比特币上涨、DeFi风险与下一笔AI交易

Welcome back to the Hivemind podcast, the show where we share our unfiltered opinions on markets, crypto, AI, and the top happenings across the industry. Proud to announce quickly, this is the first episode exclusively produced by Deli Media. This is where the Deli Hive Mind podcast will live going forward. Just like Nvidia is not a car, this is not your average podcast. So strap in because we've got a ton to cover. Joining me as always, we've got Yan, co-founder and managing partner of Deli Ventures; Jose, co-founder and head of Deli Labs; Ceteris, our head of research; and Jason, our head of markets at Deli. I want to start off because we have a lot to go through, but I want to go around the horn real quick. How do you guys feel since we last recorded a couple of weeks ago? Do you feel better or worse about markets right now, Jason? Let's kick off with you.

Two weeks ago, I was really cautious. I had a lot of hedges on, and most of those hedges have been absolutely blasted. Looking at prices alone, it's hard not to feel better about markets relative to where we were 1 or 2 weeks ago.

I don't really think a lot of the underlying fundamentals have changed. There's still an energy shock, and we'll see how that flows through to everything. There are potentially inflationary pressures coming from it, but at the end of the day, markets in the short term are more positioning- and narrative-based.

Clearly, the administration is the GOAT influencer of markets. If you had asked me 2 weeks ago whether equities would be at all-time highs, I would have said no, and clearly we're here. I don't mind being wrong in this scenario.

I'm definitely feeling better than I was 2 weeks ago, but I don't know if too much of the underlying fundamentals have really changed. It's more a question of how long this rally can go for me at this point.

Bitcoin is around $80K right now—almost $79K before we started recording. The STRC and MicroStrategy bid is crazy. ETFs might be coming back a bit, so things definitely look more constructive.

It's always easier to be more bullish in the market after it's done well. I'm still constructive. I think it's pretty straightforward in the near term: as long as STRC continues, it's hard to be bearish.

It's not just the bid from that, but also the bid from everyone else who comes in because of that bid. I don't know what the amplification is—not to use the word they use—but basically, how much of an incremental bid is driven by every dollar of STRC that comes in. There's definitely some of that.

On the geopolitical side, unless you get some insane World War III escalation, I think it should be somewhat of a non-factor in the near term. You would have assumed oil would have gone quite a bit higher a few weeks ago if you were told that this would not be resolved.

The additional shipments from the U.S., Saudi finding another way, and all of that have definitely helped cover a decent chunk of the shortfall in markets on that front. I'm not going to sit here and pretend I'm going to be able to figure out the knock-on effects better than some pure macro guys, so I don't really have a strong opinion on those. I'm thinking more near term.

As long as STRC seems to be bid, there's a reflexivity to it. If STRC bids Bitcoin, Bitcoin goes up, and then they can have more ammo from Strategy issuance, which gives them more cash to support more STRC.

They have $19 billion in headline capacity, but that doesn't really matter. I think it's more about how much they can actually deploy. It's hard to say exactly because you need to have an opinion on where BTC goes, but another $5–10 billion seems pretty plausible without any changes here. As long as that's there, it makes sense to be bullish.

I don't think I have anything particularly value-add to say here. I'm still bullish, and I haven't changed my positioning at all since last time. I still have a hedge on, which is down, but I'm still mostly pretty bullish.

I think crypto has been so beaten down that it's an interesting place to bid here, especially in some of the sectors that have product-market fit. I also think private markets are interesting.

Private markets in crypto historically have been such that if you slapped a token on your project, you would instantly get a valuation premium. Back from 2020 through 2023 and 2024, every project migrated to crypto because it saw this valuation premium that it wanted to capture. Obviously, you got a lot of people doing crypto who really had no business doing that.

Nowadays, we're seeing the reverse, where things that would get a bigger premium in traditional markets are getting punished for being in crypto. I've seen a lot of agentic trading pitches. It's clear to a lot of people that agentic trading, and AI generally, is the biggest change to the trading user experience since mobile. I would argue it's a bigger change than mobile.

Being first to a UX shift in trading has historically been a good way to win. Robinhood and others like Binance, I guess, hit crypto first, as did Coinbase. I think a lot of people are realizing that with agentic trading.

All the TradFi VC products that I've seen are pretty terrible, actually. I think True North is one of the best products in the category, and they actually have users. We incubated True North, so I'm biased here, but the valuations you see on the TradFi side are insane for some of these narratives.

You're seeing it across the board, even with some energy plays and other things from crypto teams. I think it's a really interesting time to back smart crypto founders you know well who are going after things that are effectively global businesses, rather than just operating in the crypto sandbox. Even in the private markets, I think some interesting opportunities are coming up because of the valuation discrepancy.

I definitely want to dig into that more. How are you feeling right now?

I don't know. It's hard to be too bearish. Markets have looked good, and it seems like they've been forming a bottom, at least in crypto, for a while.

You still have strong pockets. A bunch of the AI stuff is doing well. Monad still trades well, and Venice has been trading really well. The USD.AI CHIP TGE today is trading at around $1 billion fully diluted.

You're seeing more and more of these good launches. We haven't had a good TGE in a while. A TGE around $500–600 million fully diluted is now up to $1 billion, so we'll see whether that keeps up.

1. The Aave/KelpDAO exploit explained

Outside of DeFi's number-one abuser continuing to wreak havoc, I think things have been pretty good overall. It is interesting that, with the amount of money North Korea has stolen recently, crypto prices have actually been doing pretty well.

