Jeff Yan 谈 Hyperliquid 让整个金融体系上链的计划

Nothing said on Empire is a recommendation to buy or sell any investments or products.

Good afternoon, everyone. I'm Shawndra Devins. I'm a research analyst at Blockworks, and over the past year, I've had the privilege of leading our team's coverage of Hyperliquid. I'm extremely grateful for the opportunity to talk with the one and only Jeff, co-founder and CEO of Hyperliquid Labs. Jeff, how are you doing today?

Great. Thanks for taking the time to do this interview, and I really appreciate all the research you put out. I know you've been super early to the ecosystem, writing about it, and I think a lot of the insights you've shared have either come true or helped people realize what's already true. I really appreciate the good work you do.

Thank you. Hopefully, we're going to use this time today to get into Hyperliquid, what's being built, and what's still left to build. But I did want to start by saying thank you for the incredible amount of work that you and the team are doing, and for the principles that you have. It's hard not to get inspired by it. You notice it when you interact with Hyperliquid, you notice it in Singapore, and I feel like, most importantly, you also notice it in the industry. Compared to 2 or 3 years ago, there's so much more optimism. It feels like protocols actually value their community and the token, and I think that is in no small part due to the example that you guys have set. If anything, you probably set the bar too high.

No, no. It's a team effort. When I say team, I mean everyone building on the protocol.

I wanted to start by saying thanks for that. On the topic of the amount of work you guys are doing, this year Hyperliquid has shipped HLP4, which is Outcomes, Portfolio Margin, which has been expanded into a native borrow-lend protocol, priority fees, and also HLP3 Star on test net. I'm curious how you'd broadly contextualize everything that you've built, and what the overarching strategy is for your decisions.

The overarching strategy, I guess, is just to build the tools that ultimately can house all of finance. You can say that all you want, but if someone wants to build something and they can't do it, then they're going to build elsewhere. The goal is really just to be useful, and I think that's probably the undercurrent of all of this.

It's about looking at what can be done today and what builders who want to build feel like the gaps are, and noticing the commonalities there. That's really good feedback for the overall direction the protocol should take. An important property you always want is that the primitives that are ultimately built aren't too numerous, so that the system becomes complicated and it's unclear how things work together.

It should be very surgically precise, but each primitive should also be broadly useful in many ways, and they should compose well with each other. Portfolio Margin is a good example of how a borrowing-and-lending primitive can compose with a trading primitive, like perps or spot. That kind of interaction is nice because it unlocks a lot for people to build, but it also feels more elegant. It's the right way—the way finance should work: fewer things, but they work together well.

One thing that I noticed is that there seems to be a lag between when you guys build something and when the community and public understand how important it's going to be. HLP3 comes to mind. Even on testnet, it was apparent that this would enable the creation of markets for any asset with a price feed, but it was only when TradeXYZ and Shoku did an amazing job with execution that people really started to understand that. Do you think there's anything similar in what you guys are building now, where it's going to take a while for people to really understand how important it's going to be?

It's always hard to predict how long it will take, and some things may never take off because maybe something about the world is not quite as it seems. But I think every one of these primitives has some crazy-large stories, similar to HLP3, yet to be written and waiting for someone to come and really execute on the implementation, which is the hard part.

HIP-4 is one example. People often ask why they can't trade options, because options are so important in the traditional financial world. I don't think the answer is that perps are better than options. They're actually very complementary and serve different purposes.

I think HLP4 is the primitive where option or option-like payout curves can be built and scaled to a point where they're useful for users. I think that would be the parallel here, and I hope that's another story that has yet to unfold.

On the topic of HLP3, that has clearly been one of the growth stories of Hyperliquid. Open interest in real-world assets is up from 300 million in January to 3.9 billion today, a 13x increase. Along with that, it seems like TradFi is slowly starting to understand the vision of Hyperliquid, especially when you think of things like 24/7 trading and pre-IPOs.

With that, the framing of Hyperliquid changes. When I started writing about Hyperliquid in 2024, the most natural comparison was to protocols like Aevo and dYdX. Over time, that shifted to Hyperliquid being compared against the major crypto exchanges, and now it almost seems like Hyperliquid is being compared to traditional giants like the CME and Nasdaq. I'm curious what you think of that framing, and whether it changes the way you think about strategy.

