Web3 Galaxy Brain 🌌🧠

Nicholas: Welcome to Web3 Galaxy Brain. My name is Nicholas. Each week I sit down with some of the brightest people building Web3 to talk about what they're working on right now. My guest today is Andrew Huang, founder of Conduit. Conduit is a roll-up as a service provider, which lets anyone deploy their own roll-up in only a few clicks. Conduit operates Zora, Evo, Mode, Frame, Public Goods Network, and several other roll-ups, and each of these ships with a suite of helpful infrastructure like robust RPCs and block explorers. On this episode, Andrew and I discuss the ins and outs of one-click L2 and L3 roll-up deployments, the costs of running a superchain roll-up, and the different data availability options available today, including L1, Celestia, and AnyTrust. Andrew also shares how he gained the experience required to run a complex service like Conduit. It was great getting a chance to learn from Andrew. I hope you enjoy the show. This episode is sponsored by Dymo. Dymo is building a stablecoin personal bank. Their small team is shipping quickly and is looking to make their first engineering hire. Dymo prides itself on self-custody and free and open-source software. Their mission is to deliver liberatory cryptography to a global audience. If you're passionate about real-world Ethereum, reach out to founders at dymo.com. That's D-A-I-M-O dot com. And if you'd like to know more about the founders, DC and Nalin, you can search Dymo on web3galaxybrain.com and listen back to the fascinating episode we recorded a few months ago. My thanks to Dymo for sponsoring this episode. Email founders at dymo.com if you're an engineer interested in joining the team. As always, this show is provided as entertainment and does not constitute legal, financial or tax advice or any form of endorsement or suggestion. Crypto has risks and you alone are responsible for doing your research and making your own decisions. Andrew, thank you for coming on web3galaxybrain. I'm excited to talk about Conduit.

Andrew Huang: Yeah, thanks for having me. I'm excited to be here.

Nicholas: Yeah, absolutely. So we're going to talk all about Conduit and L2s in a minute. But first, before you founded Conduit, you were a backend engineer at a bunch of companies, Twitter, Pinterest, Wish, Quill, which is like a dashboard builder for startups. I'm curious, did the time spent doing backend dev at all those different places inform your decision to work on Conduit in any way?

Andrew Huang: Yeah, so one really reaching back into the direct history there. But yeah, so I mean, previously I was like infrastructure engineer primarily at this company called Wish. It was kind of like my first full-time job out of school. One of the, you know, the story I like to tell about them is they're kind of the crypto company. before, you know, we had a lot of these DeFi kind of token apps. They were so good at user acquisition, so really good at raising VC money, using like Facebook ads to get users, and then like continuing to raise off of that momentum, but never ended up building like a sustainable business model. And so if you look at them now, I think they're getting acquired in the public market. for kind of less than cash. So I'm kind of a, I feel like a nice, you know, example of like a lot of these kind of apps or protocols that just like launch tokens during DeFi summer. But, you know, when I was there, it was growing really, really fast. So one of the fastest growing companies and, you know, working to support that infrastructure was a great, you know, frankly, learning experience. I think we ran some of the largest database clusters in the world to support that kind of growth. So I spent about four years there and then ended up joining this company called Quill. So it might be hard to find on Quill now, but it wasn't a dashboard builder. It was a kind of creating a competitor to Slack. And so, you know, we're- Oh, okay.

Nicholas: Maybe it's a different company that's taking the name over.

Andrew Huang: Yeah. So we're kind of going through this conduit now where as we're scaling the company, Slack is starting to get super busy and like it's hard to find all the different kind of threads happening. And the idea behind Quill was like, if you just had more structured kind of conversations, things were threaded by default. And so maybe felt like chat, but a bit more like email, you could bring some like, you know, organization to that chaos. Ended up getting acquired by Twitter towards the end of 22, I believe. And then took six months off, joined, sorry, took six months off and then ended up joining Paradigm as an EIR. And it was kind of through that, that process, you know, conduit was born. And so, you know, I'd been following crypto since the Bitcoin days or the Bitcoin paper. And it was always really fascinating in terms of, you know, in computer science or kind of just engineering in general, you kind of solve distributed systems challenges. But the assumption is that, you know, you can kind of like trust, you know, most actors. Crypto is kind of the opposite, right? It's like you can't trust anybody and you need to validate. And verify everything. And so kind of the, in terms of just the intellectual challenges, it was kind of unintuitive to that background. And so I always found it super interesting. It was really not until I would say DeFi summer, you know, you started seeing Ethereum start to take off and these like Legos and kind of building blocks coming together, starting to form like interesting things. You could see where it kind of might be going that I started to dig in deeper. And, you know, towards the end of 2022, I think we're in some ways, exiting the bull market. FTX had had yet to happen, but that happened shortly after I joined Pear Diamonds in the AR. And I think there was a feeling like, you know, we had seen so much, you know, price increase and a lot of interest and a lot of money moving into this kind of the system. But it really felt like there was a limit to the actual utility that a lot of these apps had. And, you know, Uniswap, for example, that's like all amazing engineering at the end of the day, like facilitates swapping like two tokens between each other. And that's incredibly valuable in like all this stuff. But, you know, in terms of their ambitions of what they might want to build in the future, like I feel like apps are very heavily limited by the amount of compute and, you know, how cheaply they can actually access that compute. And so Ethereum letter one notorious, right, for how expensive things are, gas prices going up. you know, Solana and Alt L1 to try to, you know, compete with that. I think like the, you know, frankly, the what we see is kind of the, you know, the integration of these kind of kind of economic forces and kind of competitive factors is really kind of the role of centric roadmap where you have the ability, you know, to launch your own roll up, get dedicated compute, get sovereignty over that. It's a bit similar to the kind of Cosmos thesis. but you still interoperate with kind of the biggest crypto ecosystem in the world. um that is kind of ethereum and you're not discounting bitcoin it's it's not as programmable. um and so kind of building around ethereum and that ecosystem of assets of developer tooling of tech. um so being able to leverage all that work and then getting kind of your own custom uh kind of domain. on top of that um i think is the unlock to allow developers to actually build better applications and i think we're really seeing that today right. so just to rattle off a couple of conduit customers. evo is really a good example of um you know decentralized exchange built on their own layer 2 roll-up. and i saw a stat today they did like a billion in volume in a single day which is like kind of kind of crazy um zora network. another good example of you know i think kind of a dark horse for me initially but the kind of creator economy and like mints and art. um i think there's some stat where it's like they're responsible for like 70 percent of nft mints or kind of new uh you know nfts on the super chain which is like insane um. and so i think your own role is really allowing teams um apps companies to create their own dedicated block space that they can then differentiate in a meaningful way. and i think um that's allowing these kind of new um applications and kind of these new user flows that weren't really possible before.

