On The Brink with Castle Island - Brannin McBee (Core Weave) on moving from GPU mining to specialized compute (EP.258)

Episode Date: November 10, 2021

Core Weave cofounder and CSO Brannin McBee joins the show to talk GPU mining and high performance computation. In this episode:  Details on the Core Weave's latest fundraise Core Weave's hybrid data center model which combines Ethereum mining and generalized computation Transitioning from just crypto mining to the specialized compute market Why the hyperscale cloud providers aren't always suitable for firms with specialized computation needs The nature and size of Core Weave's GPU and CPU fleet How Core Weave services clients in VFX and rendering, machine learning, and blockchain Why Core Weave was able to build a generalizable computational fleet Why the demand for VFX/rendering-based compute is accelerating How film studios are going from capital intensive on-prem computation to outsourcing it to cloud providers How Core Weave is able to achieve a high utilization rate for their GPUs How Core Weave is working on open sourcing GPT-3 Why you can't access scale compute at conventional providers How Core Weave is managing Ethereum's transition to Proof of Stake When Brannin expects the final merge to occur How the market for GPU mining will change over time How NFTs are a new source of demand for large-scale compute Core Weave's attitude to Ethereum development Did Ethereum move away from some of its original principles? How proof of work was a strong distributive force during the early days of Ethereum Are miners less positively inclined towards the network now that their business model has an expiration date? Brannin's attitude towards MEV today Brannin's thoughts on Ethereum's security model as it moves towards PoS Does the abundance of capital make PoS more vulnerable than PoW? Why diseconomies of scale for electricity procurement protected PoW networks Why rendering PoW 'synthetic' in PoS makes it easier to pull off an attack How Core Weave pursues sustainability in their operations Sponsor notes:  This episode supported by Public.com. Start investing with as little as $1 and get a free slice of stock up to $50 when you join Public.com today. Visit public.com/onthebrink to download the app and sign up. This episode is brought to you by Withum, a top 25 accounting firm with a cutting-edge Digital Currency and Blockchain Technology practice. To learn more, visit  withum.com/crypto.

Hello and welcome to On the Brink. I'm Nick Carter. This episode is brought to you by Witham and Public.com. More about them later in the episode. So today we're sitting down with repeat guest, Brandon McBee, who is the chief strategy officer at CoreWeave. Now, Corweave is a cloud service provider that mines Ethereum part of the time and other GPU coins with their enormous GPU fleet. And then they train data models. and render visual effects the rest of the time. So they kind of have a hybrid model cycling back between those kinds of jobs. And today we talk about how they manage that fleet and how they've been able to build up that computational asset and how their outlook on Ethereum has changed as the merge approaches.

And this is a really interesting episode from the perspective of one of the largest Ethereum miners, probably the largest in North America, how they engage with the core developer community and how they're preparing themselves for the potential obsolution of proof of work mining on Ethereum. Let's dive right into it. Brought down by bad mortgage investments,

Lehman, which has 25,000 employees, will be liquidated. The federal government loans American International Group, AIG, $85 billion. This is a different kind of market, and the Fed is asleep. The federal government is stepping it to stabilize Fannie Mae and Freddie Mac, the two mortgage giants that have been threatened by the housing crisis. The Bank of England has pumped $75 billion. and pounds more into Britain's ailing economy with a new round of quantitative easing.

You print a couple trillion dollars and all of a sudden people start to worry. So out of this worry, we have something called the Bitcoin. Hello and welcome back to On the Brink. I'm sitting here with repeat guests, Brandon McBee. Hey, Nick. Hi, Brandon. Hello. Welcome. Welcome back. This is what your third, is this your third appearance? Second appearance. Third appearance for Correve, though. You had Brian on, I think,

sometime earlier this year chatting with you. That's right, two of my favorites. So Brandon is the chief strategy officer at Corweave and co-founder, even. Correve is North America's biggest Ethereum miner. Is that fair to say? Definitely one of. I would say we position ourselves today as an accelerated cloud provider, and we're the largest provider of accelerated cloud computer.

compute in North America now. So very exciting this morning. The announcement dropped. You guys raised a very chunky round. So congratulations. Yeah, thank you. We raised $50 million from Magnitar Capital. They're a private equity fund focused on providing growth capital to businesses.