None of the framings are perfect. It is interesting, because I do think the protocol is something that's genuinely not being attempted elsewhere, at least when I look around. I don't actually see a perfect comparison, and I think that's a good sign, because it means we're not all working on something that's really going to happen regardless of whether we're the ones to make it happen.

It's cool that it genuinely feels like there's not a clear comparison. Like you said, the target keeps moving, and I think maybe that's a sign of people realizing that it's multifaceted. But I think it really points to the fact that it's a lot more akin to public infrastructure—something like the internet or Linux—where you don't need 10 different attempts at the same thing. You really just need 1 neutral system.

All the names that you mentioned are ultimately not competing, but rather could benefit from the technology if it fits their use case. Going forward, I think that's the most helpful and productive framing. We're not here, as a community, to displace anyone. We're just here to empower them and help them help their users.

It has taken a while for people to understand the vision. One thing that helps is the amount and quality of teams that are now building on Hyperliquid, including Kraken. Another thing that I wanted to ask about is HIP-3*, which is on testnet. For context, HIP-3* is like HIP-3 in that it allows deployers to list their own markets, but it also has additional permissions, such as being able to whitelist who's allowed to trade on the exchange.

Payward, the parent company of Kraken, has already publicly announced that, pending regulatory approval, they want to use that to offer Hyperliquid to US citizens. I'm curious if you could talk a little bit more about HIP-3* DEXs and your vision of how they would interact with the broader Hyperliquid ecosystem.

This goes back to the other question about primitives and which primitives should exist. It's not any different from the other things we've discussed. Builders want to build these really cool things, and it's not currently possible with the suite of permissionless primitives available. HIP-3* is an attempt at distilling all of that feedback into a few modifications.

Whenever possible, it's nice not to reinvent the wheel. HIP-3 broadly works and has many desirable properties that deployers may want. Ultimately, the transparency, the self-custodial nature, and the redundancy and resiliency of the system are all really positive properties, regardless of which specific regulatory regime you might be trying to operate under.

Taking that as a very solid foundation and then adding opt-in controls that a deployer may decide to use, which ultimately lets them deploy the markets they want to deploy, is the entire motivation behind it. I think it synergizes well with all of the existing on-chain functionality that the protocol already serves.

Yeah. I think you said once something along the lines of, “There’s no such thing as decentralized finance and traditional finance; there’s just finance.”

Mm-hmm.

And the system we have is just a fundamental improvement, and I think that shows with HIP-3, right? You’re just giving people and institutions the option to build on top of it. You’re not taking away anything; it’s purely additive.

Mm-hmm.

So, zooming out from just perps, Hyperliquid’s vision has always been to be the house of all finance. I think what you guys did differently from other L1s is that other L1s started very general, and then they tried to improve their performance down the line, whereas you guys started with a hyper-performant L1 that was hyper-specialized in perps, and then you built out these building blocks.

We were talking about how you added spot and outcomes. We also now have native borrow-lend. I’m curious how much is left to add in terms of the blocks, or is it more about just improving the performance of what already exists?

I think it’s both. I think there will always be a performance problem. It’s a good problem to have, but with every order-of-magnitude increase in usage, that’s obviously a huge unlock and something we should welcome with open arms. I think it leads to a new set of hard technical problems, so I don’t think that will ever stop.

But that’s the best problem to have, right? Any L1 wants the problem of there being way too much organic demand for the chain and block space. Can we solve that? I think it’s a really fun engineering problem. That’s why we’re always trying to work with the best engineers in the world. It’s probably the most interesting problem in the world if you’re interested in high-performance systems.

There’s nothing more satisfying than building something and realizing that it’s not good enough because instead of 1 million people using it, now there are 10 million people. That’s such a good feeling.

But in terms of the other part, the primitives themselves, I think the list should not be finite. If anything, there’s no reason for something that needs to accommodate the entire global financial system to at some point decide, “This is it. From now on, this is set in stone.” That’s just not a practical way for a protocol to evolve and accommodate its users and builders.

On the infrastructure side, one recent thing that you added that I think a lot of people are maybe misunderstanding or downplaying the importance of is priority fees. Those work in 2 ways, right? You have read priority fees, which allow traders to stream uncommitted order flow from the mempool, and you have write priority, which allows traders to attach HYPE bribes to their orders to get priority in the queue.

