you're in these megafunds that are all shockingly similar, especially public-equity funds, it became a combination of choice, incentives, pressure, and, partly, the market.

So, we had a market for 20 years that was pretty boring in public equities. Dispersion was low; everything was going up together almost every day, and so it was a terrible period for active management. Active hedge funds really struggled to outperform after fees because, basically, if you could just buy and hold a basket and capture the index, it was pretty hard to outperform that when most things were just going up together, right? The fees and the negative tax effects pressured people like myself, as well as endowments and institutions, to look for alternative asset classes.

So they were looking at frontier markets, emerging markets, and alternatives. Investments in certain types of commodities became more and more attractive to pensions, for example. Historically, that had been viewed as off-limits because commodities aren't inherently cash-producing. But over time, it was like, “Okay, we've got to get more creative, so let's start investing in pipelines and just other commodity plays,” for example.

So I think that desperation for yield and the push to be more creative reflected the fact that basically no one had almost any good ideas. Frontier markets were kind of terrible, and everyone who invested in them lost. That was largely because of corruption and adverse local regulation in much of Africa, for example, as well as the small markets.

So, like, every crypto project that has tried to do banking in Latin America, or stablecoins in Latin America—or at least almost all of them—has done terribly. Partly, that's just because you have small markets and people without much disposable income. Even if you have 10 million daily clients who are impoverished people in Ecuador or Chile or whatever, that just doesn't support a company of that value, right? It's hard to build a VC-backed company on that.

Whereas today, we've seen what feels like limitless growth potential in giant assets that can soak up almost infinite capital. So a lot of endowments and pensions were pretty aggressive in being early direct investors in the data-center play. Frankly, I was a little surprised by that. I visited my alma mater, or my former workplace, the University of Chicago, 6 months ago. I was surprised at how early and aggressively they had been direct investors in data centers, on that same thesis of, “We've got to be a little bit more activist and creative to generate high yields.”

Crypto is certainly not the only game in town. I think people are also a little bit disillusioned because some of those pensions, endowments, and family offices explored direct investments in crypto 5 or 6 years ago. Most went badly, right? I mean, most altcoins are down. Most things that you bought did not do well. And the things that got shopped to the big institutions were mostly toward the top. They were things like Solana at the top; they were these big protocols at the top that were probably forever tops. I mean, that's objective—who knows? But the tops in most altcoins were made 4 years ago, and most—

That's how much the dollar goes down.

But, yeah. So basically, anyone who did any kind of direct investment program as an institution is probably down very heavily, and it's hard for them to make an argument that the next time will go differently. Not saying it won't, but it's just hard from that institutional perspective.

Yes, I think you raised a great point. There's so many interesting things happening in the world right now, and so much of it is AI. To me, and I think most investors feel this way, it's disrupting kind of everything, which puts everything in play. So every boring industry, every boring asset class, suddenly might be high-returning with the right thesis or the right angle, because we're in a world of almost total disruption.

Can we dive into that? This is something that I think a lot of people are wondering, especially now, because we've had such massive gains in the memory sector. We've had such massive gains from Nvidia, Google, and Facebook. You just said something that actually not a lot of people are looking at right now: You said that AI could have an impact in boring areas. I kind of want to get into that. Where do you think the opportunity is right now, then, or is—yeah.

My high-level take on this is that I kind of missed the first wave. So I basically didn't invest in AI, and it wasn't because I was skeptical. It was just because, 3 or 4 years ago, I was super focused on crypto, and I was just kind of like, “I'm sure you guys will make money. You guys will do great.” All of my time and mental energy was consumed with the mix of entrepreneurship and investing in this space.

Over the last 18 months, frankly, I've just kind of been too burnt out. I've been gradually catching up, learning, and networking. But my take has been that I want to position myself so AI is—it's world-defining, world-changing. You can't avoid it if you're an investor who's at all tech-oriented. So my take is not to ignore it indefinitely. I'm spending an hour or 2 hours a day learning, but it's to be prepared to make those investments for the next wave.

That's not—I'm not saying I'm not going to invest. I've made some small punts. But what I'm really focused on is kind of what I was thinking about with crypto in 2015 and 2016, which is: Let me get the crystal ball. Let me try to get a highly strategic, strong viewpoint that will carry me for a decade or 2 decades, and then I'll start pursuing individual threads.

I followed a lot of the rotations. This is an amazing time. I mean, we've seen some great active investors do incredible rotations, basically following the AI, optics, and memory trades. You mentioned Caterpillar making a mint selling power generation, so it's an incredible time to be an active manager.