One thing I definitely wanted to get into—and I feel like it's a question that floats around every 6 months or so, but now it's more relevant than ever—is whether the risk-reward in DeFi is worth it at this point, given all the exploits, especially the ones we've seen over the last 3 weeks. I'm sure you have some takes on it.

If you're going to use DeFi, you have to be one of the degens. You have to be someone who's looping a lot, because at least you're getting a pretty big yield. You've borrowed so much against the protocol that if something goes wrong, your net loss isn't terrible, and you're getting a high yield with all the looping to compensate yourself for it.

The pooled money market doesn't really make sense to me if you're someone who's just passively depositing assets in a DeFi protocol. I saw people talking about 15% yields on stables in some money markets in DeFi, and it's like, you need 5 years at that APY to double your money without an exploit. That's a really high yield.

I think the pooled money market is pretty rough. There are a lot of design considerations that are going to come into play moving forward. I saw someone tweet about something that Spark actually does: money markets will have a limit on deposits.

For instance, on Spark, you can deposit cbBTC and borrow against it. I think the cap is around $50 million of cbBTC that can be deposited into Spark, but there's also a supply cap by hour or by day.

What you end up seeing in these hacks—and it's what you saw in the Kelp DAO hack—is that the entity holding the bag at the end is the money market with the most liquidity. If you hack one of these assets and own $300 million of rsETH, the easiest way to get liquidity is to deposit it into a money market and borrow the stables out.

This makes it rough if you're a stablecoin depositor on one of these pooled markets, because this kind of thing can happen.

It's just a really poor risk-reward. You're also seeing a lot of the wrapped depositors and everything. So I think DeFi is going to move—you’re going to see more siloed lending, like isolated pools where there's 1 collateral asset and 1 borrow asset, to limit the damage. That way, you get a certain group of lenders that get rugged, but the overall protocol solvency is fine. This is kind of how Morpho works.

I think there are going to be a lot more restrictions on how much money can flow at 1 time, but it's going to take a while. I don't know if you want to show DeFi Llama's TVL. It's down over 40% since this, which is kind of crazy, right? If you just scroll down.

Yeah. I mean, it's an insane drop in less than a few days, right? I'm pretty sure it's still frozen. People still can't really use it right now.

I don't know.

What's interesting is the price. When you look at where we were 2 weeks ago, when we last recorded this, it's actually up 2%, even though it's had a 20% drawdown from its recent high after this.

Yeah. I mean, how do you guys think this changes? I don't know. Some people I've talked to—it just seems like there's obviously been a ton of outflows and a big loss in TVL. On the stablecoin side, it just seems like the risk isn't there. There's obviously a bunch more to DeFi beyond just earning stablecoin yield. How do you guys see this changing things?

Yeah, I mean, on the Aave side, I'm very surprised the price has held up because I assume eventually they'll probably secure some collateral against future revenue. They aren't that high of a revenue generator.

What does that mean? Secure collateral against—

They have a hole, right? So they need to fill it somehow.

They'll do the Bitfinex token thing—

—or just something programmatic.

The Safety Module, right?

Payout. Yeah, but there's not enough in there.

How much is in the Safety Module? They use the ETH or whatever is in there, right? And then it's up to 30% of the AAVE in there.

I'm pretty sure stkAAVE doesn't backstop anymore.

Yeah, they turned that off.

Oh, really? Interesting.

I missed that. I'm pretty sure they turned it off, but I know there's a hole. So you assume that there would be some kind of borrowing against future revenues, which is basically, considering the hole and considering their revenues, effectively a permanent haircut to revenue. The alternative is insolvency, so you take that every day of the week. But that kind of just haircuts—or, basically, if it's half the revenue, then you're doubling the multiple on it at current prices.

TVL has come down, so I was a little surprised. Obviously, I don't have a bag, but I hope they do well because it's just best for everyone. I am a bit surprised.

Yeah, on the yield stuff, I agree. I think people providing collateral to get liquidity on it are kind of yield-agnostic in that sense. They're just looking to unlock some liquidity.

Crypto Rules Everything Around Me.

Eventually, something goes wrong, right? And so it doesn't really end well. I think you're going to see a similar situation here where you're just going to separate out the canonical base assets from all the other wrapped forms that take on additional protocol-specific risk.

This is also why I never really understood when they would do these LST indices. That just made no sense. Let's take on all of these risks. You're basically just taking on a bunch of additional risk, and each one was gradually different. It wasn't equal and diversified, right? It's not like you're selling insurance. The model wasn't to sell insurance. It's like, “I want no issues,” and so you're basically guaranteeing some form of issue if you have this index of them, whereas it's just better to opt for the safest one.

2. Why DeFi lending models may need to change

I think looping certainly makes sense, especially in isolated pools. If you have a good understanding of both the collateral and the borrowed asset, and there aren't these one-off social-engineering attacks or anything like that that can create issues, I think looping is okay. It really just boils down to the assets.

I think cross-margin just doesn't really make sense. I was very much a cross-margin maxi. I just think that we found out that people don't really use it, because the main use case on-chain is trading, right? My idea with cross-margin initially was that people were going to use this to do cross-margin leveraged trading—borrow stables to ape whatever—and they were going to make that really easy.

But I remember I pulled the data a while ago, and most people on Aave are just borrowing against a single collateral type. They're borrowing 1 asset against 1 collateral—the majority of people. Clearly, there are some people who only use it for the cross-collateral thing, but if the majority of people are doing that, then you really open yourself up to a lot of risk with the cross-margin system in this way.

I think cross-margin ends up making sense for perps, because that's where the capital efficiency really gets the biggest benefit: you actually make use of it. I think most people who want to unlock capital efficiency out of their assets aren't putting them in Aave anymore. They're trading perps with them.