I think this does 2 things. One, it burns HYPE. Almost 180,000 HYPE has been burned from it. But more importantly, when you compare it with TradFi, the latency competition there is very capital-intensive. It can cost hundreds of millions of dollars just to be able to compete. So priority fees are a much fairer way to bring this competition to Hyperliquid.

I’m curious if you could talk a little bit more about those priority fees and what your thoughts on them are.

Yeah, that’s a good way to put it. I wouldn’t view it as bribes. I would view it similarly to how Ethereum had this shift—I don’t remember what year it was, and I don’t remember what the EIP was either. Previously, fees were basically paid to validators, and now the vast majority, I believe, of the priority fees in Ethereum are protocolized burns.

I think “bribe” was more how you would label the previous system, which I think is toxic and unhealthy. It’s analogous to the latency arms race in a more traditional setting.

I generally think the motivation behind priority fees is: Can we find a more efficient way for markets to clear so that the alpha associated with an incidental technological advantage does not have outsized importance relative to the actual unique alpha you’re bringing to the markets?

In some sense, it’s impossible to internalize unique alpha. Unless you have some radically crazy, new, innovative market structure, I think the fundamental point of markets is that people who have different alpha come in and transact. The output of the market is the wisdom of the crowd—the sum of the signals from all the participants.

But there’s a certain class of alpha, like the microwave towers or the transatlantic fiber cables, that let you see something that happened in Chicago and then place the trade in Tokyo, or whatever. That stuff is very obviously zero-sum, and in some sense negative-sum, because the people who specialize in placing those trades need to compete with each other to go faster and faster.

Even if you’re adding a single microsecond or whatever to the time in which a specific ticker updates its BBO, that’s not valuable. No matter what someone tells you, that’s just not valuable to the world.

I think priority fees are basically an egalitarian attempt to take these negative externalities and internalize them so that the system is more accessible to anyone. If you have alpha in TradFi and you do not invest the baseline minimum in latency, then you simply cannot express your alpha, and the market does not get your signal.

The vision for a system with a truly fully fleshed-out and evolved-to-its-final-state priority fee system is that anyone can come to the market with minimal infrastructure investment and trade on equal footing with everyone else.

Yeah. So you’re basically internalizing what otherwise leaks in TradFi to these capital-intensive processes.

Yeah. It leaks to some combination of the toxic participants. I don’t know—“toxic” is not a normative word here. It’s just generally what people call them. They’re picking other participants off.

In a super-equilibrium state, I think it’s the people laying the cables or building microwaves who ultimately collect all the HFT profits, which is absurd if you think about it. But that’s what that set of incentives ultimately converges upon.

Another very important part of Hyperliquid has been its neutrality. You guys are very focused on building the infrastructure, but also on creating these economic incentives for other builders to build on top. We’ve seen very successful builders like TradeXYZ and Phantom integrate it, so it’s clear you’re only focused on the infrastructure side.

Hypothetically, because you have such a deep understanding of market making in TradFi and also Hyperliquid, if you were a builder and you were not allowed to work on the infrastructure side anymore, what would you build?

I think it would depend on what hasn’t been built and what has potential. I think that just depends on what the other builders are doing. At this specific moment, I would maybe look into options, because I hear a lot that people want to trade them.

I think the perps, by and large, satisfy a lot of users, but there are genuinely users who see the perps, trade perps, and say, “That doesn’t satisfy all the views I want to express about markets.”

That wasn’t obvious to me a year ago. It wasn’t clear whether everyone who was trading options really would just prefer perps, and I think the answer is no. So I would look there.

I think HIP-4 is a very obvious way to build options protocols. You inherit the spot order books and portfolio margin, like you said. You also have the perps venue to hedge your options delta against if you’re on the market-making side.

So I think the question is: How do you take that form factor and make it something that’s broadly appealing? I don’t think anything options-like—a sort of convex-payout-style product—has really reached escape velocity in crypto, but it clearly has in TradFi. I would try to understand that gap and see how to close it.

And also give us feedback—or, in this scenario, give whoever’s building the infrastructure feedback on how HIP-4 can be augmented or extended to really fulfill this vision.