Nicholas: yeah it's interesting to thinking about things like um exchanges or mints on their own chains. i mean in general all the conduit roll-ups so far are evm equivalent roll-ups. right. there's no deviation from evm equivalents in the roll-ups that you're providing right now i guess in terms of data availability.

Andrew Huang: so we do do custom roll-ups. there's a couple axes here that um i would say are different. i think data availability is certainly one um where um you know we obviously also support like any trust there will be like more of these in the future um on the execution client side um it ultimately depends on kind of the customer and what they want. um for example on the super chain config there is very minimal kind of customizations that are supported today and the idea is you know you want homogenous block space and everything like that. um we do have some customers that are on you know um custom execution clients. so a good example is like hyper. so they have some custom pre-compiles. we talk to other teams that want custom pre-compiles. on the nitro side you might support custom gas tokens. we actually have a pretty awesome customer that um customized the execution client in a pretty novel way um for nfts. so it is something that we support. um. it is kind of a case-by-case basis um and we see a range of customizations. i think the easiest side of that is probably on the execution client side. so adding a pre-compile for example is like relatively minimal overheads. we do see some more sophisticated ones that maybe try to change consensus or change larger things about the stack. and you know we certainly you know are interested in seeing if we can support that. i'd say the biggest um you know trade-off there is. you know when the stack gets upgrades um are you going to be force compatible with those? and um ultimately you know when you strike out so far kind of on your own um in some ways you kind of need to create to continue to to support and integrate those changes. so it's something on our mind in terms of like. there's this broad spectrum. i think for the most part most teams um are fine with just building you know on the standard kind of evm equivalent stuff maybe a pre-compile here or there but really differentiating i would say the app layer um. and then there are kind of teams that um you know for whatever reason may want more customizability. and i think on our side we're thinking through like the balance because if it is just evm equivalent we're going to have to in the same kind of configuration um we can clone that like really easily um and then for each kind of customization it's like a slightly it depends on kind of the customization um. but you know there's additional overhead there. and so as a platform for launching crypto compute and for managing and operating maintaining it upgrading it. um having that consistency across stacks makes that like 10x easier um and in a world of like thousands hundreds of thousands millions of um it feels to me like that's the only way that you know we're going to be able to solve a lot of the fragmentation issues.

Nicholas: so um yeah actually i'm surprised you're you'll entertain like uh changes to consensus and things that seems like really very painful to support long term.

Andrew Huang: yeah so we we definitely you know we see some interesting ideas. i think the question is how forge compatible. those are um and so each one is very case by case basis. i would say like that's in some ways like a service we're offering for free today is we understand the stacks at a really deep level because we work very closely with these teams. we get into the code ourselves. we've had to make changes to you know support these stacks in our infra in a you know highly scalable highly available way. um so we've quickly become like you know quite quite good at them. and um one thing we definitely help with is uh guiding the customization process and um you know for folks that want to do that. you know we um we need to make the decision on our side of. is that something that we want to support? um and you know either way we're happy to kind of provide our guidance on that.

Nicholas: yeah um i mean i'm very impressed not something i have any experience with this. the the complex task of running roll up back ends seems really uh challenging and high stakes because downtime is just well you know immediately noticed by users and product teams. how do you build a team to be able to do that effectively?

Andrew Huang: yeah so i basically think that this uh kind of type doesn't exist. i would say i'm probably the one person that tries to straddle both worlds. um and you know i would say the the task of condo is really like integrating these two different profiles. so i would say there's kind of the crypto native side of things which like you deeply understand roll up frameworks. you deeply understand what's happening under the hood um how to add new da layers how to you know the whole kind of modular blockchain kind of thesis um. and then there's the like the infra kind of sre or site reliability engineering side of things which is okay. you have these role frameworks. how do you run them in an hta? how do you roll out upgrades? do you have monitoring alerting what's the security um kind of posture um and marrying those two is um. you know infrastructure types tend to be kind of their mo's like. they're kind of divorced from the application a little bit. they're just like. i don't know any details about your thing. i'm just going to run it in kind of a you know secure way. um i think the challenge is that the crypto software where it is today you have to know the details in order to be able to run it in that way. and like a good example is um we wrote some proprietary software called conduit elector which enables you know high availability sequencing of the op stack um and that is kind of an intersection of this like infra sre kind of angle around. how do you run this in a you know traditional kind of web to distribute systems kind of way with the actual crypto software and it would have been impossible to do without you know either of those uh kind of uh knowledge areas um. and so i think in terms of our kind of core thesis and our core advantage i think it's bring together that kind of talent. so the intersection of the infrastructure with the crypto nativity um to attack this problem and like to date i think like the this working set has been zero and like we're trying to grow this talent base within conduit today.

Nicholas: how many people are working at conduit these days?

Andrew Huang: yeah so we're the number is changing all the time um. and so uh last week it was 12 i believe we're 14 as of yesterday um. and so definitely scaling growing out the team um yeah listen we're we're a small team we're. you know there's a lot of surface area and the market is moving quickly um and so we're looking to you know make sure we have the resources to be able to to match that um and to kind of continue that product velocity. and so you may have seen kind of the um self-serve mainnet deployments and we followed that up pretty quickly with the um l3 stuff um.

Nicholas: and so yeah maybe do you want to explain both for people who haven't heard yet?