I think it made sense at the time, but the isolated lending products have proven to be the better design. You can literally take on the risk that you want to take on. You can lend to whatever assets you're willing to underwrite without having to accept the risk of all the others that are listed in the cross-margin pool.

I think Aave did an amazing job surviving this long without ever having a hack. The risk framework is really on point. They had some amazing people working on that. It's a shame to see this happen. It was one of the stalwarts of DeFi.

I think this transition also just makes investing in their tokens way less interesting because the revenue take-rate potential is so much lower. If you look at Morpho's take rate versus Aave's, it's like one-tenth. Morpho literally needs about 10 times the amount of assets as Aave.

Why is that? Is it just where the fee is set?

Yeah. It's also the reserve factor.

The reserve factor.

I don't know. Aave has just been able to charge a pretty big premium over the years.

I don't know why there would be that difference. There's definitely more of a network effect to having cross-collateral, because—

You have more efficient pricing.

That's the other thing, right? The reason why you want cross-margin is that, with isolated pools, it's generally harder to find the optimal combined best borrowing and lending rates. Whereas if you have the assets pooled together, there's more arbitrage happening across different pools, allowing the cross-collateral pool to offer the best yield.

Actually, in isolated pools, you can literally price what you think the risk is. The interest rate in Aave is the cumulative risk that you assign to all the assets. Someone might think that all the assets are okay, but rsETH or whatever is really risky.

Back then, if you just had BTC and ETH, when your only risks were price-related and not technical, I think it made sense. It was just Bitcoin, ETH—these assets where you're only thinking about price. I think that's what they need to go back to. Then you have all the risky restaking stuff in a separate bucket.

You were never only thinking about price. Even then, there was WBTC, right? They were still having long discussions on the forums about the multisig model for WBTC. There was always a three-tier risk process that you had. It wasn't just market-based.

I think the isolated model—and again, I was a cross-margin maxi for a while—the isolated model lets you price it better, because you can literally have the people who are willing to lend against a single asset price in the risk of that single asset. People who are willing to lend against rsETH can do that, and others might be willing to lend at a much lower rate just against ETH. It just gets priced more accurately.

I think you see that, and I also think the P2P model on Morpho actually ends up working pretty well. It's annoying as a borrower sometimes because it can spike up pretty quickly, but I think it ends up finding the optimal interest rate. It seems to be working.

I don't know.

Pricing is a combination of risk and utilization, right? And I think you have much more efficient utilization when you have the pooled model. So you don't have these spikes.

But you don't have to have—with Morpho, you don't have the utilization rate curve, as far as I remember. I haven't looked into these things in a while, but with Aave, you had this kinked utilization curve. With Morpho, you had this thing where it would basically just target an optimal utilization, but then keep moving up or down based on that. So the whole curve shifts, and I think it's actually a better model even in that way.

I don't know if Aave moved to that. Again, I haven't looked—I haven't done a refresh on the latest DeFi in a while. Do you know, Aave, if they have P2P as well?

Not 100% sure, but I will say that we have a deep dive coming out on Morpho versus Aave that we were going to release this week, and then it kind of blew up all the charts and everything. So I think we're waiting to see how this plays out a bit, but then—

We'll have a new—

Yeah.

There's been a lot going on with Aave recently, too, right? With all the partners leaving.

The kink curves never made sense as a long-term solution because you're literally just trying to centrally plan an interest-rate formula and use this kink to ensure utilization. Things can change so quickly in DeFi that you need something more dynamic. That was always a model that made way more sense to me.

Yeah.

To me.

I think, just on this topic, maybe we can move on to the hack in general. It is interesting how everybody's blaming everyone. It seems pretty messy. Basically, what happened is Kelp used this 2-of-2 multisig where Kelp signs and LayerZero signs, and then the LayerZero side got exploited. It seems like their internal RPCs were exploited or something.

So Kelp is blaming LayerZero because this is the default setup with LayerZero, and LayerZero's systems were the ones that got compromised. LayerZero is blaming Kelp because they said that Kelp picked a risky multisig model, even though it is the default for LayerZero. Obviously, now Aave is left holding the bag, but Aave also had subpar risk design. The ability to deposit this much Kelp rsETH was not a really good risk-design decision.

And if you really think about it, this is all EigenLayer's fault, right? If you really want to get to the root of it, it's all this restaking stuff that everybody got into. And it's like—

And for what?

And for what? It never ended up turning into anything. It was all these restaking tokens, just layers upon layers: you have ETH, then you have staked ETH, then you get EigenLayer-staked ETH, and then, oh no, I want the restaking Kelp version, right? It's for the points, the governance token, and all this. You just have layers upon layers of risk, and then you have the money markets underwriting these million different LRT protocols.

I also saw Doug Colkitt had a really good tweet about how this is the L2 roadmap's fault, right? Because you're now depositing—you have this asset on a million different chains. The security properties are all different. The other interesting thing here for Kelp is that the Kelp rsETH that was exploited was the stuff off mainnet. The mainnet Kelp rsETH is actually fully backed. So should you only penalize the bridges?

But then it's like, how does this affect everything? There are so many dominoes here. It's so messy. I'm not really sure how it's going to play out. I'm sure there are going to be a bunch of lawsuits that come out of this, too. I'm sure there'll be lawsuits against LayerZero, against Kelp, and, I don't know, against Aave.

I'm not sure this is going to be clean, just like, “Oh, you're exploited, move on.”

The other thing is that Arbitrum ended up freezing $71 million, right? I thought that was the obvious thing to do in this case. Let's not pretend anymore: it's 2026. L2s are multisigs. Let's stop pretending there's any sort of real censorship resistance using these things. You have the ability to stop $71 million from going to North Korea's nuclear missile program. You should probably do that, right?