And we're extremely excited to be partnered with them for our series B round. So you have an interesting business. I'm proud to say I'm a shareholder full disclosure. You do a hybrid model where partially you're, you know, hashing away mining Ethereum and other proof of work blockchains. And then partly you're doing, I guess, high performance computation. And so probably one of the only firms that's actually successfully been able to perform that hybridization.

Yeah. So it's, you know, myself, my two co-founders, we come from an institutional finance commodity trading background. And, you know, one of the most important parts of our jobs and careers was just assessing risk on a daily basis and just to have the best optionality into the markets we participated in. And back in 2017, when we founded Core, we've, we looked at the GPU. as having amazing optionality into high growth markets.

And over the last four years, we started on the blockchain side, so building out a miner. But the last three years of effort on the engineering side and the infrastructure side of the business has been fully focused on building an enterprise grade cloud, and specifically to serve the specialized compute market, or accelerated compute, which is high-end GPUs, high-end CPUs, super high bandwidth storage, so like MVME storage, SSD storage, all that kind of stuff that businesses today are increasingly

needing and aren't, don't really have an option in the market in the hyperscale clouds. The hyperskill providers tend to focus on having a lot of fairly mid-tier compute, and that's a fantastic fit for a lot of businesses out there as we go through this transition of on-prem kind of hardware resources into cloud resources. But there's absolutely a subset of the market that needs specialized high-performance compute. And they don't really have an option for that today. So we built the business, kind of bootstrapped ourselves up using cryptocurrency mining. Then, as I said, it's been the last three years or so focused on hiring the amazing engineering team that we have today, identifying the correct partners to work with on sourcing enterprise grade hardware and really building out the infrastructure that's needed to run a high-performance specialized cloud today.

And it looks a lot different than the warehouse that we started in to run specific. specifically crypto mining. Today we're located in tier four data centers across the U.S. to serve our clients. So it's been quite a run of it. So tell me about the nature of your fleet and the scale there. I mean, is it fair to say that you're one of the larger buyers of GPUs in North America? Yeah. So we are we have a great partnership with Nvidia. We are their only elite. cloud service provider that kind of puts us one tier below where aWS gCP and azure sits and as far as we understand from invidia we are you know about their fourth largest buyer of uh enterprise grade GPUs from them so we're fully focused on that uh enterprise grade GPU that's you know a 100s

a 40s the whole rtx series and that's the kind of infrastructure that um you know, enterprise businesses need today to, to operate. It's very different than the kind of retail assets that we started, started the business with in early 2017. So in terms of the breakout of your revenue, I mean, it's still still majority, you know, crypto mining, but I presume you anticipate that's going to change and that ratio will shift over there. Yeah. So we, we launched our cloud product in January of this year. And I'd say we expect the year to finish up with cloud services representing about 14% of our revenue for 2021. And based on our current growth, we expected to hit about 50% of our revenue next year. And what does that actually,

what do those jobs look like? I mean, so you're idling, you're mining Ethereum or some other Provework coin and then when a job, one of these big old compute jobs comes in, you turn your attention to that and then you go back to idling mining theory. Is that sort of what it looks like? Yeah. So I look at our compute as having optionality into kind of three main markets. That's the media market. And we look at the media market in terms of its growth trajectory as just as like the amount or the volume of content. being created, like all the Netflix's of the world and just how much they're pumping out right now. And then the complexity of the content, the like amount of VFX that's used in a production of all these media assets being produced today. The next sector is kind of really data is the best way to think about it.

That's your artificial intelligence and machine learning processes. And it's just an insatiable amount of demand. there. All the data that we've been kind of producing globally for the past 20 years is now being consumed by these algorithms to draw relationships between all those data sets. And, you know, that's just going to be how business's function going forward is looking deeper and deeper into these data sets to figure out how they can draw conclusions and better orient their products and services. but it requires a ton of compute to figure that out. So we're very active in that space.

And then the third space, I really think, is blockchain. Blockchain requires a substantial amount of compute to operate, whether it's proof of work and hashing a blockchain for consensus or it's supporting a L1. I think Flow announced recently that they're working with GCS, to operate their network. So I believe that blockchain is going to be a substantial consumer of compute, and that's absolutely one of the markets that our infrastructure spends a lot of time in.