I'm not a full-time investor right now. I don't want to half-ass it. My attitude, for my own money as well as everyone I'm giving advice to, is: Don't try to trade unless you're treating it like a full-time job. It's hard. Even if you're good at it, it's a full-time job. It's demanding.

4. Mythos Is A Military Project

With that said, I have very few meaningful, insightful short-term trading or investment views. With that said, I do have some high-level views that I think are worth sharing. One is that I've been—I don't know if bearish is the right word—but a little bit skeptical about the IP created by the AI generators, like Claude, Meta, and Google.

I was having a debate with a close friend who's pretty senior on the Meta ML team about this. We were debating it 48 hours before Claude accidentally posted a good chunk of its codebase publicly. This was about 3 months ago. They accidentally revealed a lot of weights and things. There's constant corporate espionage, both between the firms—Google's trying to poach top Meta people and vice versa—and then nation-state espionage.

Anything being produced by Claude, by Meta, or by Google, you have to assume is in the hands of Russia, China, and North Korea. I don't know specifically who has what, but that's a good starting assumption. So with that assumption, how do you value these companies? How do you value the IP? Where does the value accrue?

This is very similar to the early discussions we were having in 2016 and 2017 about crypto, that protocol thesis, right? It's, okay, there's going to be all this activity, this thing is going to change the world, but what do you actually want to own? Do you want to own the memory-chip makers, the AI generators? Do you want to own the companies that are going to utilize the AI?

Here's another angle that I think is important to recognize. If a firm like OpenAI or Anthropic develops a model that can, for example, beat the stock market tremendously, not only are its users probably never getting access to that model, so we as the public are never getting access to that model, but that's not going to be on a retail customer tier list, right? Enterprise-model users are also not getting access. Ken Griffin's not getting access to that. But neither are the shareholders, because if Sam Altman realizes what he has, why is OpenAI going to run that? Sam Altman is going to run it privately, or the engineer who discovered it is going to run it privately.

I don't mean to state that as a tautology. You could frame it as a risk, but I would argue it's actually a likelihood. So that pushes me away, potentially, from investing in the model creators like Anthropic.

Another angle, as we've seen, is the regulatory side. These are pseudo-nationalized companies. Do you think that's necessary in order to make an investment in Anthropic or OpenAI? Do you think it's necessary that they capture 100% of the value—

—of their model? Right. So it's like they could be a good investment, but maybe their engineers on the back end are the ones running it.

Both could be true. Could absolutely be true.

My concern is more—it's hard. I was trying to come up with, okay, well, what is OpenAI actually going to create that they can monetize for $1 trillion, right? I mean, we're talking gigantic numbers.

Right. I mean, these things are coming out at insane valuations.

I'm not saying these companies are not valuable or that they can't produce great cash flow. The question is more, is the market overpricing or underpricing them? And with something like SpaceX or OpenAI, there's a lot of optimism, right? I mean, SpaceX is valued at well over $1 trillion on, what was it, like $20 billion in revenue, right? Yeah.

Yeah, and that's not that insane because SpaceX does have an incredible set of prospects, and there's no question that revenue is going to be growing radically. So, yes, SpaceX is a valuable company. I don't disagree, but is it 100 billion valuable or 2 trillion valuable? I have no idea.

My impression, talking to other investors, was that they were not discounting this at all. They were not incorporating these factors as risks either: the nationalization of Anthropic, Anthropic not being allowed to sell or use models. I don't know how far this goes, but it could go pretty far. I have some AI friends who are prepared to have a Los Alamos-style lockdown. They expect the US government to fully control them within 3 years.

This is a hypothesis. I'm not sure I agree, but when you think about the most critical national defense projects the US has ever had, with the nuclear project, we literally put all of our nuclear scientists in the desert and spied on them. We wouldn't let them leave. They couldn't make phone calls, right? We treated it like a military project. AI is a military project.

At some point, the US government and other governments—I mean, other governments are already doing this, at least the Chinese government—the US government may start approaching it that way and say, “Oh, you're a top engineer in machine learning at Meta. You're only using a government-issued cell phone. You're only traveling with permission. You're not allowed to leave the country.” If the US doesn't do that, we'll be leapfrogged by China, because they can basically use corporate espionage to get to where we are and then go forward.

All of that affects these as investments. I'm making a concrete argument—just raising these variables that I haven't seen that many people talk about. So, yeah, I'm not saying they're bad investments; I just haven't seen people seriously talk about discounting these factors on things like the memory makers.