Andrew Huang: for sure for sure. so um. traditionally you know uh conduit always allowed folks to spin up their own test nets. um you know self-serve pretty easy process. um mainnet to date has always been going through a sales call um and that's primarily because you know it's especially particularly early on right? it was such a you know new thing such a technical kind of challenge and and um you know strategy challenge that um we felt it was really important to have a hands-on approach. and um you know over the last year we've launched you know over 10 of these to mainnet and have guided our customers through that. and um you know. i think now that the narrative is there now that there's a lot of successful applications i think it's really de-risked that kind of. later too um launching or kind of layer layer three launching um. and so we felt that you know there are a lot of teams out there that probably wanted to launch um and didn't necessarily need to go through that sales call um and so we wanted to enable them to deploy permissionlessly um. and so there are a bunch of details there. but tldr um instead of needing to schedule a sales call anybody can actually deploy to mainnet um without talking to anyone through conduits platform and i we're excited to see.

Nicholas: is it fully automated now like it doesn't require any intervention on your side doesn't require any intervention on our side fully automated.

Andrew Huang: there are maybe some details around like uh what are coming soon around like you know dns like getting custom dns names which today is manual but we're launching self-serve soon um. but yeah functionally you get your rbc explorer transaction tracer kind of um you know high availability sequencing um at the click of a button and that starts you know settling to mainnet right away. so it's a really good. it's pretty cool.

Nicholas: it's like three thousand three thousand dollars a month plus deploy costs. uh if the numbers are held from last week um and yeah about 15 minutes to go from zero to mainnet launch it's very cool.

Andrew Huang: about 15 minutes and that's primarily just the contract. deploys. take about 15 minutes um to reliably go to mainnet. um. there are some kind of uh ongoing da costs um but for all da there you can kind of think of them as fixed costs and tend to be you know less than point five eath a month um. so very very reasonable um. and i think you know i'm excited for what that enables. where you know anybody in the same way that you can you know permissionlessly deploy a smart contract to. you know ethereum mainnet or like base or any of these later twos um you know in some ways making it even easier to deploy your own layer two or layer three um and so i think that aspect

Nicholas: where yeah so the layer three announcement i'm curious about also yeah

Andrew Huang: so it's been. it's existed on the arbitral morbid side for some time i think the you know uh particularly with base um and jesse and everything they're doing there they're very interested in and i think this makes sense for any l2. um is folks that want dedicated block space but want to tie into the distribution of the l2 ecosystem um. and so for base. i think there's a very clear case here of um. you know coinbase is kind of bridging uh off chain on chain um and there's a very compelling case for case for bringing more of these kind of quote-unquote retail users on chain for the first time and giving distribution to a lot of the apps built on top of base um. the challenge with just using base today is that you know eventually uh you know you're going to hit the gas limit and there's only so much that can be you know deployed there. um it also uses ethereum based da. so will kind of generally be more expensive than you know like an all da solution. having your own app chain or all three means that you can you know still settle to base and you can still be able to use it and so it's very easy to deposit in um. you withdraw out directly to base. you might get direct you know centralized exchange onboarding in the future um but you get to leverage. you know the cheap gas of all da as well as you know dedicated block space and so we're seeing this for a variety of applications whether it is you know games fully on chain games um you know various apps that pay gas for their users and view that as kind of an operating cost. so launching our chain enables you to recycle those fees um and then ultimately again like tying into the distribution of base. we also launched this for mode and zora which are two other chains that we run. um and you know. our thing is that you know l3 is tied to the distribution of the l2 ecosystem. that is you know unique to that um and then you have your own dedicated space to be able to build. you know the best apps that you can yeah.

Nicholas: so you mentioned in passing um the the costs related to running one of these roll-ups. so uh if i go look at the site there's the deployment which is a few ETH if you're deploying to or I guess no matter what it's a few ETH or does it depend where your uh settlement layer is? I suppose it'll be cheaper on on base.

Andrew Huang: yeah that that needs to be updated to reflect prices more accurately. um we initially set a pretty high hurdle because um you know we we didn't want to basically get dosed um and then have like a bunch of youth flow out of our system. um and so there that that's in some ways there just to uh you know rate limit it um. but yeah tldr it should be cheaper to actually deploy these on the later two because there are some like you know fixed costs associated with that. later choose a bit cheaper um which enable that um and so I would say fixed costs generally a bit cheaper for an all three which you know makes sense. um and I think you know ultimately there are the technological improvements but we frankly see a lot of narrative like a narrative reasons for why people want to launch all threes and I think the big one is really like um. you know for a lot of smaller apps or teams or game devs or experiments um there's a lot of hesitation in launching your own all two because it's like whoa. this thing is like you know I don't know if I can compete with the bases or the maintenance or the you know arbitrary ones of the world. um. and so by launching in all three it's kind of a little more insulated it's a little more uh I would say it's a little less risky um as perceived by by kind of the deployers. um and so we ultimately see that as bringing in the long tail that you know want their own dedicated block space. but we're too you know maybe too scared um to actually launch their own layer too. so we're very excited for for that trend.

Nicholas: and so the advantage of the owning your own block space in the case of an L3 or if you're ambitious enough and have the budget to do so as an L2 is you get to save on. if you're subsidizing transaction costs you save them because you are the sequencer. and there's this narrative advantage of launching your own chain. you know sort of in in the abstract of being your own chain is there and I guess you're. you're saying also that if they're a game or something with like a really high frequency of transactions anticipated, then it's even more important. Are those the main advantages?

Andrew Huang: Yeah, I would say the biggest one is the economic reasons or the recycling of the kind of transaction fees. We have a ton of applications that, for example, spend 3k a day or more on like Arbitrum Nova, which is like the cheapest, you know, one of the cheaper later chews out there. And that's a non-trivial expense, right? 100k a month, roughly. And so on your own chain, you actually get to recycle a majority of those fees back to yourself. And so if the question is like, do I spend 100k on this a month? And like that scales with volume. So like, where does that go if this continues to take off? Or is that cost effectively zero on my own chain? I think the answer is really clear. And I think that's why we're seeing a ton of inbound demand for this.

Nicholas: Right. So just circling back to those costs. So there's the deployment. if you're going to L1, less than that, if you're going to an L2 to deploy to L2. There's the monthly DA cost. So people are typically choosing between L1, Ethereum, Celestia, or AnyTrust if they're in the Arbitrum ecosystem. What are the DA costs roughly for, I guess it depends on how much use there is of your L3 or L2, or how can people think about the DA costs between, let's say, Celestia and AnyTrust?

