On The Brink with Castle Island - Brian Venturo (CoreWeave) on the journey from mining crypto to the cloud (EP.163)

Episode Date: December 30, 2020

Brian Venturo is the CTO of CoreWeave, the largest North American GPU miner, which also doubles as a cloud infrastructure firm for general computation. In this episode:  How Brian went from hobbyist Ethereum mining to wielding a fleet of 50,000 GPUs How Core Weave was able to cheaply acquire GPUs from insolvent mining farms The implications of the Ethereum DAG file growing beyond 4 GB and its effect on miners The effect of the looming PoS transition on miner decision-making Why Core Weave specifically limited their GPU fleet to NVIDIAs with more memory Brian's stance on Ethereum's transition to Proof of Stake Are miners pro-cyclical or counter-cyclical? Does Core Weave hold inventory in the coins they mine or do they divest them immediately? Brian's options based model for pricing hardware The mini gold rush happening in publicly traded mining firms Brian's opinion on hashrate swaps or derivatives How Core Weave is transitioning from purely mining based to public cloud services The feasibility of GPU miners moving into cloud computing Core Weave's plans to train an open source version of GPT-3 Brian's view of the role of miners in the Ethereum ecosystem The motivations behind monetary-related EIPs in Ethereum Brian's analysis of incentives to make Ethereum deflationary Why Proof of Work is an underrated distribution method Brian's position on increasing the gas limit Why tinkering with the fee market is counterproductive in terms of transactional efficiency The relationship between blockspace and fee revenue for miners Brian's thoughts on EIP 1559 and whether it increases Ethereum's security Brian's view on the pace of ethereum development Why Proof of Work launches are preferable to liquidity mining launches The current state of the ASIC v GPU debate Brian's retrospective on the Grin launch Why Core Weave is not interested in miner-extractable value (MEV) Sponsor notes:  Withum is a forward-thinking, technology-driven advisory and accounting firm committed to helping our clients be more profitable, efficient and productive in today's complex business environment. Our Digital Currency group is proud to partner with members of the cryptocurrency community. Get to know us at withum.com/crypto.

What's up, everyone. Welcome back to On the Brink. This is Nick Carter. This episode is brought to by Witham, a top 25 accounting firm with a digital asset practice, more on them later in the episode. So we're resuming our mining miniseries. The episode we did with Mustafa Yohan of Bigson was incredibly popular and hopefully we can balance things out by bringing an industrial Ethereum miner on the show. Today we have Brian Venturo, the CTO of CoreWeave, which is probably the largest Ethereum miner in North America. They boast a fleet of about 50,000 GPUs. This episode is about how Brian took a hobbyist mining practice to an enormous fleet, how they made these opportunistic acquisitions during ETH's 95% drawdown and built a mining behemoth.

And are now pivoting to other netherous. non-mining use cases for those GPUs, including rendering, AI, and other cloud computing use cases. A really fascinating transformation, and it shows how those GPUs can be repurposable. Brian is fairly well known for some of his critical comments on proof of stake. That's how I first found out about him was when he pointed out that in a proof of stake system, on-chain lending can be competitive with stake deposits and can impair the integrity of of a staked system. I think this hypothesis has really been borne out subsequently. Ryan was really early to that idea. So we talk a lot about what it's like to be an Ethereum miner, especially as

Ethereum protocol policy has become increasingly hostile towards miners with multiple updates being incorporated, designed to marginalize miners and actually contract their revenue. We also talk about the tinkering in the fee market. Why proof of work is an underrated distribution method and the state of the ASIC versus GPU debate. We cover a lot of ground here. Really fascinating episode from the perspective of a large industrial Ethereum miner. 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 pounds more into Britain's ailing economy with a new round of quantitative easing. And print a couple trillion dollars and all of a sudden people start to worry. So out of this worry, we have something called a Bitcoin. Bitcoin. All right, Brian Ventura. Welcome to On the Brink. Nick, thank you very much for having me. This is really exciting for me. I've wanted to have you on the show for many months. Dare I say even years at this point. So this is a great special occasion.

I've been trying to convince you I have nothing good to say, but here we are. So let's see how it goes. No, I know better. Yeah, you have all kinds of interesting stuff. We're doing a mining miniseries right now. So our last episode with a Bitcoin miner was really, really popular. and now we're going to cover Ethereum mining. So it's going to be great.

Let's do it. So by way of introduction, Brian, you at CoreWeave, would you say you were the largest Ethereum miners in North America? Is that a fair thing to say? Yeah, so we're, from what I know, and I do try to do a bit of digging on this just to know other people in the industry. We are the largest Ethereum miner in North America that's at least talked about it publicly. I think there's probably a pretty good chance there's one or two that may be in the weeds that are larger than us, but from what I've been able to gather, yes.

So you're a pretty interesting story. Tell us about your sort of pre-crypto journey and then how you decided to become an Ethereum miner. Yeah, so I'm an energy trader by background. I started at a college actually at an environmental asset management firm. And I wound up running the U.S. energy book there after the financial crisis. And I'm specialized in natural gas. So directional trading in U.S. natural gas. It's kind of a deep end of the pool in terms of kind of the volatility that you see. And in 2014, 2015, the markets went from being crazy volatile to pretty tame. And the volatility junkies that we were, we were.

we're kind of always looking for something. And we stumbled across Bitcoin and crypto. And, you know, it was a total joke to us at first, really like, hey, let's go bet Bitcoin on the pool table. And we would like bet half of Bitcoin per pool game. And we denominated our fantasy football leagues in Bitcoin. And then one day it really turned into, hey, this was like, price started to move. And we're like, well, maybe we should take this seriously.