I think people trying to philosophize over this and be all cypherpunk and edgy are just wrong. You have bad principles if you think this is somehow the wrong thing for Arbitrum to do when it has the control to do it.

They're already taking the flak for being able to do it. You might as well exercise it when it matters.

Yeah. So, I don't know.

Damned if they do, damned if they don't.

It's just what this has all shown.

You're subtweeting Gabe right now, right?

Not necessarily Gabe. Gabe, yes, but in general, a lot of people. There are a lot of comments where people are like, “You start here, and then what comes next?” This isn't DeFi. It's like, well, yeah, it's not really DeFi. They have a security council. What do you say? They're able to do this, right? If you have the power to do this, you should probably do it.

Is this good or bad for Hyperliquid, right? Because all the Hyperliquid USDC is on Arbitrum, right? Mhm. That's—I mean—

Back in the day, I don't know. I think what Gabe's saying is, first of all, he's pointing out that a lot of DeFi is multisig DeFi, and that's not what we set out to do. But also, people don't really have consistent values about what they want from this stuff—whether it's meant to be fully decentralized, decentralized except North Korea, or meant to follow the law.

People hate on Circle as well for not freezing assets, right? Yeah, whatever. It's not a hill that I want to die on.

I think that's the right take. Back when I was in DeFi, when we were building protocols in DeFi, you literally couldn't—and probably because Gabe was our lawyer—you couldn't have multisigs on anything, right? Multisigs were a huge risk. In general, the OG crypto protocols, the ones that birthed DeFi, don't really have multisigs. They actually run fully decentralized, and that has its own set of trade-offs.

A lot of stuff was also avoided by multisigs, right? Because the other thing, if you have it fully decentralized—and this happened to a protocol we worked with—is that if you catch a bug in production, it's extremely scary. You have a 7-day governance process to make any change, and you don't want to make the fix in the binary of a governance proposal because a hacker can reverse-engineer it and hack you. So there are all sorts of game theory that has to go on there.

But definitely, most of the OG DeFi protocols don't actually rely on multisigs, like Curve, Yearn, and even Aave. It just became okay in the last few years to accept multisigs. L2s just never moved past stage 0. I remember back in 2021, there was a roadmap to be at stage 2 by 2024 or something like this. They haven't even moved past stage 0, most of them.

Then all the DeFi protocols just use multisigs or are centralized, like Hyperliquid. Hyperliquid would not have been okay back in 2020 or 2021—this kind of setup—but people just accept it. I think it's good to be reminded of the risks of that.

It's good to be reminded of the risks of relying on a multisig.

I agree with everything you're saying. I agree with what Gabe's saying. I understand Gabe's point. I don't think he's wrong, and I think Taz's tweet at him is a bit much.

Telling him to die in a fire was probably a bit over the top. But the biggest—the main problem here is how much of DeFi relies on bad multisig protocols that can be compromised, and protocols that have gaslit people about this stuff over the years, too.

People think this is the way to build things, and it's the practical way. It's practical. It's not maybe the best technical solution, but it's practical. It is kind of a joke how we've ended up here. And even—

Yeah, man. Hyperliquid—listen, I don't want to talk about Hyperliquid too much, but its bridge has always been the main concern. People don't want to talk about it, right? You're just considered FUD and all this stuff. I'm not saying Hyperliquid's going to get exploited.

Like, there have been a few now, and they can probably take a lot of lessons from this, right? They can make their systems better and everything. The reason people talk about Hyperliquid is because they have so many assets, right? It's not like people are specifically picking out Hyperliquid. It's just like, if their system is compromised, yeah, that'd be pretty damn bad for everything, right?

It'd be nice to have some rate limits on some of this—like, rate limits on their USD products. I think it makes a lot of sense. You want to move toward it, and it can still be—it's not antidecentralization, right? It can be fully permissionless; it can just run. Have a $10 million limit per hour or whatever. If you're a whale, it's somewhat inconvenient, but the downside it saves you from is so huge. It just makes sense, even on-chain.

Some protocols already have that. And yeah, that's the worst part about the Aave thing: it's nothing to do with their code. It's just poor risk management, and it's a pretty easy fix that could have avoided this.

How could you fix that on Aave, though?

You limit the deposit.

Yeah. You don't let them just deposit a bunch of trash collateral quickly, because you have limits. Every hour has a limit. You have your total supply cap, but then you have an hourly supply cap.

You'd have to have limits on borrowing; otherwise, they could just deposit slowly and then do that.

Yeah, exactly. You'd limit the damage because you'd have some time to address it in one way or another.

Mm-hmm.

And Drift, too, right? Drift's exploit was an admin key, yes, but they pumped collateral that they had just made to borrow against it, right? Why were they able to deposit this memecoin into a new market at a billion dollars of value when it was worth basically nothing? You shouldn't be able to deposit a billion dollars of collateral in one go and borrow against it. Clearly, if somebody is doing that, they are exploiting you. Nobody is organically doing that.

Money markets have always been kind of sketchy for me, but mostly because of the reentrancy attacks on EVM chains. I haven't used money markets on EVM chains for a long time because of that. But now the attacks aren't even really to do with the code. It's just, you know—

What DeFi exposure do you have left, Ceteris, if any?

The only DeFi exposure I have is holding some naked liquid-staking tokens.

And what did you have, say, 6 months ago?