Andrew Huang: So I would say today, they're roughly like. Celestia is so cheap that it's roughly equivalent. In theory, AnyTrust is like free. And, you know, Celestia has some costs. I would kind of think of it as like a fixed cost. So it depends on how frequently you want to post settlement to the settlement layer. And so, you know, the advantage of waiting longer is you get to batch up kind of more data into kind of a single data set. You know, hash that you post to layer one or layer two. And that drastically, you know, can reduce costs. Let's say you post once a day versus once an hour, that's a 24x decrease in kind of the fixed costs. And so ultimately, it depends on, you know, the trade-off there is, you know, finality takes longer, right? So finality is based on when you actually post that to the layer and then waiting, you know, X amount of time. For most of the applications that we see, cost is the biggest factor. And so, you know, delaying kind of reducing the frequency of posting. And then, you know, because the variable costs are so cheap with Alta, you know, you kind of have essentially fixed costs, but variable kind of sequencer revenue generation.

Nicholas: I see. I see. And is that the delay for this batching process? Is that directly correlated with the withdrawal time? I saw that option you have in the setup. Is that withdrawal time? that is your batching frequency?

Andrew Huang: So the withdrawal time is how long after, you know, proving that you have this withdrawal, how long you have to wait. The batch time and the proposal state roots impact, you know, after you initiate a withdrawal on layer two, how soon you can like prove it on layer one and start that withdrawal delay. So yeah, there's a lot of different components here. Ultimately, it kind of comes down to like for, you know, maximum security, excuse me, maximum security. But yeah, TLDR slower batches will delay withdrawals because it delays how soon you can prove a withdrawal against that state.

Nicholas: Right. And then there's this plan cost. So if you're deploying to an L2, it's $3,000. Is it the same for L3 deployments?

Andrew Huang: It's the same for L3 deployments. L3 deployments are actually, depending on the stack, are actually can be more expensive. Because of some, particularly for OP stack L3s, there are some configuration changes you need to make to actually have a stable chain. And that has, you know, that reflects in some of the hardware, additional hardware we have to run. But today, we're just keeping it at the same kind of flat pricing. And, you know, again, I think over time, the prices will go down. Crypto software traditionally hasn't been, you know, engineered in a way that makes it easy to of dedicated kind of software. I think we have plans to kind of, in some ways, like Web2ify, you know, the crypto software to be able to run it in, for example, a multi-tenant way or horizontally scaling way. Whereas today, you know, we run at least six or seven nodes per deployment for reliability, for HA, for backups, for things like, you know, syncing other nodes. And so it gets, you know, the costs start to add up. Over time, you know, our internal, you know, roadmap here on allowing multi-tenancy and kind of these other, I would say, like cloud style infrastructure improvements for the crypto software are going to drastically kind of bring down the prices there.

Nicholas: Yeah, I'm curious, how low do you think the prices can get, let's say, for deploying your own L3? What's the, I mean, will we live in a world where every application will have its own L3? Is this going to be a $15 a month kind of thing?

Andrew Huang: I mean, that's ultimately the goal. I think the, you know, the, I think there will always be the deployment prices, which are just deploying the contracts. And so that's not something that we are able to change. That will be like kind of core Ethereum improvements. But, you know, in terms of the actual software, you can imagine this getting down to like $50 a month or something, depending on the amount of resources you're using. And, you know, I think this is pretty ubiquitous across Web2, right? Where you can sign up, you can kind of pay as you go. And that's kind of the ultimate model that we want to get to. And, you know, we, I think like if by the end of the year, we have something that approaches this, I don't know if we'll get to $50 a month, but we'll be definitely be able to knock it down from $3,000 to something much cheaper.

Nicholas: And that would be probably a Celestia L3 or, or Antitrust, I guess you're saying is also nearly, or is free or should be around free?

Andrew Huang: Yeah. I mean, it should probably be all DA for those cheap price points. For sure. But, you know, fundamentally, a lot of the software, the same between all DA or not, it kind of comes down to yeah, basically the DA costs. And so for things that like, like Zora network, like mode network, which are Ethereum based, they tend to have much stricter uptime requirements. And so that just means we need to run kind of more hardware, better hardware, et cetera. So that's where the primary drivers of cost are. I think all DA, we see, you know, longer tail applications and like you're dealing less with value. So they tend to have, you know, it's, you know, not as stringent. but we'll, we'll continue to see what types of apps are built in, like what the requirements are and importantly, things like performance. So, you know, that's always a factor in terms of one cost, but two, you know, what that enables developers to, to create.

Nicholas: And where do you think I can DA will come in on price? I guess it's going to be more expensive than Celestia, but less expensive than L1.

Andrew Huang: That's a good question. I don't know what the pricing is going to come in at. Yeah. We're working with the team. We, you know, we think they're great. Celestia is great. I and he is great. We're working on integrating, we already integrate with Celestia obviously, but looking to integrate, I can do and other DA options as well. Yeah. Pricing, I think ultimately is, is DA a commodity? You know, I guess we'll find out, I think in theory it is. But if you look at the token price, I don't know, man, that seems like it's a winner. take all. So. Yeah, we'll certainly have to see, you know, how that plays out over time. I do think having some competition in the market is healthy, just so that customers have multiple options, but I'm excited to see how that plays out. And ultimately, I do think there is some difference in execution and, you know, we'll have to see just how much better these different teams actually get to get to market.

Nicholas: So ultimately, the business model then is not the fees that you're charging. But instead, the percentage. The percentage of the sequencer profit share, or how do you foresee the the business model for Conduit evolving?

Andrew Huang: Yeah, it's a part of the business model really depends on the chain. For example, for L3s, the fees are so cheap, it's unclear that that's going to be a big revenue stream. And so it's primarily, you know, it's probably like 95% kind of the SaaS and maybe 5% kind of fees.

Nicholas: Even when you go down to like $50 a month?

Andrew Huang: Even when you go down to $15 a month, you might still bill. Yeah. Yeah, we'll have to see what the because like today, for example, we don't charge for the RPC, but in the future, you know, we basically subsidize that today.

Nicholas: And it's quite expensive on your side, because you're running this redundant RPC tech.