So late 2016, we decided that we wanted to learn more about, like, crypto and how it was made and where it came from and what was happening with it. And the first thing that we did was we bought some cloud mining contracts. And then very quickly said, okay, like, if somebody else can do this, like, we can probably do it ourselves. And we ordered a couple GPUs online from Amazon Prime Now that were delivered to our office in like an hour. and within a couple hours we were up and running, and that was our first entry into crypto mining. We got into Ethereum, mostly because it was very accessible to us, right? It didn't require specialized hardware.

Like I said, we had the hardware delivered within an hour. And it kind of just went crazy from there, right? It was, this is something that we were doing as a personal hobby to then some of our friends and associates found out we were doing it and said, hey, could you do some for us? And then we went from our office in downtown Manhattan to grandpa's garage in New Jersey, like as cliche as you get, to our fleet today, which is over 50,000 GPUs in five data centers across the U.S. So it's been a pretty wild ride. That's amazing. So you didn't plan to become an industrial miner, and you kind of stumbled into it.

And so yet here you are. Yeah, you know, we were running it really more of as a hobby initially on, right? Like it was never, hey, let's go do this for a living. That came probably about a year into it, a year and a half into it. But early on, we were, we wound up really building things for our friends and associates. And it was during kind of the middle to late 2017 hardware squeeze where we wound up not really owning any hardware ourselves. Right. is we were so busy building for other people that we didn't have that much.

And that really continued through early 2018. And then when the market kind of crumbled, we were put in a pretty unique position where there were distressed assets in the world. And we looked at it and said, you know, this is a pretty binary outcome here. Either crypto is going to make it or it's not. And we were pretty big believers that it was going to make it. So we raised some money in a seed round and actually went out and bought distressed hardware. So that was really the big springboard for how we became like industrial-sized mining operation.

So you basically made a countercyclical trade where you looked at all these insolvent mining farms and you picked up GPUs for pennies on the dollars, is that kind of fair to say? Yeah, so we took down a couple really big installations. One of them was actually up in Canada. We bought some distressed hardware and a couple different data centers, and we were buying. We actually found some blocks of brand new hardware

that were really just never shipped to the U.S. and kind of left in a warehouse. But we had a pretty discerning eye on what we wanted to own, right? And we wanted to own NVIDIA assets. We wanted to own stuff that had enough video memory to be able to run past the 4-gigabyte. level of the DAG file, which happens at about eight days.

And, you know, at the time, this stuff was real liquidation fire cell pricing, right? And we were making the binary bet of is crypto going away or not? And for me, like, that's a pretty asymmetric bet. Right? And, you know, looking back on it, we did it for some pretty big size on our side. And, you know, I think the only thing that we did wrong was not make it bigger. Do you think that because there's this view, this kind of manifest destiny in Ethereum of moving towards proof of stake and that being the ultimate objective, does that mean that firms were kind of less willing to invest in mining solutions for Ethereum because they always sort of had that overhang? Yeah, I think you still see that today, right? I mean, even through this whole thing, right, as proof of state.

has really been delayed a number of times and then phase zero the beacon chain finally launched in december and while that is like a crazy step in the evolution of Ethereum like it's not there yet um i think that you saw two different things right you saw the lack of industrial scale miners come into the space right you don't see 100 megawatt farms of GPUs they just don't really exist and you didn't really see crazy effort on the ASIC manufacturer side to deliver a viable product to market. Bitmain's E3 was just as power efficient mining Ethereum as the Nvidia 1060 series was. So there wasn't really a technological leap there.

And I think that all of this was governed by the fact that proof of stake was always quote and quote around the corner. Right. Why invest in an ASIC if it's in the stated objective of the protocol designers to totally deprecate proof of work. Yep. So you hinted at the DAG file growth. This is kind of, I think maybe among your peer group, people are aware of it, but I would say

in the wider crypto community, people don't know about this. So tell me about this kind of taking time bomb that exists in mining and how it privileges and disempowers certain classes of hardware. Sure. So the DAG file was a piece of the ASIC resistance. that came with the Ethereum algorithm ethash. And every, I believe it's 30,000 blocks, what they call it, a DAG epic, the DAG file changes

and increases in size, right? And the idea of it increasing in size is that it really limits the ability for an ASIC manufacturer to go and say, hey, I'm going to build this super specialized piece of hardware that doesn't have the ability to grow, right? So if they're gonna build something that's supported for more than the next month or two, they have to put memory overhead on board.

The DAG file is approaching four gigabytes, which is going to eliminate the feasibility of using a lot of consumer cards, specifically the AMD radion series that was at four gigabyte of VRM on them. That happens, I think it's like around midnight on December 23rd, so like right to start of Christmas evening, where this big block of four gigabyte GPU, is supposed to drop off the network. So, you know, us as a company, and when we made the investment early on,

we didn't purchase anything that was less than 6 gigabyte at V-RM, knowing this was coming. And, you know, at the time, the industry was making the bet, oh, well, my 4-gig GPUs will be usable through the end of proof of work, and proof of stake will be there, and it won't matter anyway.