Yeah. I think when HP4 came out, it was viewed as basically bringing prediction markets to Hyperliquid, but I think it’s much more than that, right? It’s about the composability of having this within the same ecosystem, where you have perps and spot. That’s where the value unlocks, right? Combining it with things like this.

Another thing that I wanted to ask you about was privacy, because privacy is a narrative in the meta right now, and there's a lot of discussion about whether it's beneficial for blockchains to be fully public and transparent. There was a study that looked at Hyperliquid order flow. It looked at T-webs, comparing T-webs that were hidden with T-webs that were publicly telegraphed to the market, and it found, contrary to what a lot of people think, that the public T-webs had better execution because they allowed market makers time to absorb that flow. So I'm curious, broadly, what your thoughts are on that, and whether there's a need for a private layer to be built on top of Hyperliquid.

Yeah, my thoughts are a bit controversial here. I think I've expressed them in various tweet essays over the past few years. But I approach it more from a first-principles perspective. If you look back to before electronic trading, trading was done in pits, and people were yelling at each other and trying to get the best price. By and large, it was not first-in, first-out. It was, "The tallest person, the loudest person gets a better fill."

If you compare that to electronic markets, the core difference, other than the increase in throughput and fairness, is that everyone can see everything that's going on. You don't need to be tall to see everyone in the pit. You can just look at the feed. Anyone can subscribe to the feed and get the data.

I think it's not at all controversial to say that there's a causal relationship here: when everyone sees the order book, the order book itself becomes more liquid because everyone knows what's going on and can incorporate that into their algorithms. I don't see why that's a very clear win for transparency and openness in electronic systems. I don't see any reason, a priori, to believe that the line should be drawn there, or that the current setup of order books is the optimal level of transparency.

If you look at Hyperliquid, it's built with order books on-chain from day 1. This actually comes from the principle that order books should not be run by a privileged central intermediary, which, by and large, is still how order books are run even in crypto. There's just some fundamental philosophical thing there where someone who's operating an order book has a lot of power, and a truly global, neutral financial system should not concentrate that power in one hand.

A side effect of that, when it comes to execution quality, is that the order book is open. It's more open than TradFi because you can see not only the total size and perhaps the number of orders and the orders themselves, but you can actually see the addresses associated with each order.

TradFi is not actually fully aligned on exactly what should be shown. If you look at markets in Brazil, for example, it's a very interesting situation where, whether intentional or not, the order book feed also publishes which broker many of the orders are associated with. This detail in and of itself is a very coarse classification of orders, but it leads to a ton of downstream, interesting things you can look at on the Brazilian exchanges.

Hyperliquid is just kind of like that, but many, many steps further. Now you have one of probably the finest-grained things you could ask for, which is an L4 order book of literally all the on-chain addresses. I think if you just draw that trajectory, it should, by first principles, lead to more efficient markets.

It's cool that papers kind of see the same results empirically. I think in TradFi, people do know this, sort of. If you look at the levered ETFs, they need to rebalance every day to stay at their leverage ratio, 2x or 3x or whatever. If the price goes up, they need to buy more; if the price goes down, they need to sell. They largely choose to do this in the closing auction.

That's an example where, empirically, you get better execution if you just tell the world, "This is our specification for the contract, and we're going to execute in this way according to this formula." Then the market maker is going to say, "This flow is exactly what it's going to be every day, and I can account for that and provide liquidity competitively, knowing that this flow is coming from a non-toxic source."

I guess that's many answers. I gave both the empirical and the first-principles arguments for why transparent markets are ultimately better for the end user when it comes to executing large size.

Yeah. I think it's also about fairness, right? In traditional platforms or centralized exchanges, it's not private. That data exists; it's just a matter of who's getting it and who's privileged enough to receive it. Having it public is fair.

I've noticed in my research, comparing Hyperliquid to traditional exchanges, that it is just a world of difference. It's completely different with Hyperliquid: it's completely accessible, versus the amount of gating and payments needed in TradFi.

Another thing that I wanted to ask you about is the community. You always mention that Hyperliquid is more than just Hyperliquid Labs. The community is part of the broader Hyperliquid team, and I think a lot of us appreciate that because we consider ourselves part of the community. As a co-founder and CEO, do you have any feedback for us specifically—what we're doing well, but also whether there's anything we could be doing better to support Hyperliquid?