So our stack kind of works, as you were discussing before, such that whenever we're not allocated to clients within the more traditional markets like media or data clients, it defaults to operating within the permissionless revenue stream that is blockchain, because we're able to kind of enter and exit that market very efficiently. So the exact same compute that a client is leasing to train their machine learning model on, when they're done with that, it gets released back into our hardware stack and begins to mine whatever the most economic cryptocurrency is. And my understanding is that we are the only operator in North America that has done this on an enterprise scale.

And why would you say that's the case? I mean, obviously there's a lot of publicly traded Bitcoin miners. And, you know, they don't, I mean, some of them might make noises about HPC, but that doesn't really make any sense because all they have is ASICs. whereas you guys obviously have a ton of GPUs and other sort of more general hardware. Is it your sort of unwavering optimism around ETH that caused you to continue to build at that fleet

even as maybe other folks thought that the proof of stake transition would come sooner? Was that sort of the key advantage you had? Or was it just this view that you could move in this direction of being a sort of hybrid computational resource from inception. Like what is it that led you to, you know, occupy this unique position, basically? Yeah, that's a great question. You know, I think we've been looking at this from day one as an option for us for the future,

but we certainly didn't understand the complexities involved in it. you know, there are plenty of pitchbooks floating around for GPU-based miners and even ASIC-based miners out there talking about building a high-performance cloud, but at the end of the day, you can't service these enterprise clients from a container in West Texas or Colorado or the Tundra in Canada, right? Like you need to be in a Tier 4 data. center, you need to have low latency, you need to have all the backup redundancies that come with sitting within a tier four data center. So that that's a path that we have taken pretty much 90 cents on the dollar of what we've generated on the mining side of the business for the last

four years and invested it and building this cloud product. And this latest financing round is really going to allow us to aggressively pursue a market where we believe we've identified a fantastic product market fit and now really comes down to just scaling the business. So when you look at these sources, these three sources of demand, I want to take them in turn. So the media, the rendering VFX business, I mean, is that kind of a COVID thing that all these studios have gone from IRL, you know, live action to more rendering? Or is there just structural growth there because the world has a forever increased demand for that?

What is that? What are the dynamics look like there? Yeah, I think there's a couple of tailwinds going on there that we're really excited about them. So one that you mentioned, I think just the volume of media content being created and the complexity of that content is really driving demand on compute resources in order to produce it. Right. Like there's just a lot more content being created today than there was five, 10 years ago. The second question about, you know, COVID, remote work, all that kind of stuff. The, the

VFX industry was already in a transition to move from on-prem resources into the cloud before COVID. It was at the very kind of beginnings of it, but it's also a huge lift, right? Like, you're the, the production platform is the heart and stole of these VFX studios, right? And messing with that or messing with the way that artists interact with their compute to work on on a daily basis can be incredibly disruptive. So it's definitely a slow mover from on-prem to cloud pre-COVID. But COVID certainly accelerated it. And it provides a number of advantages for these studios. You know, if you think about it, right, these studios would have to procure a substantial amount of resources that ultimately went pretty underutilized because you have it's very bursty demand in terms of the way that you kind of consume compute.

Like you have artists who are working on machines on a day-to-day basis, but, you know, every month, two months or so you go and to a large rendering project, which is like putting together the final production of the asset you've been working on. And that consumes, you know, multiples more compute than on your daily basis. So you kind of have to have that sitting idle until you're ready to use it. And thus very low utilization rate, poor use of capital. And I think the VFX studios have understood that. But they really didn't have an option before. There wasn't a kind of virtual student.

in the cloud product before. So that's exactly what we set out to build. And late last year, we partnered with a fantastic VFX studio to design the most optimal product for their team. And in a way that we could scale it across to other VFX studios across the US and the world. So now the stack, instead of all sitting on-prem and people having workstations at the office,

their workstations are in the cloud. out with us. So they log in via BDIs and their desktop kind of sits on our cloud infrastructure and their storage of all their production assets, like all their files and everything, sits on the same infrastructure as well. And when they go to actually render these films, they can scale up to thousands to tens of thousands of GPUs to get their project rendered in a timely and cost efficient manner. And it's, it's, It's just something that these guys didn't have access to previously. And from what we're told is super disruptive to kind of legacy workflows in the VFX environment.

So when we hear about like interstellar, you know, rendering their image of the black hole, if you remember with the whatever it's called accretion disk, if you saw that scene. Yeah, yeah, yeah. You know, that apparently took, you know, 24 hours to run to each frame. Would you be able to do that faster? Oh, no promises, but I'm barely certain that we could have got that done a lot faster. So, I mean, these studios, did they literally have, like, server farms, like in the closet or something? Yeah, they literally had server farms in the closet.