So I think we're about to see a bet on proof of stake, even on the mining side, people that were mining proof of work kind of go really wrong. That's fascinating. So because miners had this kind of short-termist attitude, they loaded up on AMDs and cards that didn't have the ability to deal with the larger DAG file, whereas you had sort of forecasted this, and you figured that the transition of proof of stake would take longer than everyone was expecting. Yeah, my take on the transition of proof of stake is like, I think it's going to happen, right? But I think that every day that this network, work operates at the scale it's at today makes it harder for people to really pull the trigger,

right? It's just you've got a however many billions of dollars enterprise and how many other coins and projects running on top of it. Like that's a pretty big bet to make and there are risks associated with it. So I think that, you know, as this platform as Ethereum has grown, they've rightfully gotten more cautious with how they roll things out. Right. And I think that people look at, people have historically looked at Ethereum as the fast mover. But on the main net, I think that you're seeing that slowed down over the past two years, right? And it's kind of you hit this point of critical mass where people aren't really willing to mess with it as much. So due to this obsolescence of lower grade hardware, I mean, do you, I presumably expect a hash rate impact from this around Christmas Eve?

Correct. What do you, what kind of magnitude you're expecting there? You know, this is always the crazy question, right? So the Bitmain E3 A6 disappeared a couple weeks ago. And it was a question of, are they like 1% of the network, are they 15% of the network? It was a measurable event when they disappeared, but not as much as some people expected. I'm hoping it's going to be a pretty big impact just because I'm motivated to hope that. but you know it could be anywhere from 10% to 40%.

That's material. So one thing that I asked Mustafa, who is part of Bikshin, which is a Bitcoin mining firm, among other things, and I'm always curious about this, is do you think that miners are generally pro-cyclical or countercyclical in terms of affecting the unit price? Do they hoard coins in bull markets and then liquidate them in bare markets or do they do the opposite? So this goes back to the this kind of goes through a question of like the types of miners and how they operate. Right. I mean, I can speak about us just because I really the only thing I'm certain about.

We don't hold inventory, right, in crypto. We actually have a suite of auto trading software that kind of goes out. that we built that sells coins as soon as they come in and goes back from whatever the crazy kind of alt coin we're mining to Ethereum to US dollars. We go back to fiat as fast as we can. We look at our mining exposure as it's actually in our in our hardware, right? So every GPU that we own is worth some amount of synthetic Ethereum. Yeah. Right. And that's really where we look at our length and our exposure to crypto in the hardware itself. I think that when you see miners that go out and spend $100 million in mining hardware,

and then they also hold all of the mining rewards in inventory, right? Like they're just increasing the bet. Right. And if the idea for me is to go out and to produce coins below market price, right? I'm really a spread player. And I'm selling those coins to lock in my margin. But I don't think that you see firms like commodity production firms like Exxon who go out and produce produce crude and then sit on massive stockpiles of it hoping that price goes up.

Right. Right. They're producing and selling to market as it comes on. As it comes to market, right? And, you know, I think that some of these miners are almost like many hedge funds. And some of the publicly traded ones, too, is pretty crazy. Yeah. So because you already have exposure to ether and to a lesser degree other proof of

coins, which is instantiated in your hardware, you don't feel that you need to double it up by also holding onto the coins that you mine. Yeah. Like, if I'm going to try to increase my exposure, it's better for me to go out and purchase more hardware than to just hold the coins. Right. I mean, if I'm just going to produce the coins and then hold them, I might as well have never joined the mining enterprise at all and just bought coins. So we were talking about this recently. You were talking about your model for pricing hardware.

I saw a good piece from Leo Zang about how to model A6. How do you think about pricing hardware, whether it's A6 or GPUs? Yeah, so this goes back to the distressed asset time in the market when we were looking at, like, what price to pay for things and what their real value was. And it was never really, okay, like you can never look at it and say, well, if we expect Ethereum to be $100, what is the GPU going to make? right because it's kind of disingenuous. It's usable for a lot of different things. And we had an idea of what the volatility around our Ethereum and alt-coin earnings could be. So we actually, at the time we developed a model to look at GPU hash rate as a string of daily options.

And the string of daily options was struck at our cost of operation. And as we go out, we would have, there would be more extrinsic value in that option because we have more flexibility around turning it on or turning it off. But it gave us a really good idea of what we thought the market price for hash rate at the time should have been. And that's really what we use when assets are very distressed to do that. And that's really when you don't really know what it's worth at the time because it's kind of at risk of failure.

Today, you know, you look at Bitcoin Asic prices and you look at some GPU prices out there and it's all kind of crazy, right? You get this market fervor to own hardware. to own hash rate. This is a time that we typically step back and aren't really adding to our fleet. Right. We make these bets when the market's really in more of a dislocated moment. Do you think that there's kind of a structural premium for a certain mining gear because

there's like a romanticism about mining? Like a lot of potential acquirers would rather mine their coins and buy them on the open market and that causes them to be sort of less sensitive to price and more willing to buy at times like these? Yeah, it's everybody looks at it and says, oh, if they look at it and see what their earnings are today. And I think it's some of the players that are getting involved, particularly in the ASIC side, there's always new entrance to the market, right?

And they say, oh, well, hash rate difficulty may not increase, or network difficulty may not increase as much as I did last time. And they look at it and they're being rewarded. Like some of these public companies are being rewarded for just grossly going out and buying tens of millions of dollars of ASICs, right, with no plan on how to operate them if the world kind of goes back to like 2019 levels. And I think it's kind of a gold rush. It's pretty, it's just, it's almost like it was in 2017, 2018, the last kind of mining boom.