My concern there—and this explains why I put some thought into this and yet haven't pulled the trigger on many investments, or anything in size, for these reasons—is with things like Caterpillar, power generation, or memory chips. As an investor, Avi, something I've often talked about—a framing I often use—is the kind of extrapolations investors make throughout history that are incorrect.

Every tech bubble follows a similar pattern: investors correctly identify a world-changing technology, then extrapolate it forward. They extrapolate recent progress and the pace of that progress forward. And so they say, “Oh, this is going to change the world in 5 years.” They're always right directionally but wrong on the timeline. The 5 years is almost always 20. The reason they make that mistake is that it's a series of mistaken extrapolations that are pretty simple if you analyze them.

So, I worry that the buildup in data centers—and this is not something other people accept, for sure—is analogous to the overbuilding of fiber in the internet era in the mid-to-late 90s. It's not that demand for the internet didn't continue growing exponentially; it's that other bottlenecks appeared that had to be addressed first.

5. The AI Bubble Is The Fiber Boom All Over Again

With AI, what we found is that data centers were the bottleneck. We threw massive, unbelievable amounts of money at that and built a crazy amount of capacity; it was basically like a man-on-the-moon project. Building a trillion dollars of data centers in a couple of years is incredible. But now what we're realizing is that there are other bottlenecks.

Yes, we could get better AI if we threw another 5 trillion at compute, but everyone's realizing that's not the good-ROI investment at this point. What we're starting to see is, week after week, new AI papers being published that dramatically cut the hardware needs to achieve the same result.

So, if you think about progress in almost any multivariable area, you hit a bottleneck. The smartest people throw lots of money and brainpower at that. They work at it and work at it, and then there's some exponential breakthrough: we discover the Transformer; we discover, you know, a much better chip fab, whatever.

Then, often, we go too far in that direction because all the money, all the incentives, all the everything says that this is where it's at: build data centers or whatever. But then we realize, “Oh, if we do a little bit of work here, we get much bigger gains,” right? We've overbuilt physical capital; we're underweight human capital. Now we have all these data centers, but we don't have enough power, rare earth metals, silver, optics, or gas turbines. Those are now the limiting factors, right?

And so the challenge with investing in that is I think you either need to be really fast and on the ball, like the best active managers are today. I'm envious of them, and this is an amazing time to be a sharp investor of that type. But I think that's a full-time job that I'm reluctant to dabble in.

And the concern there is, okay, once you move on to the next bottleneck, it's very much a rotational type of trade. What happens once you move on to the next bottleneck? Obviously, in crypto, at BlockTower, we used to love trading those 1–3-month rotations.

I think they were, in some ways, the best part of crypto, because if you paid attention to the pulse, you could figure out where that capital would flow and park in it. You could basically say, “I don't know when this narrative is going to take hold, but I know in the next 6–8 weeks it's very likely. Even if it goes down 20% in the meantime, it's going to go up 5x at some point in those 6–8 weeks, and then we can make a trade out of it.”

But again, the issue with crypto is that it was all a trade, right? Almost everything—or everything—that we touched when we were at BlockTower, except for Bitcoin, was a trade. I don't think there's a single one, other than—I mean, you led an investment in Polymarket, which obviously has done extremely well. But I think if you look at the portfolio that we managed at BlockTower, almost everything we touched is now probably far below its all-time high.

Yeah, to be fair, we didn't know that at the time.

But I think there was a little bit of an implicit understanding between both of us that—for example, I actually remember this very clearly—in Q4 of 2021, we put on this huge NEAR and Fantom trade, and it delivered a lot of return to our portfolio. We were both like, “Okay, we know what this is, right?”

I think part of being a good trader is obviously understanding what made you your money: were you an incredible investor who could see the future, or were you able to navigate these types of cycles? At least for me, I think one of the reasons that we actually work together very well—and you guys will hear on this podcast as we keep going—is that Avi's very good at figuring out all of the different possible paths that could exist. I think he finds angles that other people can't find, and then you kind of just have to pick one to voice a trade. I think that dynamic worked out really well.

But I guess the whole point of this is to go to AI, where I'm of the opinion that these cycles are not necessarily trades. It's very possible that, 10 years from now, everything that touches AI is just higher, and that's what I'm trying to parse out, right? That's the difference I see with crypto.

So, I don't think I disagree. I fully agree there's more of a real secular tailwind with AI than with crypto. With that said, with every real birth of a new technology that was purely positive, like the personal computer or the internet, 95% of the startups still fail, and 95% of the public companies still go bankrupt.