And it's all, I mean, it's a, it's very capital-intensive. It's a balance sheet item that just sits there and depreciates. And again, the utilization rate was just super low. So being able to bring them on to our infrastructure as resulted in a substantial cost savings. I mean, most of the VFX studios that we work with are reporting anywhere between a 40 to 60 percent reduction in costs for their production. And by contrast, your utilization rate is almost 100. Correct. Our infrastructure, because as soon as, you know, that artist who logged in for the day,

on his workstation, he stops using it, then we take both the CPU and the GPU compute that the artist was using and we allocate it to the blockchain. So moving on to data, what kind of, I mean, can you talk about what kind of data projects you've done? I mean, anything exciting there?

Yeah, you know, I can talk about stuff fairly broadly. You know, one example that I'm really excited about is we are working with an open source developer team called Eleuther AI. Aluthor is working on open sourcing GPT3. GPT3 is a large machine learning model that Microsoft purchased from OpenAI a couple of years ago for pretty large sum of money. I think it was close to $12, $15 billion. dollars or so. And this group would like to open source it. So we offered our resources to help them on training. And it looks like we're about this kickoff training that project for them, which is really exciting for us. That will leverage NVIDIA's A100 GPUs of which we have a pretty sizable cluster of.

You know, another great example is we have been working with a protein folding simulation company that works with pharmaceutical companies to help discover new medical treatments and pharmaceutical treatments for various diseases. And they will run thousands, tens of thousands, hundreds of thousands of permutations of folding these proteins to help in drug discovery. You know, another health science application we've been working with is radiological outputs or x-rays for cancer screening and kind of matching that up with the radiologist's interpretation of those x-rays to support and the algorithmic identification of cancer.

and to help that radiologist see things that they might not immediately recognize or at least prompt questions, right? So it's really a broad array of different entities that will come in and leverage this compute. And I'd say the differentiator between us and some of the hyperscale clouds is the hyperscale clouds will provide a lot of software resources to have. help teams that don't have like a strong developer or engineering staff. Whereas core we've, we focus on offering pretty much strictly compute, bare metal access or kind of virtualized access to infrastructure. And that allows for us to take a lot of the costs out. Typically on a, you know, skew by skew or GPU comparable basis,

we come in at about 70% less than the hyperscale clouds. And that just really allows for, entities that are doing all this fantastic research to accrue not only cost savings, but they can run more jobs at the end of the day and just allows them to operate their businesses better. So my mental model around this is, I think it might be called the Jevins paradox, but basically as you make a commodity cheaper, and actually those total analogies to block space here, but as a commodity becomes cheaper, maybe you discover some new industrial process to produce it, the world learns new use cases for it because simply the price point has come down

and so you can just do more with it. And compute in my mind is a commodity. And it seems like it's becoming cheaper all the time. And so my guess is that we're just going to have greater demand for it as the price comes down and it's become more and more ubiquitous. And we're just going to it's going to become completely normal for just regular folks to you know outsource enormous computational jobs to to the cloud whether you know edge computing i guess they call it whether it's you know me using Photoshop and it running slowly on my laptop and faster somewhere else or you know gaming i guess is also happening and is that kind of the way you think about it just as it gets cheaper just massively stepped up demand for computation.

I do agree with that. I also think that there's a little bit of a bottleneck today as well and just being able to access scale compute. You know, we get told by clients pretty frequently that they can't just step into AWS and request for 3,000 GPUs on demand. I mean, we operate over 50,000 GPUs at this point, And it's pretty wild to us that clients, and these are fairly normal mid-sized clients, right, that are looking to burst up to 3,000 GPUs. That scale just isn't available at the traditional generic clouds, I guess is the best way to put it.

So we think that we are one of the first providers really offering these resources at scale. And it's changing the way that these businesses are able to approach using customers. compute in the cloud. And in terms of actual data sharing, I mean, if you're dealing with terabytes, do you DHL it or, um, are you, you know, is it all still digital transmission? Like, have you ever mailed anyone a hard drive? No, no, no. We don't have to step down to that.