And you see this in ASIC prices and GPU prices right now? Yeah, you see it everywhere, right? And on the GPU side, it's a little bit different because there have been some new series rolled out by Nvidia and AMD. And they were kind of in a period there where they were like clearing the channel, right? So they had such a big overhang of inventory from the last mining bust on the market that they really had to allow it to work off before they introduced new cards. So, I mean, the 3000 series from Nvidia is a really awesome mining card, but we really haven't seen that come to market and scale yet.

So last question on the financial dynamics of mining. So we keep seeing these products being rolled out, you know, hash rate swaps, hash rate futures. Is this a product that you as an industrial miner would take advantage of? No. Why not? I definitely have a sour opinion on this piece. Probably not a widely held opinion. But I mean, I've been in markets for a long time.

And I've seen esoteric instruments like this come up. And they always sound better than they actually are in practice. Right. And I just don't see myself going out and hedging my hash rate. If I'm going to hedge my hash rate in any aspect, right, it's going to be price only. And then the rest of it is really my operation. and that's really where I provide value. So if you're looking at it and you're expecting to be able to go out and lock in hash rate swaps or whatever it is that allow you to make some excess return, I just don't think it's going to last long.

So I don't think those instruments are going to have a kind of high level of acceptance or utilization. I think that people have been talking about them for two, two and a half years now. and I don't think I've seen that many trade. Okay, well, somewhat contrarian opinion there. If you listen to this show before, you know that we take accounting pretty seriously. And you might think that's a little strange in the context of digital assets. But whenever there are trust relationships or centralized counterparties, that's where these firms become relevant. I think the industry is in a good place as it pertains to these accounting firms with specific digital asset practices.

Today's sponsor is Witham one such firm. They are a top 25 accounting and advisory firm in the United States. They have a digital currency and blockchain group with experience navigating tax, legislative changes, managing token sales, and managing stablecoin audits, which is pretty critical. To contact their team, go to Witham.com slash crypto. That's WITHUM.com slash crypto. We're going to change gears a little bit because, At CoreWeave, you guys don't just do mining.

So tell me about the other side of the business, what you're doing with the GPUs when you're not mining. Yeah. So CoreWeave, principally, when you go to our website, you'll never see mention of cryptocurrency mining. Cryptocurrency mining is really what we do to monetize our idle compute. And for the past two years here, after we made those kind of principal distress buys, we've been transitioning to provide public cloud service based around accelerated compute. So we service a number of different markets. And it's really like our data centers are split pretty, there's a pretty big divide between

quality of data centers where we operate hardware, where there's a couple data centers that we run that are purely running mining hardware that aren't used for anything else. And then we have tier four data center space where we're running AI and machine. machine learning workloads. We run visual effects rendering solutions for a few Hollywood studios, whether it's HPC and simulation processes for protein folding and molecular dynamics customers. We run the gamut of the whole typical cloud compute space. And then when we're not running higher-end workloads or when customers don't have jobs scheduled, we are running all of our GPUs back in crypto. What would you say the ratio is between

these higher performance compute jobs and then mining so um it's it's changing rapidly you know when you look at our revenue versus last year as a private company you and you look at our compute revenue today from the cloud like it's this time last year our cloud revenue would be a huge percentage of our total revenue but then Ethereum one up five x right right so we've kind of we're a rising tide here has lifted kind of all ships um so we've had significant growth in both sides of the business this year. I'll tell you that in terms of what our focus is internally and where engineering resources go, I'd say it's like 95% to the cloud side.

And then it's really, on the mining side is really just a maintenance, is a maintenance only type of engineering and development effort. So was it a deliberate move to get into the cloud compute as opposed to mining ether? Or did you kind of realize that that was an opportunity once you were established as a miner? Yeah, it was, it's an opportunity we, we really stumbled on to when we were a minor, right? And it came at that time where we made those big acquisitions. And it was, you know, we want to own this compute. We think this compute has value, but we have no idea what the heck we're going to use it for yet.

And then we spent a year and R&D of like, what markets can we serve? How quickly can we serve them? What do we need to do it? And we made a few really key hires that helped us move into that business. And, you know, it's at this point, so the purchasing decisions that we make today are never, are typically never motivated by mining. They're definitely informed by the mining revenue will make off of them. But we're purchasing high-end hardware to serve cloud clients, not necessarily purchasing GPUs specifically from mining today. Relative to the big providers of compute, you know, the AWS and the Googles and so on, do you have,

an advantage that comes from your experience as a miner what would you say your competitive advantage against those like established providers is yeah i think that we're definitely scrappier right is we're a startup and as a mining startup you if you spend a ton of money on infrastructure and making things look really pretty like you're out of business and i own your hardware already but you know i think that we're pretty scrappy in the way that we build things whether that's hardware selection or doing our own chassis design and manufacturing to kind of custom fabrication of components, where I think some of the bigger cloud providers are building things to expect that if they have one server, that server needs to be able to be used by everybody from Joe Bob's garage or the

website down the street all the way up to a government entity that has defense contract requirements. Right. So they're building for the highest end use case across their entire data center fleet. where we're definitely building to the higher end for some of it, right? But we also have a tremendous amount of raw compute that our customers can use to really run highly parallel batch processing jobs. And that gives us an advantage kind of from both the cap-x cost as well as operation cost.