You look at the leading internet stocks in 1999; very few survived 5 years. You look at the personal-computer brands of the early 60s; very few survived. Same with railroads in the 1850s. The same is true of basically every technology.

My rough mental model on this is—and a challenge with AI is that it enables disruption, including to itself—that the AI leaders today are less than 10 years old. OpenAI, Claude, and new firms being created seemingly out of nowhere are getting multibillion-dollar VC-backed valuations. Right now, capital's not an obstacle for anything AI. If you're someone who is very credible founding an AI startup, you get money thrown at you.

I think advances, because AI disrupts tech and allows leveraging of tech, mean that the AI leaders can just as easily be leapfrogged again as they leapfrogged and became new incumbents. So, I'm super bullish on AI. I'm not at all an AI skeptic, but I'm a skeptic of individual AI companies, and I'm just trying to retain that basic investment realism.

It looks very similar to discussions in crypto: just because a protocol has billions of dollars of activity on it a day doesn't mean the price goes up. It doesn't mean the protocol is valuable; it doesn't mean the tokens are valuable. Just because everyone is using it—I mean, OpenAI had their incredibly popular—what was it?—Sora, their video-creation product.

Everyone loved it. Amazing. They shut it down because it was massively money-losing, and they could not find a way to make it profitable. So, it's a great example: They built a working, great product, got the market, got the users, and yet it was massively money-losing.

Kind of like all the early delivery apps on the internet before.

Yeah.

No, no, it's true. It's true. It can be very smart and can be great—great for the world—but also can be bad for the company. I think we're going to see that again and again, and I think it's hard. So, how do you invest in this?

You can, very similarly to crypto—and actually, I think I use some similar mental models everywhere. You can either take, as you said, that kind of trading, rotational mentality, where you don't care about disruption in 3 years. You're really just looking at the current pipeline: Who's coming out in the next 6 months? Who has pricing power over the next 6 months?

Or you take the VC mindset and you say, “Okay, I'm going to make a bunch of bets. I'm going to have—I know that a bunch of these startups can fail.” Similarly, if you're approaching personal computers in the 1960s or the internet in 1995, how do you win knowing that 95% are going to fail, even in this incredibly ripe, fertile ground? Well, the answer is you have to be somewhat discerning and make good bets, and you make a lot of bets. If one of the bets in your basket is IBM, you win. If one of the bets in your social media basket is Facebook, eventually you win.

I think you can take that attitude with AI. The challenge there is—and this became the same challenge with crypto later on—when the assets are so expensive and highly valued, it's not easy constructing a passive basket. It's not clear what that means. This brings me back to late 2017, when investors would ask me, “Are you just doing a market-cap-weighted basket?” I'm like, “Well, that would mean I'm putting 20% of your money into IOTA, which is valued at 20% of crypto's market cap.”

Oh, wow. I have not heard that name in a long time.

Yeah, and it's like, no, this is garbage. It's going to go to zero. Why would I put 20% of your money in it? Basically, if too much of the market cap is heavily overpriced, then a somewhat passive or market-cap-weighted approach will produce a bad outcome, even if you still have a secular tailwind.

Even if all of crypto—if you bought that basket in 2017 and didn't touch it—we can easily imagine that, in another few years, maybe, just saying hypothetically, maybe Bitcoin is at $200,000, maybe Ethereum is at $6,000, and yet IOTA is still probably at zero, right? Or we're very, very close to zero. So, that initial basket maybe does okay, but the point is I don't think there's a free answer at this point.

I think AI is maybe where crypto was in 2021 or 2022, which is to say the easiest money is gone. The tailwind that was so strong that almost anything with an AI name gave you 100x—I think that's over. Now this is more like 2021 or 2022, which is to say we still have a strong tailwind. Prices in general are still going to go up. But I think some people will lose money on AI over the next 3 to 5 years.

Right. I'm curious to take it a step further—not just talk about the AI companies themselves. What is something that's undeniable—and I think you agree with this, and I agree with it—is that AI will disrupt certain industries and change certain things.

One place to start is obviously: What is AI good at? What are LLMs currently capable of doing, and what industries are potentially the most affected by that? Going downstream and trying to think to yourself, “Okay, does this mean I just need to be short Accenture for the next 20 years because it's going to be a great short? Or is that not going to be affected?”