But, you know, a reason that we don't have to is that we are operating within these tier four data centers, right? Like we have that 10, 100 gigabit, um, connection at our sites. So things that you can't really quite attain again in like West Texas in a container or something along those lines. Like you really do just have to be positioned within these super high grade data center environments and to be operating, you know, enterprise grade infrastructure in order to

work with clients who are looking for high performance compute, just hard stop. And that's that's what we've found. is what really has allowed us to differentiate ourselves against our peers. I mean, I kind of look at the GPU crypto mining space as a version one of what operating infrastructure for the blockchain will look like. And I think that we're kind of a better definition of what version two will look like. Version two is someone who operates these enterprise-grade resources that can be used for multiple high growth use cases that exist today and might exist in the future.

You mean, you know, one analogy I really like to explain it is I think that we're the equivalent of a power plant, right? Like, we're providing this resource that is used by countless types of consumers and is just critical and part of a backbone of the way that business functions. We are the power plant of high performance compute. Let's take a quick break to talk about our sponsors. Now, on this show, we only really like to work with sponsors that we actually know and can recommend. And both sponsors today are great examples of that. So our first sponsor for this episode is Witham.

They're a top 25 accounting firm with a cutting edge digital technology and blockchain practice. So wherever your company is, from pre-seed to IPO, they have tailored solutions just for you. They've helped some of the largest companies in the crypto industry with audit, tax, and advisory needs. And they've also helped some of our portfolio companies, too. To learn more about their advisory, audit, and tax services, go to on the brink. Dot link slash with them. That's on the brink. Dot link slash with them.

Next up, we have our newest partner, public.com. Getting started investing, especially in public equity, can be really difficult and confusing. It seems like there's a lot of barriers to getting started, especially with cryptocurrency. Public.com is a platform that allows you to participate with fractions of shares instead of having to buy whole shares. It's also very community driven, so there's a large community of investors discussing stock picks and trades. And now they've also added support for cryptocurrencies. Now, I've really enjoyed my time on public so far. One of the things I really like are the audio discussions, Public Live, where you have

public market CEOs and equity analysts sharing their views on particular stocks and themes. It's been a really interesting experience so far, and I like that there is public equity alongside cryptocurrency. Now, one of the interesting things about public.com is that they take the privacy of the community really seriously. They don't participate in payment for order flow, which is a common practice in which the brokerages actually sell your trades to third parties. Instead, they route your trades to the exchanges directly, so there is no middleman. You can start investing with as little as $1 and get a free slice of stock up to $50 when you join public.com today. Visit public.com slash on the brink

to download the app and sign up. That's public.com slash on the brink. Now this is valid for US residents age 18 and over. It's subject to account approval. See public.com slash disclosures. This is not investment advice. So I want to get to Ethereum because you've, you know, pretty intertwined story with them. But before that, so just, you know, seems like you're kind of hinting at this convergence between proof of work, you know, the normal, the the normal operation of proof of work and then proof of stake validation becoming much more computationally intensive. And there being kind of a convergence there. So maybe you speak to us a little bit about what that might look like.

I mean, is there a world where like even if, you know, Ethereum moves away from proof of work and then a lot of these smart contract chains for proof of stake, there's still a role for your industrial data center in blockchain value. validation. Yeah, absolutely. That's a fantastic question. You know, one of my favorite topics when I'm chatting with investors and potential investor groups. We are firmly of the opinion that the ETH2 merge will come. You know, my opinion is that's probably something, sometime around Q2, Q3 of 2022. Tim Biko recently tweeted that he put forward. He put forward. fourth a EIP proposal EIP 4345, which will delay the difficulty bomb till June of 2022.

And that's all in the spirit of allowing enough time to properly test the merge process and execute on the merge as well. And I think that that's fairly conservative estimate. You know, maybe it slips till Q3, Q4. But overall, we believe, that that will move forward this year. And that's a goal that Ethereum's had since day one. And I think it's a fantastic milestone for the blockchain. But that being said, we think that there are an amazing number of opportunities outside of just Ethereum for this enterprise-grade compute infrastructure. I think that some of the more legacy miners that are operating in Tier Zero data sitters that don't have the storage and the CPU compute to really back up and run a cloud might struggle with this

transition. But we've really been positioning and front running this merge for several years at this point. So to offer a little bit of perspective on what Ethereum mining and other blockchains look like today, Ethereum emits about $64 million worth of proceeds. to miners on a daily basis right now. If you were to look at the other 35 top GPU-based networks that are kind of only proof of work-based, they emit an aggregate about $3 million a day, right, with front runners on there being coins like Ethereum Classic. But I think, and to offer context for those numbers, Bitcoin emits about $60 million a day currently to minors, right?