Some of it also is just like some of our competitive advantage is just the fact that we're smaller, right, and that we're scrappier on the engineering side. And we do a lot of engineering when we onboard clients and to help them find the right solutions and help them find the right GPUs. And, you know, it's not really a self-serve business like you find at AWS or GCP. Is this transition that you made? Is this something that other miners can also trivially undertake or did it require a lot of sort of overhauling of the business? Yeah. So this has been a massive, massive CAPEX exercise as well as human resource exercise, right? We've, I mean, I'm millions of dollars in on spend and development in this

already. And a lot of that has been self-funded, right? So it's been stuff that we've done through operating cash flow. It is not something that I think you're going to find a lot of competitors moving over from the mining space. I think a lot of people talk about it. I've yet to see very many of them do it. And it's a, it's the type of thing that you can't do unless you really have substantial substantial compute scale, right? The guy down the street who may be running a thousand GPUs in his garage, that's not scale enough to be able to go out in service industrial level clients. So speaking of scale, before the call, you're telling me a little bit about GPT3, which got a ton of buzz early this year and sort of alternatives to GBT3. So tell me a little bit about what's happening

there with OpenAI and potential alternatives to GPD3. Yeah, so we run an inference service, which is effectively a model serving service for natural language processing models. GPT2 and GPT3 are two of them. We have a lot of clients that serve GPT2 off of our infrastructure, but as soon as GPT3 was released and the impact that had on the model performance and quality of the outputs, we lost a lot of that business, right? And that business went to Open AI and their quote unquote exclusive partnership with Microsoft Azure. Open AI has made the ethical decision that they don't want to open source that model and deliver the model weights to the market. They're saying that they're trying to protect the world from, I guess, bad AI. You know, it's a pretty crazy stance to make.

It seems like it's directly opposed to what they started the company or the organization as. And if that was the case, I don't really understand why they're commercially licensing it via Microsoft. So, you know, Corrieve as a company has actually been looking for and just committed to helping an organization train an open source version of GPT3. So we're hoping that we get to play a part in bringing that model to the world. But I guess we'll find out in four or five, six months or so. So the barrier to entry there is, is it simply? that nobody else has undertaken that much effort in training a model because that's just very expensive computationally to do yeah and here's here's the thing right it's like so they say oh we don't want this

to get out there but any organization that has the means and really wants to do it like this is not a 300 million dollar expense right this is a couple million bucks in compute time um i don't think it's a large barrier to entry to some kind of larger organization doing it and and getting access to it already. And when the genies out of the, out of the lamp on this stuff, I just don't think it goes back, right? So I look at it as it's better to have this open sourced than it is to have it only controlled by one entity. So if you committed all your resources to training an alternative model, how long do you expect that it would take you? If we committed all of our resources, it wouldn't take very long. You know, the type of model that this is, though, is it's really, it's, it needs to be

run in what's called distributed training, right? And it's very dependent upon network throughput. And your, the, the GPUs you're using to train it may never be fully saturated because of the network capacity issues, right? Even if you go to the craziest levels, you're not going to saturate all the compute. So, you're actually incentivized to do it on your highest, the lowest number of high-powered GPUs you can with the best interconnect, instead of putting it across like all 50,000 of our GPUs. Right. So it's the type of thing that we'll train for a month or two with like 500 or 1,000 GPUs that are in the exact

same homogenous build that are very well connected and very close to each other so that the model can train more efficiently. So let's shift gears a little bit. I want to talk about Ethereum because you've kind of made a name for yourself, budding heads. with, I would say, Ethereum leadership and sort of Ethereum stakeholders even at certain points, I mean, especially around Pervis Stake. So I guess the question we can start with is, how do you see miners? What's their role in the Ethereum ecosystem?

Are they just service providers? Are they more than that? How do you see your role in Ethereum? So there's pretty differing views of miners out there from some of the developers. look at us as kind of the scum of the earth. But, you know, when I see miners, like, I think of anybody who's doing crypto mining as a sole operation or kind of a sole purpose, like they're very committed stakeholders in the network, right?

They may not be buying tokens outright, but their tokens are bought through their allocation to hardware, right? And they're buying them over time. And in the process, they're providing security, right? So just because somebody buys tokens today in Ethereum and I generate them over the next three years, it doesn't really make us very different. We both have capital allocations to it. In terms of the minor as a, like, is it a service provider? Like I think that minors are typically service providers, right?

We process transactions and the idea of them being purely a service provider over time is fine until it's not. right and the time when it's not is when the network actually needs them or something's happening and they're there to enforce consensus or to whatever may be but i think that like i said i think that the market assumes that they're quote unquote service providers for the majority of the time but you know like personally it's kind of offensive man like you have these calls with people and they're like you're just a service provider i'm like dude like i spent every day of my life on this Yeah. Like literally, I'm in my office every day, like trying to be better at this and thinking about ways to like to reduce like uncle rates and to and that it just they throw it in your face because they think that all you care about is making money.

And like if all I cared about was making money, like I don't necessarily think I would be like work. I feel like I wouldn't be working so hard. Or maybe you wouldn't be in a theorem minor. Yeah, maybe. I mean, maybe I'd just be a public. Bitcoin miner, I can trade it an absurd valuation. Awesome. Is the latent

hostility towards miners a function of the fact that mining was just meant to be the interregnum before, you know, the glorious progression of proof of stake? Yeah, maybe. You know, I think that, like, it's pretty easy to see that some of the decisions

and some of the EIPs that come out of the Ethereum community are all like, they're all driven by greed, right? And whenever you look at it, like you talk about issuance reduction and monetary policy. And it's like 10 guys that are sitting around, 10 people sitting around that have a big bag of Ethereum. And they're like, oh, how do I get this to go up in value? I don't know. Like some of the decisions that have been made are like clearly just greed based.