Andrew Huang: It's more like the request volume and the amount of data that you're addressing and the compute that you're using. So for Azure network mode network, we run free public RPCs that are quite generous, and we kind of view that as a way to incentive compatible growth. And so we've run into cases where we've scaled up to like 2030 nodes serving all this RPC demand, and it's like, this is starting to get expensive. And so if you don't price something, you're going to get spammed, quote unquote. And so there, you know, we want to put some pricing there, I think, like, generally, my take on RPC providers today is like, they charge a lot for what they offer. And so that's why so many people run their own nodes. I think our goal is to offer the RPC at a competitive pricing such that, you know, you don't think about it as this huge trade off between cost and like operations. And, you know, you kind of just get a great service that you're happy, happy to use. So that's our goal there. But yeah, the blend of SaaS and like sequencer fee revenue today is I would say is around 50 50. And so it just really depends on the networks. It depends on, you know, yet their model. And, you know, I'm excited. I'm excited to explore in the future what those look like. For example, an all DAL3 might generate significant sequencer revenue if, you know, they have a lot of congestion, right? Because EIP 1559 will like force the price up. In that case, they might benefit from, you know, being their own later too in the future, for example. But ultimately, it will really depend on the type and style of the applications. And we're excited to see what folks kind of build.

Nicholas: In what cases do Optimism and Arbitrum charge a fee to roll-ups on top of them?

Andrew Huang: Yeah, so they charge fees to every app on top of them in terms of the built-in sequencer premium. And typically, you know, it roughly amounts to, I think, like 30% blended for both in terms of like the rough, you know, over the call data cost that they sent out one, what they charge on the later two. So roughly 30%.

Nicholas: And then there's also additional fees if you're in Superchain or the Arbitrum equivalent. There's fees associated with that.

Andrew Huang: Yeah, so launching your own Superchain chain is a 15% net fee to the Optimism Collective. You can, of course, launch your own fork that's not part of the Superchain and pay zero fees. On the Arbitrum side, 10% for your own later two through their Arbitrum Orbit Expansion Program. And L3. Sure, free, I believe, on Arbitrum 1 and Nova. And then, yeah, you might need another 10% for L3s on other chains. So that's the kind of rough cost there.

Nicholas: How do you see the advantages? or like, is there substantial enough advantage to being a part of Superchain or the Orbit Expansion Program for people to participate? Is it, what specifically do you get for being a part of those groups?

Andrew Huang: Yeah, so on Arbitrum. On the Arbitrum side, you can't launch a later two because it's BSL'd unless you kind of contribute that fee revenue back to the DAO. It makes sense, right? If you kind of treat this world framework that, you know, it's frankly a really great world framework and a lot of great tech in there. And to be able to protect the brand and, you know, fund future development, I think it makes sense. But, you know, you basically wouldn't be able to launch it. Yeah. Unless you did that. So it's more of a requirement. On the Superchain side, the idea is, you know, one, shared upgrades, kind of decentralized governance of these chains. And then in the future, like interoperability. And so the idea is that, you know, L2s might fragment the space in some ways where, you know, liquidity gets shared across a bunch of different chains. In that world. So you may want to simplify bridging between L2s and, you know, obviously there are third-party solutions today. They might be expensive. They might be slow. And so what Optimism is really building with the Superchain is like trying to build a unified kind of role player that facilitates, you know, easy composability or like not necessarily native composability, but like easy interoperability between these networks. And so that, you know, development of that. And to be a part of that is kind of what you're buying.

Nicholas: What is the, is there like a technological solution that enables them to facilitate the movement of liquidity between the different chains that are part of the Superchain? Or is it more a social agreement?

Andrew Huang: Yeah, there will be a technological thing. I believe they posted the first interop spec. And I believe Arbitrum also had like a research post on this. So it's seems to be an idea that's in like, obviously. I think every. Ecosystem is trying to build their own version. Generally, the requirements here are you need the roles to go through the same bridge so that, you know, the assets don't have to L1 doesn't need to be notified essentially. And, you know, you don't need to go through that separate seven day withdrawal or kind of settling. And then to the, the chains that pass messages to each other need to be able to validate the other chain and like roll for optimistic roll ups, like roll back together. And so. That, you know, kind of opting into shared fault proofs and the additional infra burden of running other, other full notes. So it's, you know, it's, I would say an expensive infrastructure solution to a, you know, seemingly valuable interop problem. And so, you know, excited to see how this develops amongst the various frameworks.

Nicholas: Awesome. Okay. I have a bunch more questions about running rollups as a service, but first it's time for the sponsor message. Today's episode of web3guide.com. Galaxy brain is sponsored by Dymo. Dymo is building a stable coin personal bank. I just played with the app. Actually, it's really slick receiving a USDC from a link that creates your pass key based smart account. Very, very cool. They've been shipping fast with a small team and they're looking for their first engineering hire. Dymo is really dedicated to the mission of your keys, your coins and free and open source development. So if you're interested and you're a developer who'd like to join Dymo, their mission is to deliver the best of the best. They've delivered this liberatory cryptography to a global audience and they're looking for a passionate real world Ethereum developer. So you can reach out at founders at dymo.com. If you're interested in, uh, in working with the Dymo team on their exciting new project and you can check it out in the app store today, if you're interested. So back to conduit, how involved are rollup customers in operating their rollups? Uh, we talked about like one click deployment basically, but is everything pretty much handled by conduit or do you need to have some protocol engineering on your side? If you're a customer.

Andrew Huang: Um, pretty much everything is handled by conduit, uh, and that's kind of the, you know, the product that we want to offer, right. It's like you get your own chain, click of a button, um, you integrate it with it as if it's any other chain that you've developed on. Um, and so we want to make that as seamless as possible. Um, and so that's kind of the, the conduit guarantee, if you will, um, it does tend to be a little more like we provide a bunch of support to our customers, um, because, you know, launching our chain, it's kind of a big deal. Um, and sometimes they have, you know, a bunch of questions about it and it's kind of a new thing. Um, but generally, you know, again, it's as if you're building on any other chain. Um, and, you know, I think with the customer list that we've had, we've like, we're super excited at kind of being able to facilitate that, that process zero to one, like in a really seamless manner. Um, I think for, for example, for like Zora network, they went from test net to main net in, I think like four weeks. Um, and most of that was just testing the test net and like. Integrating and making sure to like, they're one of our first customers. They're just like making sure this thing is real. Um, we've actually seen folks, uh, I think, yeah, we launched base L threes on Wednesday and then we got a request for a main net actually that day. Um, and then we deployed it the next day. And so that the timeline to getting to me and that was just getting faster and faster. And I think, again, that's where the self-serve aspect is really going to accelerate things.