I had a decent amount in Solana DeFi protocols for a while, like the RateX stuff and Project 0, which is like a prime broker. Why did I have it in there? Because I'm farming a make-believe token that might come one day, you know? You're kind of just getting rugged. I wouldn't even use it, really. I'm not someone who—

It was the most insane free roll of all time for many years, to be fair.

Yeah. But over the past—especially the MarginFi thing—it's like, dude, I had money in that protocol for how many years? I never got anything out of it, except I had some Drift collateral.

As soon as the Drift thing happened, I removed all my DeFi exposure everywhere. It was an admin-key thing, and I was like, you know what? I just can't keep doing this. I have the most assets on centralized exchanges I've ever had. What about you, Jose? I know you've generally played around with a lot of DeFi stuff over the years. Are you just not doing that, or are you still pretty active with some money?

I had a lot in DeFi for a lot of years, just farming Ethena and some of the points stuff, Pendle, and things like that. I still hold a bunch of USDe. I trust them. I think they really care about security and stuff like that.

But I definitely think the risk-reward has never really made a lot of sense on Aave. I guess it depends. Sometimes in bull markets, the rates get to a point where it does make sense. But almost always, it's because people are borrowing to do something that pays a higher rate. Why wouldn't I just do that and not take the whole money-market risk?

The last time rates were really high on Aave was because people were looping to do Ethena points farming, and people made a killing borrowing there. So why would I want to lend to them? If the rates are high, I'd rather be doing the thing that people are borrowing to do. If they're low, why would I take the risk?

I do think there are opportunities that come along in DeFi where, if you know a protocol that's coming out and you're really excited about the founder or things like that, some of the private farming opportunities are interesting. I just think you can't bucket DeFi into one category. You have to really look into the specifics of each protocol.

3. Where DeFi yield still makes sense

People used to do this back in 2020, 2021, and 2022. You would look at where the multisigs are, how many admin keys there are, and do a lot of risk analysis. But there haven't been big hacks in a while, or the ones there were weren't punished, right? Wormhole—Jump just made everyone whole. So people didn't have to think about this stuff.

I don't really use the perps, to be honest. Do you think this winds up being— I feel like in every one of these situations, it's bearish for certain protocols and bullish for others, right? You mentioned Ethena; that could be one. I think, as Ceteris mentioned earlier, USD.AI's CHIP token went live yesterday. The thing's absolutely ripping since it launched.

4. USDAI and real-world AI infrastructure yield

But you look at diversified yield opportunities, and it seems like USD.AI is launching at a really good time. People might be pulling TVL from other places, but they'd be interested in going into a diversified yield opportunity that's well above T-bills, into something like that. I don't know how you guys feel about it. I know some of you have gone deep on DeFi. How do you feel about the launch? Is it the right timing, just in general?

Yeah, I think it is. I mean, maybe Yan—you know it better, actually?

Decent. I know a decent amount.

Yeah. Okay, maybe you start then.

I'll bite.

Yeah, I mean, it makes a lot of sense, right? There aren't really asset-backed loans for these products in the market otherwise. It's actually solving a pain point with crypto that doesn't really exist elsewhere. Part of it is just that banks don't want to do it. The size isn't big enough, and so you have this kind of—it's not even a long tail anymore, right? But it started as a long tail.

Can you explain how it works, actually?

Yeah. Basically, you collateralize your hardware that sits in data centers, right? The data center is just the host, and then you can purchase the hardware. It sits in the data center, and you're allowed to borrow against it. When liquidations happen, they take it over and sell it.

It becomes a capital-efficient way for buyers of hardware to scale. Initially, the concern was that these assets were going to depreciate, but that clearly hasn't been the case. That's even more bullish for this as a form of collateral, because if they do depreciate, then you have to price in the risks of what happens with liquidations and all of that, and that results in higher borrowing costs.

But with how these things have been trending, it becomes a pretty comfortable thing for lenders to lend to. It basically is a capital-efficient way for people who are buying hardware to continue doing that by unlocking the liquidity of their existing hardware through these loans.

Yeah, it makes a lot of sense, actually. I think this in general is something we've seen, even with Tori and things like this. It's going to be a big theme: bringing real-world yield opportunities onto the open API of crypto rails and allowing for looping and integration into different protocols and stuff like that.

I think there's going to be a gold rush to do that over the next few years. Just figure out what some persistent sources of yield are that you can tokenize. It's unfortunate that they're all going to be branded as stablecoins, because they're more like funds.

Yeah.

Well, you need a stablecoin component to eventually get there.

Because you kind of play the long tail of, “Oh, if it gains adoption, then you can actually unlock quite a bit more yield,” because for every stablecoin that's not earning yield because it's used somewhere, that results in more yield for everyone else.

Yeah.

It's actually the best marketing ever for a fund, because you're basically saying, “I won't lose money,” right? That's what you're saying when you say you have a stablecoin.

You can always redeem this for a dollar, which is essentially saying, “I won’t lose money.” So it looks like free upside, which is always, I think, a worrying way to market things.

Well, you basically tranche it out. There’s the component that gains and loses, and then there’s the component that remains flat. If you have a loss that exceeds the tranche that’s taking on the risk, then the stable part takes a hit. But I think it varies.

You’ve got tranches in USD.AI?

No. USD.AI is different, right? What happens in a world where the data center gets bombed and the collateral is gone? That’s the risk that you’re taking on here. It’s not these things, unless there’s some technical breakthrough announced and something happens to chip prices where you don’t need the same level of compute anymore. They could get nuked, and you can’t sell fast enough.

But it seems like that happening on an acute basis is less likely. I think it’s more about gradual depreciation, and you can liquidate those assets over time. The vaults that are taking on risk are the trickier ones, where bucketing that as a stable is definitely a bit harder to do.