6. Where AI Value Actually Accrues

If there are publicly traded law firms, are you going short, or is AI going to increase the caseload? That's actually an argument that people are making, just as a tangent, with law firms specifically: AI is enabling lower-ticket cases to come online, especially in the personal-injury world. And so the caseload is actually exploding there, not decreasing. It's kind of interesting.

There are all these different effects, and I'm curious whether you've tackled this or thought through it.

I've spent some time thinking about this, but I don't have really strong conclusions. I can riff with some high-level thoughts. There's a hedge-fund manager—I think it's Chris Han at Founders Fund—who's one of the top-performing managers of all time, and he's had an interesting thesis that's a bit of a tweak on Warren Buffett's value investing.

He invests in companies with locked-in distribution. Sometimes those are government monopolies or regulatory capture. Sometimes those are literal distribution pipelines that would be very expensive to replicate. He doubled down on that thesis in the era of AI, and he's been doing very well over the last few years.

His thesis is that an AI company can't easily disrupt a distribution company. Engineers kind of get this wrong, or the intuition isn't super clear. Why couldn't OpenAI be Coca-Cola? Why couldn't OpenAI be Walmart?

The answer is because it's actually really expensive and hard to replicate all that local distribution, because you're dealing with local regulation and ordinances, town by town, city by city. Now AI may help with that. I'm sure AI lawyers and all of that may help, but it's still—anything that touches human beings is the bottleneck for all things AI.

Anything that's purely digital moves at infinite speed, or near-infinite speed. Anything that involves humans goes at human scale. If you need a human to sign off on it, it doesn't matter how fast the AI is. Anything involving unions, government officials, regulation, bureaucracy, or lobbying—all of that is still a human-first endeavor, even if it's helped by AI.

Chris's thesis—which I'm just bringing up for discussion—is that he basically doubled down on some very boring, old-world companies, with the thesis that these companies—and I don't know his current portfolio, so I can't say whether he owns this now—will benefit from AI.

You take a company like Visa, which at face value should be the easiest company in the world to disrupt. You and I have seen a million pitch decks saying crypto is going to disrupt Visa. AI should disrupt Visa, and yet maybe not, because Visa can use stablecoins. Visa can use AI to drive its costs down dramatically while providing the same service.

Why is Visa going to get replaced? OpenAI could offer a Visa with half the credit-card payment-processing fees, but does that really matter to consumers? Does that really matter to merchants compared with the brand and the ubiquity of it being accepted everywhere around the world?

His argument is that by reducing costs for companies with locked-in brands, consumer bases, very high switching costs, and all of that, those companies are going to see falling costs due to AI and flat or growing revenue, which makes them great buys. So, that's just one kind of old-school Warren Buffett investing. Is there still a place for Warren Buffett in this world with that style of investing? Yeah, and Chris seems to be proving it.

I think that's an interesting thesis. Again, I think that's a full-time job: really diving in and understanding these companies and how they're going to use AI. I don't think that works without channel checking and really getting to know the companies, not just doing industry-level analysis.

Other than that, I have more fears than bullish conviction. What I mean by that is disruption just looks incredibly attractive almost everywhere. LLMs are amazing, and AI agents are getting better exponentially. One person—one human being—can now be a company.

If I was launching a hedge fund tomorrow, I don't think I would hire junior analysts. I think I would just use AI in their place, basically. BlockTower was very thinly staffed because I didn't like managing people, so we were always understaffed. At one point, we were managing almost $2 billion with a team of about 33 people across 5 different strategies and franchises.

But if I did it over again, I think I'd probably have a third of the headcount, because most junior-analyst work, I think, would basically be me and an LLM. Or I would hire a single analyst who was an AI power user, and he would do the work of 8 analysts in terms of producing research reports for me on any topic I want in an hour.

That would have been work I would have asked you or Blake to do, or work that I would have been doing myself, either for BlockTower or for a boss before. That's almost all pure LLM at this point. It requires a smart human to give it some basic instructions and basic formatting and set up the agent, but then you get a million stock-specific reports and a million daily reports on cryptocurrencies.

One passion of mine recently has been personal health, which may turn into a startup. I've been rabbit-holing on neuromodulation. LLMs are incredible for medical research. I can pour through 200 peer-reviewed medical studies now and create my own medical meta-study in 10 minutes, whereas that would have been probably 10 hours of hard work before.

Are you using that to drive outcomes in your actual behavior in life? What does that look like to you?

7. I Can Beat Most Doctors With An LLM

Yeah. I’m fairly early down this journey. Coming out of BlockTower, I was super burnt out, anxious, depressed—pretty miserable. It just felt like I had the body and mind of a 7-year-old.