And some of the proposals that are out there today are like you can see through it pretty quickly. like yes, they do have technical merits and they think they're actually going to make some changes that make adoption better. But at the same time, like, there's always some deflationary kick in there. I think that there's a very big misunderstanding of how users are acquired through proof of work and proof of stake chains. Right. And in Ethereum, like, I got involved because I was able to mine it. And I know, like, I know hundreds of other people just in my direct network that I've gotten involved in crypto because they were able to mine. Ethereum. And if you look on the different public mining pools, there are hundreds of thousands of

wallets of people mining Ethereum. And, you know, like today, like after the last two years, I get phone calls from people I haven't spoken to any years and say, hey, you do Ethereum, right? Like, what GPU should I buy? And that's a pretty powerful thing, right? Tell somebody, hey, you can go buy this GPU and you're earning Ethereum tomorrow. Like, that's powerful. And proof of stake is really there and it seems like it's going to be rewarding the incumbents. Yeah. So your view on all this is that mining is a distributive phenomenon and helps disperse the coins and then proof of stake is the opposite. It's likely concentrative. Yeah. I think that the proof of work distribution model might be like the best user acquisition study ever. I guess the other benefit is that it's not KYC'd. Like anybody can acquire an ASIC or that might be a little trickier, but definitely anyone.

can acquire a GPU and get started and they don't have to disclose their purchase to anyone. I think it's really powerful that people around the world can take a GPU that may already be in their computer and use it when they're not gaming or not running their Excel spreadsheets or whatever it is for the regular job and earn Ethereum, right? It gets them bought into the ecosystem very quickly. Now they're thinking about, okay, what do I do with this? How do I spend it? Is there any other thing that I can do to make this worth more to me? And it's like these proof of stake chains, they just don't have that. Right.

They're trying to invent ways to get people involved. I mean, look at all the airdrops, right? They're literally just giving money away, hoping that people will come to their chain and use it. Is it fair to claim that Ethereum sort of lucked into this extended distribution period through proof of work, even if that wasn't really the intention of Ethereum initially? Yeah, I think so, right? And I think you're going through it again, right? you're getting rewarded right now for the fact that that price like it's a positive feedback loop

right is price is going up people more people are using the chain mining rewards are going up more people are getting involved it's driving more and more on chain usage and price and usage just keep going higher right and people that are doing it they're not like people that are buying GPUs and whether it's one GPU or 10 or 100 right they're not just going to be there for a day they're making a commitment that they want to be involved in this network for some amount of time that's really valuable yeah why throw that away like just to to have something be super concentrated and security be based in people that have been in the network since the genesis block like you don't need to do that so you hinted at a couple of these changes that were

potentially deflationary these protocol tweaks let's talk through a couple of them so one change which is a more of a spontaneous thing, I guess, is the increases to the gas limit. What is your view of that? I mean, do you think there's a natural limit where we should stop increasing the block space available? So this is another hot button. There are people that argue that increasing the gas limit

increases state bloat, which makes it harder for people to sink the chain, which means that the chain is less operable, blah, blah, blah, blah, blah. Right? I mean, all I, like, no matter what you do, you're going to get to that point of usage or that point of, stable load anyway. I think that increase in the gas limit accelerates the time it takes to get there, right? But the idea that you're not going to allow the network to grow and for transactions to happen because you're trying to give researchers more time to solve those problems, like,

I don't think those problems are really solved until you have like the dire need to solve them anyway, right? So you finally see people starting to do some roll-ups because transactions, fees this summer were so insane that now they're actually doing it. But that like those rollups were available six, eight months ago. They just didn't, they weren't so driven to use them. Because they weren't being financially compensated to do so. Right. So the market's really going to drive innovation here. And, you know, I think that one of the one of the biggest risks is like if you mess around with market mechanisms to solve these problems. Right. And everybody says, oh, well, if I did X, Y, Z, I can keep gas prices lower. And like, yeah, you might be able to, but I'm pretty sure that you're not

going to be smarter than what, like, your contrived engineering exercise is not going to be smarter than what the market's going to be able to bear. Right. And like the, the saying in commodity trading is always high prices, cure high prices and low prices, and low prices, right? So, you know, I think that as we're expanding the gas limit here, or as it has expanded a couple times now, as long as it doesn't have an impact on the network's ability to really stay in sync, I think that's fine. I think that there's going to be a point where, like, it will get crazy and it will cause problems, but I don't think that we're there yet.

But I don't think that, like, I think that you're kind of, you're alluding to your next question is going to be about EIP 1559, which is a fee market change. and basically gets rid of transaction fees for the miner and blah, blah, blah. And I mean, it's, of course, there's a deflationary component in there to help them pump their bags. But like any of these attempts to mess with the fee market as it was originally designed, I just don't think, I think it just delays the inevitable, right? Which is that you have to deal with these things. And if the market drives people to innovate and deal with them sooner, I think the market, the network's probably healthier in the long run. Yeah, so just to be clear, so on the gas limit, you support increasing it up to the point where it interferes with got validation and then sort of keeping it fixed so that people can design around that stable parameter.