Even this is obviously way less risky. I think the business model is amazing, especially considering that these chips, like the H100s, are trading higher on secondary markets than they were when they came out 2 years ago, which is insane. Michael Burry was arguing that the real depreciation schedule is 2 years, or a year and 8 months or something, and it’s like, they’re literally more valuable now than they were at launch 2 years ago.

I think that’s going to keep being the case. I’m so bullish on the market for inference. I just think it’s going to be up only, so you’re going to need so many more chips and so many more data centers. But it’s not a stablecoin, right? There’s no data center that can be bombed where you lose money with a stablecoin.

It’s just that people have always branded these things as funds, and you try to understand the risks. Obviously, you hope that it won’t go down.

And they do have—

It’s going to do really well.

They have trusted partners, so it seems like they’ve tried to insulate or mitigate as much of that long-tail risk as possible.

For sure. Everyone’s doing this, right?

Right.

You can make the same argument for Ethena, so I’m not trying to have double standards here. Everyone is doing this in DeFi. It would be silly not to do that, but it is something to look out for.

In crypto, you’ve seen yields compress. They’re a direct proxy for appetite to go long. The question is, how do you combine the amount of capital that exists in crypto looking for a yield with these off-chain sources of yield?

In particular, the crypto benefit is the tokenization of that asset. The fact that you can now provide it as yield creates a lot of these opportunities, because it’s all basically a carry trade, right? This is a carry trade for people who own hardware to unlock more capital.

Looping sounds really bad, but I think it is a key unlock in crypto. It’s just a byproduct of unlocking collateral.

Why is it a key unlock in crypto?

I was just going to say it’s also how a lot of capital-intensive businesses operate. The concept of looping is basically borrowing against collateral—in this case, assets that you’re building out, or hardware you’re building out for a business to run. But go ahead, Yan.

Real estate developers have been looping forever, right? That’s literally what they do. That’s why you have these drawdowns; they get whacked because there’s not that much equity in there.

Yeah.

But I think it’s the ability to allow anything to be a form of collateral in crypto. Obviously, you take on some other risks, but I think that’s unique to crypto. You use USD.AI as an example: you didn’t have asset-backed loans being provided at scale for these products in a cheap and efficient way. That wouldn’t be possible without crypto.

There’s a Noya Protocol project of ours. We’ll be dropping a pod on that in the coming weeks. But that’s another example of carry trades and other exposures that you can tokenize and loop.

The idea is to borrow in a lower-yielding currency and deposit into a higher-yielding one. I think crypto, in particular, has enabled that. It really boils down to the quality of the underlying asset, right? You don’t want to loop garbage, because then it can implode pretty quickly.

But hardware like this, or carry trades on currencies, are some of the biggest markets in the world and are highly liquid. Those are the characteristics you want, and those are the ones that can scale and ideally bring yield back to crypto. You’re seeing Ethena diversify outside of pure carry trades on crypto, looking at carry trades on equities and other things. I think it’s bullish overall, just bringing capital into crypto.

One interesting thing with them right now, though, is that they’ve got $320 million in TVL, and $280 million of that is in PayPal’s PYUSD. There’s a lot of demand to have something like this, but it seems like the majority of the yield is still just in another stablecoin.

What you’re getting with this right now looks like mostly stablecoin exposure, but the APY boost is pretty real. PYUSD is yielding 4.5%, but their stablecoin is yielding 7%. So you’re getting an extra 2.5% just by having a lot of these 12% loans. It’s an interesting construction.

5. Venice, Bittensor, and AI token narratives

The trend here is important, right? If you looked a couple of months ago, it was predominantly PYUSD, and now they’re getting a lot of these contracts. You assume those are sticky customers who are going to keep coming back.

It seems like a ripe time for AI-related tokens in general. Venice, which I think we talked about last time, has continued to outperform and do well. This sector seems like a real bright spot in a lot of darkness within the broader crypto market.

There’s a TAO-sized hole that needs to get filled.

Do you say, “Go on Venice, and then we’ll go to TAO”?

Venice is basically a VPN for LLMs, right? You get access to the open-source models that they use, AIOZ, and other decentralized sources for compute. I think that’s where they make most of their margin.

You also get access to Opus and others, but that ends up being pretty low-margin. If you do the math on Sonnet, they have around a 20% margin, but on Opus it’s pretty much a pass-through. You assume they probably have some enterprise-scale API agreements with Anthropic, so they’re getting better rates than retail users do.

That’s probably loss-leading or maybe mildly break-even. I don’t know enough about what they’re paying on that front, but it’s a source of customer acquisition. Ultimately, you go and use the free models that are nearly as good, with complete privacy.

You keep seeing these hacks and other risks coming out, so the narrative is pretty strong. User growth has been really strong, and it’s a pure recurring-revenue business, so it’s hard to see a reason why people wouldn’t continue using it.

There are a lot of people, just digging around on Reddit, who use it for porn and things like that, which makes a lot of sense. It’s the ability to interact with these models without worrying about whether Anthropic or one of the others is training on your data, or about exploits, leaks, or anything like that.

The narrative is certainly strong with this one, and they’re putting everything behind the token. I have bags.

I think a lot of people have bags.

Do we think TAO is cooked? Shifting gears.

I wouldn’t say it’s cooked. It has a pretty big cult following, so there’s always going to be a lot of interest. There are a bunch of funds dedicated purely to TAO, so it’ll probably continue to see a lot of interest in it.

It’s never really been one I’ve been able to get behind. It’s chopped around for about 5 years as well, which may be better compared with other coins, honestly. But I think that subnet leaving was their number-one subnet, right? That subnet basically farmed TAO emissions for a while, made a lot of money, and then said, “The emissions are drying up.”