Running a hedge fund in crypto can do that.

Yes. Yeah.

I think that’s probably actually very close to home for a lot of people.

Yeah. I certainly am not unique in this. I wanted to prioritize physical and mental health. It was rational, right? I have a well-earned respite from work and I have the resources. Why not?

That’s been a slow, laborious process. It’s been very hard getting good medical advice, even though I tried to see the best doctors I could find. U.S. doctors are rarely holistic in their approach. For example, I have some minor back issues: slight scoliosis and a herniated L5-S1. Not a single U.S. doctor ever mentioned my feet or hips in relation to my back issues.

I hired a medical quarterback who has a network in London. I went to London and met with a few spine specialists—4 different people in back rehab and osteopathy. They all diagnosed me differently, with different methodologies, but they all came to the exact same conclusion. Every one of them was focused on remediating hip imbalances as the cause of my back issues.

I gave that as an example, and that’s specific to me. The physical was one side of it, and that included diet and supplementation. My learnings on the physical side are not going to be interesting to anyone listening to this. It’s the same thing everyone hears on every podcast: sleep, exercise, and nutrition. The 80/20 is to get the basics right.

I’ve been dabbling with some slightly more nuanced things, but I think it’s very much 80/20 on that. Whether a supplement helps you is largely dependent on whether you yourself have a deficiency in it. Is selenium going to help you? Only if you have a selenium deficiency. Otherwise, no.

The advice that I would probably give everyone watching this is that there are some things you can do very cheaply that are equivalent to the best-in-the-world concierge health practices. For example, you can get a monthly blood test for $75 a month. For another $100, you can have someone come to your apartment to take your blood, or you can just go to any of the normal blood-test clinics.

Having that data monthly—or maybe it’s $150 for a comprehensive blood panel—is great data. You may notice, working with an LLM or a doctor, that you’re not in the red for anything, but you’re a little low on vitamin D. For me, I was a little low on vitamin D, vitamin B12, selenium, and boron. It wasn’t at the level where my GP said, “You’re fine,” but this medical quarterback said, “You’re a little low. Why don’t we just supplement? It’s harmless. It can’t hurt you, and we’ll do another blood test in a month.” Being a little low on those things reduces your energy level.

There are also all sorts of tests you can do at home pretty inexpensively: gut biome, DNA, and all of that. I do think it’s reasonable to use LLMs to analyze the results, because that’s the expensive part, and it’s so hard finding good doctors. I’ve seen a lot of good doctors.

LLMs beat me. Working with an LLM beats them for most analysis. The very best doctors absolutely outstrip anything the LLM can do, or anything I can do with an LLM, but it’s so hard to find those really good doctors.

So I think it’s a resource now. With that said, if you’re going to make a major medical decision, of course consult with a doctor and consult with an expert. I’ll caveat all that. But realistically, it’s not reasonable to tell someone, “Go do your own research.” That’s just bad advice. And the whole “talk to experts for everything, all the time”—none of us have the time or money to do that. I’m not calling up a doctor with every question every time, especially when I think you can get a lot done very cheaply today.

Where I’ve been more interested and have maybe more nuanced thoughts is on the neuromodulation side. I fell down a cognitive-augmentation rabbit hole. I was introduced in London to a psychiatrist for TMS treatment—transmagnetic stimulation, or am I messing up the acronym?

You lost me there.

I’m messing up the acronym. It’s magnetic stimulation to the brain, which triggers electric pulses. This is well studied; it’s been around for about 50 years. It’s FDA-approved for pregnant women and children, and it’s very safe. It’s primarily used to treat anxiety and depression. It’s also used for stroke victims and Parkinson’s, and it’s pretty widely used.

There are all sorts of different types of TMS. I tried it for anxiety and depression, and I felt like I got moderate benefits over a few days of sessions. The psychiatrist-neuroscientist I was working with became convinced that TMS could improve learning, and there aren’t many public studies on this. I dove into it. He believed that, using TMS with a few other things, he could increase his speed of learning by 5×.

In terms of retention of information?

Literally learning a foreign language 5 times faster, learning advanced math 5 times faster, or learning sports skills 5 times faster.

It was plausible to me that that might be true, and it might be underappreciated in the medical community. Both of those things could plausibly be true due to medical bureaucracy. If you want to do a medical experiment, the process is very rigid and narrow: the medical ethics application and all of that.

The company that makes the device I used didn’t bother getting FDA approval in a clinical setting because of the cost and bureaucracy. It’s only marketed as an aesthetic device. Even though it’s zapping your brain and being used for brain treatment, they only put it in clinics that do stuff like aesthetics.