And the growth in fees would force people to use technologies like roll-ups to scale. Yeah, and it's not even so much as like I support the immediate increase of the gas limit. I'm not opposed to it, right, but I'm not arguing for it either, right? So I'm less opposed to changing the gas limit than I am to changing the way the fee market works. So the follow-up question there, and everybody has a different answer on this. I've seen some of Vitalik's answers, which I didn't really understand. There's too much math involved. Is like in your just intuition, you know, in your opinion, what would you say the relationship

between block space and fee revenue is because you your compensation is in a large part fee base. So are you of the opinion that increasing block space actually increases fee revenue as well in the long term? I don't know. I think that all of this is based upon increasing adoption, right? And fee revenue is going to increase as you increase adoption. And if that means we can allow another 20% of transaction in each block and that helps adoption, then yes. I don't have a really good forecast of what it's going to do or an opinion on if it's good or bad. I just think that, like, yes, a large portion of our revenue with this point is based on

fees. And those fees are happening because people are using the network and they're doing things like they're doing defy tokens and which is the craziest thing on the planet this year. Right. And they're making these trades and they're sending these transactions because it's economically profitable for them to do so. Yeah. So even if, like, I think that if you increase the block space, you'll probably be back at the same spot in two months in terms of like how much capacity you're using, it'll just be that the network is larger, the network of people using it are

larger. And I guess it's good. I mean, like I said, I think you eventually get to a tipping point there where it stops going from being good for the network to debilitating to the network. So you mentioned EIP 1559. I guess we have to talk about it. It seems directly hostile to miners, I guess, in that it. Oh, it isn't.

So one of the crazy things about EIP 1559 is that there is an, there's an admission that some portion of Ethereum may never entirely go to proof of stake. Right. And there's a discussion of the fact that networks like Bitcoin, that will go to a zero block reward model and rely only on transaction fees may have trouble or may not have trouble, we'll see, paying for network security. So EIP 1559 is like, it's a deflationary mechanism kind of guised as a fee market change to help keep transaction fees lower.

What it does is it basically takes the transaction fees that are programmatically calculated now based upon the last block's utilization and it takes those transaction fees and burns them every block right so they're ether go proof um and on top of that there's what they call a minor bribe which is really bad nomenclature like i don't know if they thought about it good i'm gonna have to put on my balance sheet like revenue from minor bribe minor bribe yeah that's not going to cause any problems yeah like just somebody please be an adult um And that's really like the fee market today where they're incentivizing miners to include their blocks. So like the last iteration I saw was the block like gas limit would increase to I think it was like 20 million.

And then after that point like the fee market would kick back in anyway. So it just seems to me like under normal operating conditions is like all this is going to do is keep transaction fees away from the miners and burn them in favor of the token holders today. So what's the verdict? Does it increase or decrease security for Ethereum? It probably decreases it, right? It reduces the amount of funding that goes to securing the network. There's like there's so many different opinions on how much security is required for network, but we won't digress too much. I think it opens up a couple interesting attack vectors, right, where you now have entities that

say, you know what, I'm going to mine empty blocks because I'm not being rewarded for them anyway until the fee market winds up just like, it's possible that transactions become more expensive because now you have to pay miners just as much as you pay today as well as the fact is like you're now burning Ethereum because of this stupid mechanism to do so. Like it's just one of these things, man, it's like I just don't think that this contrived like process is going to be better than what the free market can bear. Would you say that there's a culture of sort of over-engineering or tweaking the protocol too much? There's a whole lot of I'm smarter than the market going on.

Yeah. And I think that's pretty dangerous. So when we talk about ETH governance, so you mentioned that the rate of change for Ethereum was slowing a little bit, but then also we are seeing these tweaks being periodically pushed through. So what's your verdict of, you know, where Ethereum sits on that innovation versus security continuum? Like, are they moving too fast or too slow? What's, are they in the Goldilocks zone? So I think that in some places they're not moving fast enough, right?

And in some places, they're probably, like, to say that they're moving too fast in things is a little bit disingenuous, right? I mean, EIP 1559 has been being talked about for like a year and a half now. and it's still not slated to go into a fork. So their rate of the rate of change has definitely declined pretty dramatically. I think that their focus now should solely be on scaling. And like stop messing around with fee markets when you need to scale. That's the only thing that's important here is get more like you're scaling your user base, you're scaling your transactions, whether that's layer two scaling solutions or whatever it may be.

but the silver bullet and proof of stake is like it's coming and it's making progress but it's not there yet right and betting everything on that happening in the short term or medium term has so far proven to be pretty bad for the scaling of the network so i take it you feel that your eth mining business is not going to be obsolete in the near term yeah that's a it's a good question I get asked that question like five times a week. Yeah. My answer is always not yet. I do think that there's a day coming when that happens.

I just don't think that we're there. And, you know, I feel like I have a pretty good read on when it's going to happen. And I'm not convinced it's happening in the medium term yet. I think we're still, we're still a few phases away in ETH 2.0 before that really, that chain really becomes usable like Ethereum 1 is today. So it's kind of a wait and see situation. So once we get to ETH 2.0's more functional forms, do you anticipate a proof-of-work version of ETH still existing? You know, in parallel, is it maybe a hostile fork? Is it just that, you know, maybe the bigger smart contracts, DFI projects won't want to redeploy on the new fork?