Nicholas: So the dev X really is just like change the chain ID in your front. End in your Forge deployment and you're all good.

Andrew Huang: RPC URL, chain ID, um, and then you're, you're good to go.

Nicholas: So cool. Super cool. Actually. Um, I listened to that bankless episode you did a month or two ago and you mentioned flash roll-ups. I'm curious if, uh, there's any like interesting, uh, applications of flash roll-ups or other kinds of, I mean, we're so like conditioned with this, uh, certain perspective of what an L two or even L three should be. This kind of long live to change. I mean, that is EVM equivalent. I'm curious either within the idea of flash roll-ups or other alternate constructions. What are some like cool ideas you've seen or heard about related to roll-ups that maybe people aren't as familiar with yet?

Andrew Huang: Yeah. So caveat with like, we hear a lot of crazy ideas. Um, I don't endorse any of these or recommend any of them. Um, but, uh, just to tell you a couple kind of crazy ones, I think flash layers is definitely something that we've heard, uh, actually the first version of conduit. Um, was, you know, thinking through flash layers and basically you run a bunch of, you know, expensive computer, um, you run a game around, right. Kind of off chain. And then you set all that results on chain. Um, and so people can kind of like fast withdraw from that. Um, I think it was just a little early, a little ahead of its time. I think L threes are probably a better form factor for that. Um, where, you know, it's kind of cheaper to do this on an L two and there's probably, you know, more activity or retail activity going on to, to enable that.

Nicholas: Um, so you have like a time limited game. Experience and then like, uh, endless, uh, claim period for the rewards. Yeah.

Andrew Huang: So, I mean, there's various ways to do it. I think like one version we've seen is you've run like, uh, for example, like, uh, there's that recent crypto, uh, game, I forget what it's called. Maybe it's something crypto, the game or something. Um, and so you can imagine that happening on a custom rollup that they made, it runs for 72 hours a week or whatever. Um, and then after, you know, the game is done, um, you just. Uh, settle everything to, you know, the L one or the L two wherever, um, and then you can spin down the role. So you kind of only pay the costs when the game is actually running and you might, for example, be able to spin it up or like reuse it again, kind of at a later date. Um, these use cases, frankly, today are probably a little more, um, experimental, um, I would say, um, and that like, again, blockchain is a bit about kind of this persistent state, how that grows over time and building on top of it. Um, I think the withdrawing aspect is interesting. Um, but there's also, I feel like, uh, such an advantage to building your own roll up and building, compounding that state that the flash stuff, um, we just haven't really seen yet in a meaningful way. Got it. Some other crazy ideas that we see if you, if you want another one is, um, well, it's that natively settled to multiple chains. And so the idea is that instead of just settling to one chain, like Ethereum, you might settle to, you know, arbitrary and like optimism. Um, or, you know, a Solana chain, for example, um, at the same time. Um, and that way you get to avoid, you know, I guess the fragmentation of a bunch of these different bridging protocols and like wrapped assets. Um, but these things are again, like super experimental and I wouldn't recommend anybody invest a ton of time yet just cause like huge lift and, you know, maybe, maybe unclear if they'd actually work in practice.

Nicholas: How are, uh, rollups that are, uh, Um, Dex centric, uh, like the one that you mentioned, uh, earlier, I understand how something like Zora where they can simply swap out the default network where you mint, uh, an NFT, for example, and then, you know, even do things like the kind of mint to bridge, uh, transactions in order to get liquidity onto their chain for people to be able to mint in the future. But if you're, if you're L2 or L3 is primarily, uh, about trading tokens that live natively on other chains, how. Do such rollups handle the liquidity problem?

Andrew Huang: Yeah. Um, I think it depends. There's multiple ways to do it. Um, I think AVO has a unique model where their architecture allows for, um, Like a lot of capital efficiency, frankly. Um, and I think it's in some ways a marrying of this off chain order book with an on-chain kind of settling. Um, and so it's kind of a centralized exchange experience on top of rollup. Um, and so the advantage is, you know, you have, uh, you can see the assets and the bridge, you know, they're not going anywhere. It's not like FTX where you might have some money over here in FTX, some money over here in Alameda, and, you know, you lost track of, of where, where it's gone. Um, and so in some ways that security guarantee, um, you know, has enabled AVO to, uh, in addition to their shipping speed, frankly, um, has enabled them to kind of get to the, it feels like dominant position they're at today where, um, you know, they have. Great liquidity on the network, um, with a mix of, I believe third-party bridging as well as, um, you know, onboarding market makers onto the platform. So, um, that's primarily, I believe where kind of that experience comes from. And, um, yeah, again, I think off chain order book, being able to facilitate tight spreads and very quick kind of UI UX, um, feels like it's been key to that.

Nicholas: So it really is just like a superior application that appeals to its users that gets the convinces them to bridge.

Andrew Huang: I think that's right. I think that's right. I think like it would be too expensive to build on anything like this on layer one. I think a layer two uniquely enabled the cost structure to build something like this. Um, and yeah, I mean, they've done a killer job at, at executing so far.

Nicholas: You mentioned that there are people have been including some precompiles, uh, into their rollups. What are the most popular things people are doing to modify the vanilla EVM?