Let's go. What's the point of us being on this network? Let's just go do our own thing, right? And I think that's one of the challenges.

I don't know. It would be interesting to get a TAO debate. I think Kyle Samani was going to debate Jason, but I feel like Jason is not a good person to debate—not Jason on this pod, but Jason from the All-In pod. I don't know if anybody else is interested. It's just not interesting to me. Honestly, I don't find the model interesting. It's tough because—

I've never found it interesting, I think.

Yeah.

Yeah. It's like the cult thing, right? You can't discount it, but that on its own is just not a good enough thesis. I think cult is a great amplifier for something that's standalone good, but I don't think it can bandage over massive emissions and the headwinds of people evolving out of it when they're big enough, or just using it for free training, which you hear a bunch.

You have a lot of structural selling that you need to offset via the cult and via DCG products, or whatever else Barry Silbert is putting together to try and create sinks for it.

Silbert.

Silbert. Yeah. It produces a bunch of consumer surplus for projects and teams that use it, but it's just subsidized. The point about these teams eventually evolving when they get big enough and going on their own—I think there was a lot of debate about that, like protocols moving to their own chains, right?

We had a few years ago where everybody wanted their own chain or app chains. It's kind of a similar thesis: once you get big enough, you wind up moving away from the place where you started, or the place where you're adding value to. Yeah, the economics have never really made a lot of sense to me.

6. How AI agents are changing productivity

But elsewhere in AI land, Zay, I want to hit on you quickly because I know you watched the Jensen interview recently and wanted to have a little debate around it. I know you also did a bit of a mini AI boot camp last week, which had some good takeaways.

Not to go into too much detail, but were there any epiphany moments or things that you came out of that with a renewed understanding? You mentioned you're really bullish on inference—you're even more bullish on inference, right? What can be built with this stuff?

What were some of the takeaways for you after actually getting even deeper in the weeds and going hands-on?

Yeah, I think I had a couple of people in my life who were developers. I'd seen their setups, basically integrating Claude Code with Cursor or some kind of file-management system or context-management system.

I had been extensively using AI, but primarily just with projects—a bunch of different projects with instructions and context within each one. I think for me, the epiphany was seeing the power of a file-context-management system running with Claude Code.

I also tried to build this thing that I'm building, which is an AI chief of staff integrated with all your main applications—your email, Telegram, WhatsApp. It can read messages, draft them, knows everything about your life, can tell you your to-do list, that kind of thing.

I tried to build it a year and a half ago on Claude Code, and I couldn't really get past the first or second integration. Even integrating Telegram was such a pain in the ass. There were so many bugs, and now it's insanely easy.

You just tell it to do it, and you don't even have to fetch API keys for it. You can say, "Do it yourself," and it will open your browser, click to wherever the API key is, copy it, and paste it in itself if you give it permission to.

The speed at which you can do stuff is incredible, especially when it starts knowing you well. All of my tweets from now on, you can assume, have been co-written with AI. I haven't done one that was fully written by AI because it still can't do them quite right, but I uploaded my entire tweet history and this voice-AI skill DNA that eliminates some of the AI-isms.

I've been thinking throughout my whole life about what I can use to automate with this. The power of it is insane. Even follow-ups after calls— a lot of my job is doing introductions to people, and I hate it. I'm really bad at repetitive tasks like that, brainless tasks. I tend to procrastinate. It's the only thing I really procrastinate on, and it costs me twice the amount of time.

Now, after a call, the transcript gets uploaded. It knows the 2 follow-ups I have to do, queues the emails, and I just click send. This kind of stuff just feels magical. I've never been so excited to build things as I am right now, and I have a whole idea of what I want to build out.

I recommend everyone find someone in their life who's really good at this. For me, it was this guy, Sever Deutschman, who lives here in Lisbon and did ours. He's a really good teacher. I highly recommend it.

It also made me even more bearish on software, honestly, in the sense that I think if you want software, everyone's going to have their own CRM. I don't know exactly how it works at the enterprise level with permissions and stuff. I think there's still going to be a role for software there.

But in terms of personal software, if you want something, you can just ask for it and it will be built, tailored exactly to what you want. For smart generalists and entrepreneurs, this is a panacea. It's your superpower to such a huge extent.

Yeah. I will say I do find myself spending a lot more time almost building with agents and having my agents build things for me versus doing some of the things that I've historically really enjoyed.

Long-form reading is a good example, at least for me personally, and even writing. I've used AI and some of these agents to build out templates for me to produce more content, let's say. But I still haven't been able to crack the code of having it sound like me.

I also think that I learn by writing, and so I've done less of that because a lot of my writing is going toward agent prompts and trying to build this stuff out. So, I don't know.

I've got to send you this skill. I'm the opposite. I feel like I can spend way more time reading and writing now because I got a lot of time back from this. I'll send you this voice-DNA skill because it's—

It's getting there. It's getting there to where it can sound enough like you that you only need to make a few edits, but—

Your entire workflows have to change, and you have to change your approach to life. There are a lot of things that you couldn't do before, and your workflows were built around not being able to do them that you can do now.

Before, if I had an idea for writing something, a business, or a feature for a project we're incubating, I would sit down, put blinkers on, and just write it out and structure it in my head. It's quite a grueling, hard-thinking process.

Whereas now I just use Wispr Flow with AI, and it structures my thoughts for me. Not only does it structure them for me, it normally goes away and executes on whatever I had in mind in a really seamless way, then adds it to my context bank so I have it for later.