So, I mean, in total, you know, we're looking at something that's over $120 million for kind of institutional crypto mining right now. And it's just an amazing market that you're seeing a lot of capital flow into. But what we see other opportunities in the mining space are these more high performance products, such as Filecoin. Filecoin currently emits about $24 million a day in revenue to infrastructure participants on their network. and we think that there's going to be more opportunities like that on the blockchain in the future. Another example with a network that's still starting up, and I believe they're moving towards launching on the GPU side sometime later this year, is the Akash network.

Akash is trying to be a cloud service provider on a blockchain using distributed compute. So that kind of covers where. where things are today, right? I think, you know, one of the founding accelerators, really, of Ethereum was its ability to reach out to hundreds, thousands, millions of people so easily because their initial distribution mechanism was, hey, if you have a computer, CPU, GPU, you can participate and become part of this community. and we'll reward you for that.

And it got people involved. It got people to learn about what this network was. And they built this massive community of contributors of hardware that still exists today. I mean, plenty of people are still mining today. It's still very much profitable. I can certainly attest for that. And by moving to proof of stake, you know, you're gaining a new community, but you're also, you know, potentially losing this.

you know, hardware enthusiast community out there. And I believe that there are going to be new layer one proof of work chains out there that come in and take advantage of this already established community of tens of millions of participants. You know, one of those chains that's out there is called ALEO. This is a smart contract based L1 that is focusing on anonymity and zero knowledge. knowledge proofs to be their kind of distinguishing component and differentiator for when they launch sometime in Q1. But again, they're going to be focused on proof of work. And I think that they'll be able to pick up the community that will lose access to Ethereum later this year. And that'll be a huge growth opportunity for them. And then, you know, aside from that, blockchain itself is just

generating a fair amount of demand with, you know, like NFTs, right? Like we are rendering NFTs on our infrastructure right now for issuance. I think we've done two or three runs in the past few days alone. Anything good? NFTs, yeah. Like it's like it just kind of comes down to. Blockchain itself is generating an immense amount of demand for compute. And we are just very well positioned as an enterprise provider to sit there and help folks kind of fill their demand profile.

So that's interesting. I'm sure that wasn't something you would have anticipated when he started the company. No, it's not. But, you know, again, kind of goes back to this inherent optionality that we've built out. We have optionality into high growth markets, no matter how small or big they are. We can pivot our compute into all these various growth sectors quickly and at scale. And that's something that we're just not aware of anyone aside from the hyperscale providers that are able to do that.

Yeah, I love the idea of just building an enormous computational resource and then just throwing it at anything that demands it. It's really fun to work with. Regarding Ethereum, you know, you guys built your fleet based on. on Ethereum growth primarily. Is it fair to say that you have a kind of like a bit of a battered spouse style relationship with the Ethereum community in terms of them

constantly marginalizing and being relatively hostile to miners? You know, it's been somewhat of a volatile relationship, I guess. I mean, we certainly try to participate in the network during 2017, 18, 19 or so. We even offered resources to the network for various test nets. I think that we attempted to be a strong part of the community,

but at the end of the day, I'm just not sure the community really wanted us there. It's certainly been on the roadmap to move away from a proof of work-based mechanism since day one. But I think that there were other kind of roadmaps, milestones and promises on there. such as being ASIC resistant that were kind of ignored by the network. I think EIP was it 1057 or so was meant to stop ASICs with the implementation of progressive proof of work.

That was meant to limit ASICs on the network, which I think estimates today are that ASICs represent about 30% of Ethereum's hash rate, which is a very material number. and the network overwhelmingly voted in support of 1057 when it was proposed, but the Ethereum developer team still chose to ignore that signal and ultimately was never implemented. I think it was a pretty political source of tension at the time when that came out, and there was a lot of people for it, a lot of people against it. I do think that that was a promise that was never delivered on. And I'd say that we were certainly pretty disappointed in that. So your point is that it's just a little bit arbitrary, which of the original visions, you know, ended up being executed on?