Yeah, it's hard to give super-general feedback because the community is so diverse. I generally think everyone's amazing, and I couldn't be more proud of what people are doing, so I don't have any broad, sweeping feedback. I don't think such feedback exists.

But I can say one thing I hope people continue to focus on: remembering the underlying values of why we're building what we're building. The point of the thing we're building is ultimately to help end users, and not to get caught up in whatever you may get caught up in, whether that's beating some competitor, proving a point, or things like that.

Ultimately, we should stay humble, and we should remember that just 3 years ago, Hyperliquid was nothing more than an idea and a small group of people working day and night to make it happen. We're very fortunate with the growth we've all had up to this point, but I don't think that means we should feel complacent or better than other protocols for whatever reason.

I'm not saying that people aren't doing this already. If there was something that might be helpful to say, I guess it would be that.

I'm curious how you guys measure success as the infrastructure layer. I think it's easy to say for the builders and the deployers because there are a lot of responsibilities on them, and oftentimes they have the role of deploying markets, especially with HIP-3 and HIP-4. I'm curious how you measure your success on the infrastructure layer. Are you looking at any metrics?

We do look at metrics, not as much as you guys do in your research, but we do look at them broadly. I personally measure success more qualitatively. Are there net-new, cool things that haven't ever been built before that are being built? If the answer is yes, and those things are cool and users are using them because they prefer to use them, then I think that's the biggest win possible.

The infrastructure exists to be used and built upon. There's nothing that brings me personally more joy and fulfillment than seeing these zero-to-one qualitative shifts, whether it's HIP-3, kind of like 24/7 markets, or all these other things. There have been many such examples in these short few years.

And I think the quality of builders who are choosing to integrate and work with Hyperliquid speaks to the quality of the underlying infrastructure, right?

Hope so.

All right. I have one more question. Something you've said is that building Hyperliquid is similar to playing Go, in that it's impossible to calculate all the outcomes and exactly where the path is going to be, but you can look forward maybe 1, 2, or 3 steps ahead with high conviction and then follow your intuition based on that. I'm curious what you think. What are the obvious next steps for Hyperliquid?

Mm-hmm. Well, I mentioned options; I think that's very obvious. You have spot, perps, and options. I feel like that's kind of the trifecta: probably assets, a levered linear vehicle, and then some sort of convex vehicle. I think those three cover the bases on trading.

Obviously, perps are what Hyperliquid is well known for, but I would hope that it grows to cover everything. I guess we didn’t actually talk about the spot side. I think that is very obvious to me as well: this is a huge unlock.

We talked about portfolio margin, but portfolio margin is by and large only as useful as the assets that are supported. If people in the world tend to hold a specific thing and then you can use that thing as collateral, it’s super useful. So you have to get the things on the chain, not just the prices.

It’s harder to bootstrap that kind of network effect because holding things in and of itself is not an economic activity, so it’s understandable that it takes more time. I think there are also just a lot more difficulties in actually tokenizing a real-world asset. You can do perps as a way of tokenizing the price of the real-world asset, which is, in some sense, a lot more elegant and simple. It’s more of a mathematical problem.

It makes sense that it takes time, but it also feels very obvious to me that, on the way to housing all finance, or being the infrastructure that various financial institutions all plug into, there has to be a high-quality asset-issuance layer.

Yeah. If you think about it, 1 year ago, HIP-3 barely existed. Now it’s become basically everything people talk about. I could definitely imagine that 1 year from now, spot and options on Hyperliquid will be dominating the conversation.

Fingers crossed. Yeah, knock on wood. That’s a tall order, but I know there are some really talented teams working on it. If any of you know teams that are interested in doing this stuff but haven’t thought about deploying on Hyperliquid, I think it’s a really valuable proposition.

If you’re issuing assets or trying to define options contracts—convex contracts—the only thing you really want is for people to use the product. That’s what building on Hyperliquid also brings: you tap into this existing, global, expansive network of market participants.

Awesome. I think that’s all the time we have. I just wanted to say thank you so much for being here. Thank you for what you’re building.

Yeah. Thank you for the thoughtful questions. I hope the discussion was insightful.

Hyperliquid.

Hyperliquid.