I think there are loads of examples of things like that. Even when I'm trying to teach people AI, they're like, "I got stuck here. I don't know what to do. It's bugged." I say, "Just ask it to fix it." You literally just say, "Fix it," and it goes and fixes it, right?

Whatever you're asking, you need to be asking AI. Any task you have, I think you need to be doing it with AI just to realize its capabilities and build it into your muscle memory.

Yeah, yeah. My system needs to get more optimized, for sure, but that's always a work in progress. I get so nerd-sniped by it that I spend so much time on it. I've built a bunch of things and created a bunch of things that I'm probably going to wind up not even ever using, but it just fascinates me.

No, I think that's super interesting, and AI is going to be a recurring topic for us. I love hearing those insights on how you guys are actually using this stuff.

7. What the team is watching next

The last thing I wanted to hit on is that we've covered a bunch, but there are always things that fall off the radar. Is there anything you guys are watching or any important things you've read? What's top of mind that we haven't talked about or touched on yet? Does anybody have anything they're looking at that we didn't hit on?

I'm looking at Bitcoin hitting $80K.

Midstream.

That's my—

That's the worst—

—that it's pumped on our stream.

I think the AI fallout is still pretty interesting to me. I'm following that pretty deeply. I'm still very interested in MetaDAO and everything. It hasn't been trading well recently and hasn't had the best launches, but I'm still bullish on them and still keeping up with them.

And then, yeah, my new Zcash thesis is just that people don't want to mess around with DeFi anymore, so just put it in the shielded pool and chill.

I don't know how true that will end up being, but I still think the Zcash chart looks pretty good. There's been pretty big distribution after the big pump last year. It seems like a key point for it right here, honestly.

Yeah, draw that line on that chart. Yeah, it's a great line.

Draw some squares. What was a good line?

Before we leave, what do you guys think of the Kalshi and Polymarket announcements going after perps? Do you think people are actually going to trade perps there?

Yeah, I think people will try. Are they going to make good perps, though? I don't know. I don't think it's super easy just to make perps, right? I think people who use Kalshi and Polymarket will try. Will Hyperliquid users go use them? No.

My thinking was that there's more reason for people on Hyperliquid to trade the prediction markets that they're going to launch with HIP-4. I think the overlap is bigger there. The perp traders on Hyperliquid will 100% trade the prediction markets that pop up there, but I don't think that's true for the majority of people trading on Polymarket and Kalshi. I just don't think the majority of people there want to trade perps and would go there to trade perps.

Until we start ripping and then everyone's just like, “All right, let's just go long.”

And it's super-reflexive.

But yeah, I'm just not sure.

I agree with that. But I think it's a lot easier to build this on centralized rails. It's just a backend, so the technical lift isn't nearly as high. It is just a customer-acquisition play, and if they have a bunch of users, then it makes sense.

One thing I'm wondering is, basically, if you see how successful STRC has been, when does Tom Lee do this for BitMine? It's actually a much more compelling pitch because you get yield on your ETH, right? So you can fund quite a bit. He's sitting on, I think, like $12 billion of ETH, right? You can say 3% on $12 billion is $360 million on a 10% preferred. So you could fully subsidize close to $4 billion of ETH purchases. And then there's obviously reflexivity there and all of that.

I feel like it's a very logical thing for him to do, with how clean the capital stack is there, and it should be pretty bullish for ETH if that can happen.

Yeah, I can see him. Yeah.

Doing it for sure. You should call it LEE. That'd be a good ticker.

The other funny thing I noticed for MicroStrategy is that they shifted their mNAV calculation to be enterprise value instead of market cap. So it's 1.2-something, but that's the preferred in the numerator, whereas if you did pure market cap, I think it's flat or maybe slightly down.

If you look on Strategy.com, they have it at 1.28 or something like that—1.27 right there.

1.27. Yeah, but they throw in enterprise value, which—I don't know if that gets around anything. I think that's maybe just for this dashboard and doesn't actually apply to the documentation and filings for the SEC.

But the STRC thing for BitMine just seems—if you have a yielding asset, the biggest concern right here is that you need BTC to go up so you can sell more Strategy shares to fund the dividend and keep the flywheel going. It's obviously the same dynamic that's helpful for BitMine, but you have a much bigger safety net in ETH yield.

And it's not even activity-driven; it's just straight-up.

One thing I wanted to pivot off this quickly, in line with the MetaDAO thing, is that I'm still interested in PUMP. I think this PumpCade thing recently is pretty cool. PumpCade was a project that launched a coin on Pump.fun, then they ended up doing a raise with Jump, which led the raise, and now they're doing this SAFE round, where if you have tokens, you can get some equity. This is something Gabe has been working on.

I think Pump.fun as a launchpad for non-memecoins and legitimate projects is something to keep an eye on. I've been pretty bullish on MetaDAO and everything for this vertical, and there's stuff like Zora and Meteora doing things, but I also feel like Pump.fun could make a dent here. They have the resources, right? They got a lot of money.

It doesn't seem like they're really focusing on streaming too much anymore, so we'll see. I still hold PUMP. I know people are worried there's a ton of unlocks coming, but we'll see. When are the unlocks, actually?

July, right? Like a year after the ICO.

Mm-hmm.

Yeah. Plenty of time.

Damn it.

8. Outro

Yeah, it might as well be years from now. How smart.

Agreed.

Awesome. Well, appreciate you guys joining again. This is the first episode with Deli Media, and this will be our new home here going forward. Subscribe here—you don't want to miss an episode—wherever you get your podcast, Apple Podcasts, Spotify, you know the drill. Thanks, guys. Excited for the next one.

Thanks for this. This was good. Thank you.