Andrew Huang: Um, we've seen a variety of things. Um, I don't know that. I don't know that any of them are, you know, super compelling yet. I think a big one that we see is like passkey verification, um, which, you know, I think, um, yeah. Um, it is possible today on chains and like, since L2 gas is pretty cheap, um, you know, it's not too bad. Um, particularly, you know, Dymo, uh, your sponsor is actually a good example of somebody who has done some great work around compressing the call data. Making, you know, kind of traction transactions, a lot cheaper on layer twos, because again, like compute is cheap, call data expensive. Um, and so, um, it's not too bad. Um, today we see some other ones around like CK precompiles. Um, so for things like, uh, you know, doing, doing some kind of, uh, you know, poker game or random number generators or things like that on chain, um, to actually enable those, uh, types of experiences. We've seen some, I mean, this one, you know, used to come up more, um, but, you know, for example, if you want, um, you know, the black Shoals kind of pricing, um, as a precompile to enable, you know, options style, um, Dex's fully on chain. Um, but yeah, generally we see a wide variety of things. I think like the way I think about precompiles is that over time, the compute will get better. Um, and you know, precompilers are in so many ways. Like a hack around the EVM, um, you know, being kind of expensive, um, over time, you're going to run it on, on the EVM itself. And I think on the arbitrum side, stylus is a good example of the compute getting better, right? It's a Watson VM that interoperates with the EVM, something crazy like a hundred X or like 500 X cheaper on compute and memory. Um, so things that you couldn't build in the UVM today, you can, you know, kind of go to town with. Um, on stylus. And so I'm very excited to see, you know, what types of applications that spawns and you can imagine, um, over time as the EVM gets better, some of those applications even come back to the.

Nicholas: That's interesting. Yeah. I had, um, Nick Dodson from fuel on the show last week and, um, they have also a very different take on what a roll up VM could look like, uh, with like a UTXO parallelizable UTXO model and, uh, taking advantage of like multi-core. Um, as opposed to like the sequential, uh, validation of, uh, of an EVM style blockchain, um, do you think about, do you think stylus is, is the most interesting path forward? Uh, or how do you think about like deviating more wildly from the EVM tradition?

Andrew Huang: Um, yeah, I think the challenge with any new VM is the tooling, um, and the, you know, just huge developer base that Ethereum already has. Um, it's also unclear. I mean, like, I think in the long run with, um, you know, uniquely enabled by rollups, right. Being able to experiment, you'll see, you know, parallel the VM. I think what, for example, Georgia's a paradigm is doing with ref, it has just been like a 10 X improvement on, you know, the Ethereum client and the performance aspect. Um, and so you can imagine that combined with parallel EVM, more future improvements gets to a place where, you know, the performance isn't the issue. Um, and you know, you can actually build anything that you want and it's just a question of what do we build and like, is there the demand for it? Um, and so today we have a lot of, you know, like, uh, speculation or gambling or, you know, shit coins and whatnot of that, you know, pump prices and like, uh, gas costs on the theory of main net. And these are kind of inherently, you know, trying to all access the same state at the same time because everybody wants the MEV against it. Um, I think what does the world look like? Where, you know, you have parallel EVM, you have different fee kind of costs for different accessing different types of state. This is somewhat similar to the Solana style, um, fee market. Um, and then, you know, you can build a ton actually on the same layer without getting congested. Um, and so in that world, do we change application design? What does that look like? Is there the demand for those types of applications? Um, and in particular, it has the ability to bring fees down, um, really, really low, even lower than, than they are today, frankly. Um, and so I think that's the ultimate question for me is I think the VMs are really cool. I think EVM will probably get there. And then I think it's really on app devs to ultimately as an infrastructure guy, uh, it's on the app devs to kind of build the compelling experiences that are going to bring more and more people on chain.

Nicholas: So then if you're, uh, talking to an app dev today, maybe the, what I'm hearing is what matters most is that you generate demand with interesting applications and the infrastructure will evolve to support the demand. Once it is. Revealed, uh, but actually the, the biggest challenge right now is finding applications that are compelling enough to sort of make greater, greater and greater demands upon the infrastructure.

Andrew Huang: Yeah. I mean, I think that, um, you know, app devs have to build with what the infra is capable today. So the infrastructure always has to get better, um, before the app devs can build for it. Um, but you know, what will accelerate this are applications that demand more of the infra. Um, and I think it has to happen in a. Gradual way. I think there, you know, maybe a poor example is like web two company wants to come in and like port everything on chain and like wants like, you know, 500 billion gases. I kind of, it's like, yeah, that's a little out of scope today and probably, um, but the application that consistently needs like 10, 20, 30 million gas a second and, you know, continuing to climb is a really good use case for the, okay, let's gradually kind of expand the performance capabilities of the stack to grow with these. Styles of applications.

Nicholas: Uh, is, is centralized sequencing a problem? Do we need a solution to this problem or is it not a real problem?

Andrew Huang: Um, I mean, I think the, you know, going back to the history of like. why wolves, I think the general idea was like, even if you have a centralized actor, you design the protocol in a way such that, you know, it's not a big deal. Um, and you know, you can kind of look at, um, a lot of what kind of Ethereum has, uh, kind of in its core dev process as a way. To insulate from, you know, the failures of any particular centralized actor. And I think rollups are kind of an extension of that where they're designed in such a way that, Hey, you know, it's actually, you know, with fault proofs, they can't do the wrong thing. Um, and that, you know, even if they try to censor you, you can force transactions in, even if it's a bit slower. Um, and so I think the exit opportunities, the fact that, you know, they can't steal your money with fault proofs and things like that, I think are really compelling reasons for, you know, the, the kind of. Scalability and, and kind of cost structure benefits that they ultimately bring. Um, and like, you know, if you don't want that risk, you're free to just use Ethereum main net. Um, and like, I think for a lot of whales, you know, that might make sense, but, um, ultimately if we want to bring, you know, the next billion or kind of the entire world on chain, they're going to need to be, you know, cost structure improvements to, to make that a reality.

Nicholas: But I guess there also need to be sort of, I mean, in theory, if a sequencer stopped producing blocks, I don't know. Oh, has it, has it happened yet that an alt, a third party has taken up the mantle and figured out the, you know, spun up what's necessary in order to use the data to, to create their own blocks?

Andrew Huang: So it depends on the role of framework. I think with orbit, uh, or nitro, um, you don't need a live sequencer to exit. You can actually permissionlessly propose these like, you know, messages that, um, allow you to withdraw. Um, if this, you can. So it goes down. for example, yeah. For OP stack, I believe it requires a live sequencer today to actually process withdrawals. Um, even if you force, uh, include it, it needs to make its way into a block and then get posted. Um, that hasn't happened today. It's still, you know, today someone with permission system. Um, but this is something that over time, you know, one with the super chain, having a decentralized kind of, um, security council for kind of holding the keys of the role actually enables like, you know, a kind of recovery there, um, in the event that that happens, um, but to, over time, you can imagine permissionless forms of this, where anybody can actually post a state route. Um, and so in the future, you know, uh, particularly in the, uh, in the case where, um, you know, a sequence or goes down, you, the protocol enforces a way of producing blocks such that anybody can post a state route against that. Um, and then you'd be able to withdraw there in the current system.