Interesting. Okay.

They do that because they just don’t want to deal with getting it registered as a medical device. So it’s exclusively used off-label.

Right?

Right. They sell this device for purpose A, but almost everyone who uses it is using it off-label, which is not illegal, but it brings extra liability with it. It means you can’t market it for that purpose. There’s all sorts of obscure, capricious medical regulation, similar to what we dealt with in financial regulation.

I’ve been rabbit-holing on that and experimenting. I think this neuroscientist was overoptimistic. I don’t think you can get 5× learning benefits, but I do think there’s evidence you can get a 30% to 100% improvement.

Have you done it yet? Sorry.

Kind of. I’ve been doing TMS daily. I actually smuggled a TMS device into the—I say “smuggled.” I did it legally, but it involved tricking a lot of work.

Yeah, I was going to say maybe we have to cut that part out, then.

No, no, no.

Okay, we’re good. We’re good. We’re good.

It’s a tiny bit of a gray area, but no, you’re welcome to share this publicly. I’ve been doing TMS daily in my apartment, and I’ve been starting to experiment with it for learning.

There’s one combination that I think produced a meaningful learning enhancement: combining TMS with an antibiotic called DCS, which sounded like “descloer.” It’s a scary antibiotic that’s mostly used to treat tuberculosis when other antibiotics can’t. It’s scary because it’s heavily neurotoxic and heavily habituating. It’s not addictive, but your brain adapts to it very quickly.

In small doses, it has an incredibly powerful neuroplasticity effect. It gives you the brain of a 7-year-old. Neuroplasticity is your brain’s ease of rewiring and forming new connections. As we age, our brain becomes less and less neuroplastic, and it gets harder to learn.

DCS proved—and this is well studied—that in small doses, it increases your neuroplasticity by 10×, but only for a few hours, around 2 to 3 hours. You can only take it about once a week—

Or it’s quite harmful.

Okay. So with the TMS—

You’ve basically got to really make use of those 2 to 3 hours.

Yes.

You can’t be slacking off.

My TMS learning enhancements so far have been, I think, positive, but within the placebo range. I think it’s enhanced my learning by 30%, but that’s small enough that it’s hard to be sure, right? That’s like, okay, did I learn a week of guitar in 4 days, 5 days? I don’t know, right?

But with the DCS, I did 5 minutes of juggling and learned a new move that would normally take me an hour or 2.

I did 5 minutes of slacklining and got meaningfully better in 5 minutes.

This entire conversation makes me so bullish on the idea of these biotech companies. There's so much that's coming out right now in the research and in the literature, also aided by drug discovery using AI. Eli Lilly just came out with this cholesterol drug. We're kind of at the precipice of radical change for human health.

It's like there's going to be an overwhelming amount of drugs coming out, and that's, I think, where the danger probably comes in.

I know personally a ton of people who are on retatrutide that got it from who knows where, and hopefully it's okay.

Yeah, I've also been reading up on peptides a bit, and I actually did, a little reluctantly or just very skeptically, start taking them. I'm actually on the Wolverine stack now. Those are the only peptides I take, and I started taking them a couple of weeks ago. Those are the most popular ones. It's BPC-157 and TB-500, called the Wolverine stack because they promote healing and reduce inflammation.

Very few peptides have good evidence behind them. The Wolverine stack only has meaningful animal studies. There are no meaningful human studies. The human evidence is anecdotal and mixed. Plenty of people have been on the Wolverine stack and felt nothing.

Adverse effects are quite rare if you get them from a good source. I'm sourcing them through a very reputable supply chain, and my medical quarterback does site visits at factories and things. I've gotten that side taken care of, and I'm under close monitoring for adverse effects, that kind of thing.

The point, though, is I think the peptide industry is functioning very much like the supplement industry 20 years ago. Most of it is garbage. Most people will be hurt by it because if you walk into a GNC and buy 30 supplement bottles, most of them do nothing in a healthy person.

It's not that if you supplement with selenium, and you have the amount of selenium your body needs, the extra selenium does nothing. It just gets urinated out. For most people, the common line is that you're just producing expensive urine when you take all these supplements, which is accurate.

But it's worse than that because, as you noted, some percentage are tainted. If you're taking 30 supplements every day, you're getting some toxic heavy metals and some contaminants. There's some existential risk, especially with injections, where you can produce an infection. So, yeah, people shouldn't be rushing to take tons of peptides. They should be cautious about the supply chain.