Yes. Yeah. They certainly seem to be not in favor of the infrastructure providers of what has allowed this network to operate and grow for the past, you know, was it, five? years now. So it's, I think the miners have been a huge part of the community. I think that they're, you know, not necessarily thanked for their contribution directly, but, you know, at least guys are are still making a living, participating on the network. Yeah, I always find it curious that Ethereum's, you know, essential doctrine includes this notion of being proof of stake, whereas they, for half a decade now, they've not been proof of stake, which is kind of amazing.

because in my view, and I think you probably share this idea, is that mining has been a very strong force in terms of distributing the coins in a permissionless way, in a no-K-Y-C way, in many cases, to anyone with commodity consumer hardware. And it actually really wasn't that difficult to acquire some ether through mining with your gaming rig, especially in the earlier days. And so that's been a strong distributive, dispersive force that did get a lot of people into. it and yet, you know, the more common narratives are that mining is this, you know, burden and this sort of this industrial process, which needs to cease. I completely agree with that. I mean, it's the same way that that Bitcoin was distributed, right? And again, I think it's a community that won't be ignored in terms of its size and its, you know, technical ability to participate in these networks. And so I think you will see new L1s pop up that take advantage of this, you know, already established group of people.

So here's an interesting question. Well, I suppose I can't decree that it's interesting, but I think it's interesting. So, you know, to the extent the miners are like positively disposed towards the network, they're, in effect, stewards of the network, you know, their general incentive is to make sure the network is working well. And this is part of the sort of security model analyses that a lot of people do is the idea that miners have hardware and whose value derives from the unit price of the native token, both the eth. And, you know, Bitcoin, obviously. I mean, if Ethereum disappeared, you know, the GPUs would surely be worth less. Not zero, but certainly worth less. And so, you know, I think that's sort of an established idea that miners have a lot of hardware.

and so they generally have a long-term view that they should support the network. Now, as Ethereum moves closer and closer towards effectively obsoleting GPUs and the general hardware that's used to mine the tokens in a proof work model, do you find your time preference accelerating, you know, your discount rate increasing?

Do you find yourself sort of less positively disposed towards the network and more willing to, you know, maybe take actions which are profitable but aren't necessarily, you know, strictly good for Ethereum. And, you know, obviously the main example I'd think of there would be something like M-AV. Have you noticed that or, you know, are you going to, you know, not to put words in your mouth, but stay responsible stewards, you know, right till the end? Yeah, you know, I think that minor extravers minor extractical value was something that we were pretty against when it first came out because we viewed ourselves as stewards of the network, right? Like we don't think that you should be able to be paid to prioritize which transactions go through on a network in a way that disadvantages other market participants.

right I think though that the mining community slowly iterated towards being open to MEV-based transactions because of the lack of response that there was from the Ethereum dev community because of you know things like EIP 1057 not passing even though all the support was there for So today I think that MEV is a pretty consistent part of most mining pools. We participate on mining pool that has MEV enabled on it. And now it just seems like it's part of the network today in Proof of Work. And it seems like it's certainly going to be part of the network as we transition to proof of stake. And arguably, it could even be more pervasive and centralized.

in the manner by which players will participate in it. Now, in terms of, you know, who's actually reordering transactions, are you doing, I don't know if you want to answer this, but are you doing that on a first party basis, or are you participating in one of these systems whereby you auction off the ordering of transactions to the third party transaction pickers? It's definitely more kind of second, third,

derivative. That's certainly where we sit. I'm sure there are miners who are on that kind of first layer working directly with the parties that want to either reorg the blocks or prioritize the blocks. There are a number of mining pools now that just offer M.E.V. to their kind of pooled participants and thus it's just a frankly it's a fairly small increase in revenue for those mining pools. We're talking about like ones of percentage points, right? It's not, it's not massive or on the scale that, you know, some people talk about that being, you know, like the primary source of revenue for miners. That's not where we sit. We choose to be passive participants in MIV. I'm sure there are some very active ones that do accrue material earnings from participating,

but that's just not where we are. So I want to talk about the proof of stake world. I know that you're going to step down your relationship to Ethereum once that occurs. But returning to the mining security models, you know, there's a few papers that talk about the ability to borrow ASICs or hash rate as a security threat in proof of work. And of course, you know, the canonical examples here, the sort of nice hash augmented attacks on Ethereum Classic or Bitcoin gold, where, you know, simply put, there's just a liquid market to rent hash rate, meaning that you could acquire a large portion of the

validation of the hash rate relative to the aggregate amount and potentially interfere. with the blockchain and conduct a reorg. And I don't think that's controversial, but those rental marketplaces probably made those attacks easier. Now, if I think about the analogy to prove a stake there, the fact that there is a lot of capital that's very liquid in these lending markets, and even on a highly short-term basis, is effectively costless to obtain enormous amounts of leverage through flash loans, for instance, which we know are.