Nicholas: I mean, this is a bit in. The weeds, but how would the, if there are multiple proposers, how is, how are, how is the, the winning block decided?

Andrew Huang: There are multiple proposer. So essentially what would need to happen here in this, uh, so there's something called the sequencing window. Um, and so like, let's say, uh, sequence or goes down for the sequencer window, which is typically 12 hours on the OP stack. Um, after that period, every block produced, um, basically is forced to be an empty block. Um. Except for the inclusion of deposit transactions, which are the kind of the forced transactions from layer one. Um, and so basically you have this 12 hour kind of shifting window, and then, uh, you'd be able to, um, essentially enforce at the protocol layer because that, uh, condition is part of the fault proof. And then you can post a state route, um, from basically the last one that shows that these are the force included blocks. And so even if, you know, the sequence were to go down. Um, in the future, you'd actually be able to permissionlessly post a state route that shows that the sequencing window is passed, that your deposit should be force included. Um, or, um, you know, the sequencer is live and post that data because they have to, um, and then you, you get to withdraw anyway. So I think that per transaction basis, the force inclusion per transaction basis. Um, although the state route, uh, in theory basically includes, um, like at a certain point in time includes all the previous, uh, kinds of withdrawals. Um, so, um, but yeah, basically TLDR, I would say today, the, you know, role frameworks are still in development. They still have some of the training wheels on, um, but a lot of these are going to come off, uh, really in the next, you know, six to nine months or so.

Nicholas: Uh, how do I, how should I think about MEV in roll-ups, maybe the state today and where it's headed?

Andrew Huang: Yeah. So, um, Arbitrum is probably the most sophisticated here today where they have the FCFS, um, and like first come first serve. So. So basically the first transaction that comes in gets processed first. Um, so they have like kind of a more, uh, transparent policy around it. Um, and then they're extending that with like something like time boosts, which, um, you know, give. it is, uh, similar to, you know, paying for, you know, tighter connection to the stock exchange, for example, which is a lot of, um, you know, ARP shops, uh, in TradFi. Um, so that's kind of their policy. Um, and you know, it's transparent. It's transparent in a certain way. Right. Um, and so I think the, the important thing is that the rules are clear and everybody knows what they are and there's no special kind of treatment. Um, and I think they're doing a good job at that on the optimism side. Um, you know, it's, it's something that isn't quite as developed. Um, today it's just using a private mempool with the standard, uh, you know, PGA auction. Um, so the higher your priority fee, you just get ordered by that. Um, in practice, I just think that blocks. Um, blocks tend not to be full, so it's pretty easy to get included and there's not quite as much competition today. Um, and you know, most of the MVV challenges are frankly just on the, uh, later one side versus the, the later two side. Um, over time you could imagine similar solutions that L1 has. So, um, flashbots is maybe a good example, MEV boost and, you know, its successors, um, being co-opted to get integrated with the later twos to enable, you know, the same kind of, um, you know, MEV. Got it.

Nicholas: Um, we've reached the end of my list of questions. Is there anything that you think developers who are thinking about, uh, roll-ups, maybe deploying their own roll-up, maybe working on applications and wondering if their application is a good opportunity for spinning up their own L2 or L3 or any of the tech that we haven't discussed? Is there anything that comes to mind that we should, uh, discuss before calling it?

Andrew Huang: No, I mean, I think that it's, you know, launching your own roll-ups. Roll-up is really becoming the default here. I think that, you know, one thing that we spend a lot of time coaching, uh, a lot of potential customers through is that it's much easier to start on your own roll-up, um, than it is to migrate later. Um, there's a lot of apps that, you know, we're thinking about it launched on, for example, base or some other roll-up and are now going through the pains of, okay, we have this app. It's, you know, kind of working and we have users and they'll be mad at us if we try to launch our own chain. Um, and so it's just one of those things that I think as it becomes more of the default, as we bring the prices down, as it becomes easier and easier, it kind of just makes sense to get started as your own chain. Um, and then if it doesn't work, it doesn't work. You can always deploy on a shared layer, but it's much, much harder to migrate that state after the fact. Um, and so building for the success case where it massively takes off, you have the built-in cost structure advantages. You have the built-in sovereignty of the built-in kind of performance. And, and app and everything, um, I think is a very compelling option and we're seeing this kind of flood of demand kind of coming in. Um, and so that would just be my recommendation going forward is just to think thoughtfully about the evolution of the app, the success case, and making sure that, you know, you're prepared for that.

Nicholas: And if, if an application does choose to launch directly to say their own L3 on top of, let's say base, um, what is the user experience for their newly onboarded users? They're going to have to bridge to the L3 to get started. Right. It does add like a one, at least one step of friction.

Andrew Huang: Yeah, it can add one step of friction. I think the benefit of, you know, that style is actually. there's also, um, once they're bridged in, it's a little bit, uh, sticky, right. Because it takes seven days to withdraw. Um, and so there's in some ways that if you can get them over with incentives or an initial pop, um, the retention is actually quite good given that, you know, it takes seven days to withdraw. And only a couple of minutes to bridge in. And so, you know, frankly, that might be another, um, you know, pro launching your own chain is kind of the native assets or just, uh, kind of that asset.

Nicholas: Awesome. Andrew, thanks so much for coming on the show today. This is a really interesting conversation about conduit and all the roll up ecosystem. Thanks for explaining everything.

Andrew Huang: Yeah. Thanks for having me.

Nicholas: Absolutely. Hey, thanks for listening to this episode of web three galaxy brain. to keep up with everything. web three, follow me on Twitter at Nicholas with four leading ends. You can find links to the topics discussed on today's episode. In the show notes, podcast feed links are available at web three galaxy brain.com. web three galaxy brain airs live most Friday afternoons at 5:00 PM, Eastern time, 2200 UTC on Twitter spaces. I look forward to seeing you there.