With that said, I'm sure somewhere among the 100 peptides being mass-sold right now, at least a few definitely are good. We know that GLP-1s and GLP-3s are certainly effective, right? I don't know if we know all the cost-benefit analysis, but certainly many of these peptides are quite effective and powerful.

I'm not pitching the Wolverine stack, but we're heading into a future of biohacking, like the new age.

[Speaker?]

I do want to get your take on one thing, especially before we wrap up here, because we're hitting the hour mark. One thing that I think you've always been very thoughtful on is the concept of cultural change and the idea of how different generations react to new technologies and, basically, where the political winds might be heading over the next 15, 20, 25 years.

Right now, what we're seeing with dramatic change—I mean, we're seeing a huge concentration of capital among an even smaller group of people as people leverage technology to really set themselves apart from the pack. I think, as you noted, a motivated human today can suck up a lot of the value in a way that they just couldn't 20 years ago. It would quite literally be impossible. They would have to share that value with a substantial number of people, and now you don't.

8. Peak Chaos: The Fourth Turning & The Coming Reset

I'm curious how you think that plays out. You've talked about the book The Fourth Turning. Are we headed into a period of political unrest because of all this, or do you think we're good and actually heading into a period where we're just totally fine, like nothing's going to change?

Yeah. We've been talking about these models for 7 years now, and I think we're at or close to peak chaos. It does feel like things could get a lot worse. I don't know how bad things will get in terms of political chaos.

Compared to 10 years ago, things feel horrible in the sense that there are political assassinations, widespread distrust, and all that. But we get used to stuff. The world didn't fall apart. People's lives are pretty good. The average American has a pretty good life.

I could see things getting a lot worse before stabilizing. We might have another 5 years of things getting worse. What I mean by worse is less social cohesion and lower social trust.

The concept of high trust—I've talked about it forever—within a business, within BlockTower, within countries. I'm seeing that framing getting used more and more in the social discussions around immigration, around social and cultural values, and even around the rule of law: how that works, and the breakdown that happens when you have judges who are saying, “No, we've got to favor empathy over justice,” kind of stuff.

It looks to me like it gets worse before it gets better. The rising left in the US is terrifying to me. I called myself a progressive liberal until 4 years ago. My politics have been unchanged for 20 years now. I'm probably a moderate.

But the “Mandani” crew (likely Mamdani)—many of them openly say they're communist—two of the recently elected New York State Assembly members literally wrote that their goal is to destroy the United States. Literally, no exaggeration: those are their words. That's potentially the rising force on the left in America.

With the right, I don't know. I'm actually maybe more optimistic about the right, in that after Trump, I think Trump probably would try to retain power if he were younger and healthier. I don't think he'll try in the current context. So maybe it's Rubio, maybe it's Vance.

I could see the Republicans actually moving toward the center and moving away from identity politics, more focused on governing. I don't know if, from the investment perspective, both sides are socialists now, right?

It's funny: Trump ran while railing against socialism. He's been the most socialist president the US has had, at least since FDR. He literally had the US government take direct stakes in multiple companies. He shut down [unclear: “mythos”]. There's just a very long list of literal socialist actions. But the left is embracing that even more aggressively.

In much of the world, basically, the trend for the last 10 to 20 years has been rising populism on both the left and the right—the horseshoe, right? Both sides are identity politics. Basically, both sides are woke now. We just have woke identity politics from the left and the right.

That's horrible to me—horrible socially, culturally, economically, scientifically, horrible in every way. It looks like it's going to get a little worse. I don't know how much worse it can get. It's a scary thought, right? It can get a lot worse.

With that said, from a time perspective, I think we're closer to the end than the beginning. If you look at big themes, we probably have the blowup of American capitalism ahead of us sometime in the next 5 years. We probably have something that feels like a more holistic collapse of Western democracy.

But that's then the foundation, and then we get to enjoy 50 years of growth, or 50 years of secular progress and trend. If AI doesn't kill us all, who knows? Any long-term or cyclical-type prediction is kind of falsified by AI.

I do view AI as a singularity. It means history will not repeat. It means that any historical analogy we want to make, we need to be cautious about.

I think that's honestly—let's end it on optimism. Let's say we're headed for 50 years of peace and prosperity, and thanks to AI, we're going to get there.

This was a great conversation. Thank you, Ari, for coming on the podcast. It's always a good time talking to you. We'll need to catch up again in the future.

My pleasure, Avi. Thanks for having me.

Take care.