used in most defy attacks. Do you think of those things as analogous is the ability to borrow capital for an attack and proof of stake as somewhat analogous to the ability to borrow hash rate to pull off an attack in proof work? Yeah, look, I think that's a really interesting question. I think it's analogous, but arguably it's even easier to pull off that attack using capital. for proof of stake versus securing hash rate and proof of work. I think the most secure part of proof of work is that it requires at-scale access to electricity. Right. Like you don't have 50% plus of the Bitcoin network available on Nice Hash right now.

You don't have 50 plus percent of the network of Ethereum available on Nice Hash right now. don't have 50 plus percent of the network of Ethereum available on Nice Hash, right? It just, it just doesn't sit there. And thus isn't even a mechanic to attack those networks. Certainly smaller networks have been attacked by leveraging Nice Hash. You know, we actually stepped in and helped secure Ethereum Classic when they were 51% attacked. We threw our entire hash rate into them to stop that attack back in, what was that, 2019 or so, I think.

But that security measure gets taken away when you move to proof of stake because at that point, all you need is access to capital. And if you can get these massive flash loans, the size of which, in terms of the variability, is just going to continue to increase, I think that that will result in material vulnerabilities on these proof of stake networks and L1s. And like you said, you're already seeing it as an attack vector on DFI protocols. I think it absolutely is something that it is more of a security risk than what exists on proof of work today. For the at-scale networks, of course. Yeah, I guess with proof of work, there are ultimately dis-economies of scale because you're talking about installed power. You're talking about actual physical resources, you know, your ability to draw, electricity and deploy hardware against that electricity.

In a lending of a crypto asset perspective context with staked asset, whether it's for validation or even for governance purposes, I mean, we've seen flash loans being used to exploit governance processes. There don't seem to be that many dis-economies of scale as you gain more capital and hence sort of political power. You know, you're only bounded by the size of the lending pool. Yeah. And so, you know, it seems like there's a difference.

And rendering the whole thing synthetic makes it potentially easier to pull off an attack. I agree with that. You know, it's something I'm sure that the teams working their way towards these proof of work networks are considering. But at the same time, you know, these attacks. usually happen and the least suspecting corners of the world. I don't think anyone really anticipated, you know, like the sandwich-style attacks that occur on the Ethereum blockchain right now, or really thought through the way that gas fees would evolve on the network as they are today. So it's

certainly a risk, and I hope that there's a lot of caution and thought and less urgency to rush something to market that could potentially risk what is such an amazing ecosystem that is truly still in its infancy. So regarding sustainability, that's been a big topic for Bitcoin miners. And Bitcoin has some interesting properties that I think mean that it is potentially more tractable to go after some renewable sources, even if they are interrupted or don't have perfect uptime with when I think about your operations seems like you do require you know very strong uptime and you know tier four data centers as you say so how do you think about

sustainable energy and your energy mix generally and you know what are you targeting there yeah so our data centers are strategically located or optimally located within electricity footprints that have a high penetration of renewables. So we approach it with, you know, either offsetting or working with data centers that offset their carbon footprint entirely, or working within power pools, again, that are very substantially penetrated by renewables, like 40, 60 percent, no carbon emissions. And we think that as the grid continues to progress towards a low carbon environment that, you know, we're going to be choosing to operate within areas that advance faster than others.

Well, Brandon, this has been really excellent. Always an interesting new topic. I think last time we had you on, we talked about the Texas meltdown on the failure of the grid there. So very different topic this week. Any last words or just. general solicitations. Yeah, you know, I heard that you're starting to hand out tungsten-based fud dice

to everyone who comes on the podcast, so I'll shoot you over my address, and I'm very excited to get those. Yeah, I can't confirm that I've made those. I mean, it is possible. I've been told it is possible to make a 12-sided tungsten dye, which is engraved. I believe it'll be slightly more expensive than the plastic version. So the weight is just so pleasing. Yeah, the pleasing feeling.

We'll need to sell some more ads to finance that. Thanks again, Brandon. This has been great. Thank you, Nick. Chatsu.