Yeah, I think that a lot of the value in the authority. of network today is in the usage on top of it. Right. And a lot of these things are defy tokens and projects that are not going to go to some crazy proof of work fork. I think it's pretty low probability at this point that there is like some holdout proof of work chain that runs after that has any real viability. I do think that people are learning that the proof of work distribution model is awesome.

And specifically one that is ASIC resistant, at least in the short and medium term. is awesome for program and network growth. So I'm hoping that we're going to see some fair launches on the proof of work side over the next couple years. There's definitely been some tried, but for better or worse, I think they've been unsuccessful for a number of different reasons so far. But I think that you're going to see some folks

that try to find ways to use the asset that Ethereum has today as Ethereum moves to that proof of stake part of their life. So you mentioned fair launches. It's been really interesting this year to see a revival of the idea, albeit in a different format through the liquidity mining, which is somewhat similar to a proof of work fair launch, but also really different. So what do you make of liquidity mining as a novel issuance mechanism? And, you know, is rendering the whole process process synthetic, is that a fair and complete substitute for a true proof work launch? So I've been personally involved in some of these defy things.

And, you know, a couple of them I've gotten totally hosed on. A couple of them have been insane. Like, I think early on I didn't necessarily understand how insane it was, right? Where you're going to just, you're basically staking X amount of money for 24 hours over that 24 hours, like your tokens aren't really at risk unless you get hacked because they're stable coins or the program gets hacked and then you just make off with a bunch of money. And like initially I was sizing this expecting that like all my money was going to disappear. And like by the third one, I was like, wait, like, why am I not just dropping the hammer on this?

This is like the dumbest thing I've ever seen. I just, I don't think that scenario is rewarding to people like is driving user adoption outside of the incumbents. So the cost of capital, the years quote unquote surrendering by participating in a liquidity mining initiative, you don't view that as a true replacement for actually burning electricity to mine something. Yeah, it's, I mean, it's not just electricity, right? It's building physical plant.

It's having labor on the ground to deal with problems. It's developing software and custom monitoring solutions. it's the R&D that goes into optimizing operations like I think it is like it's the I'm going to call the lazy millennial approach to I'm sure that'll get some good responses but I just I don't believe in it so okay so you want to see some fair launches but you don't like the liquidity mining, how would you, if you had to launch a proof work coin today, how would you do it? And would you go for A6 right off the bat or GPUs? I definitely would not go for A6 right off the bat.

Right. I mean, nobody wins except for anybody that has fab space and that point. I mean, if you have somebody that has a better R&D budget and everybody else, they win by leaps and bounds. I don't have a good I don't have a good idea of how I would do a fair launch right I don't think that architecting the launch of a network is really my strong suit the only thing I would like to see is that it would be accessible on commodity hardware

it would be very hard to build specialized devices for the algorithm you know and it wouldn't have a like no free mines, no founder's rewards, none of that nonsense. It would be pretty cool. And I hope that next time something like that happens, I think it was pretty close with Grin. And then I think Grin got some really unfair bad press,

where people were going out and talking about how there were hundreds of millions of dollars in private equity firms backing mining operations. And that was total BS. Yeah. Right. And like I know because I was running a facility for one of the, SPVs and like it just wasn't even close to that order of magnitude and I knew what I knew what their percentage of the network was. And I knew what our percentage of the network was while we were mining

it. Right. And like I think that that kind of that soured the entire launch for people when they said, oh, it was all private equity back. And the source there like this is one of the problems with crypto like journalism as a whole is that the sources are just total bunk. Right. Like these people have no idea what they're talking about and they've got 70,000 Twitter followers, no offense, Nick, but, um, and they have this crazy, like crazy soapbox. Yeah. So that's why, you know, you don't trust the mainstream crypto press. This is why you got to come to independent journalists, uh, such as on the brink, where you only get the best facts. But yes, I agree. I think grin was unfairly maligned, uh, which was a shame because it was,

It seemed to be a really earnest approach at a truly fair launch. Yeah. And, you know, I think that Grin has actually has been catching a bit over the past couple of days, which may just be spillover from everything else crypto. But, you know, I think that one of the problems they had is they made this commitment to going from GPUs to ASICs on some defined timeline. Right. And now you had to have this network be able to go out and fund the development and mass production of ASICs and price just didn't warrant it.

Right. And I think that was part of it. But I really don't think that project ever got over those initial stories that it was all private equity money and people were afraid to participate. It was pretty sad. And it's funny because every other, you know, token launch where there's a pre-mine and there's a corporation behind it, it's literally venture funds and crypto hedge funds that, you know, get that early access and cash.

at the expense of everyone else. So it's so strange that something like Grin, which was, you know, seemed to be a fair launch was the one that got maligned for that and everything else somehow got a pass. Yeah. Right. I mean, like, and some of these proof of stake networks that launch that have like super concentrated block producer designations.

And yeah, Grin definitely got a bad rap. So last question, Brian, we've had you on for a while. Hot topic. in the theorem community, M.V, everyone's talking about it, you know, minor extractable value. You're telling me before the call that you guys aren't really interested in pursuing it. I know this is going to shock a lot of the sort of researchers that, you know, think of minors as trying to squeeze every last drop out of their privilege as miners or block arrangers. So what's the logic there?

why not go after m a v so some of this goes back to my experience as a commodity trader and you know i was always competing with high frequency traders on the screen um it was like it was the it was horrible right and you go to put on a trade and all of a sudden liquidity disappears uh and during the time i was kind of like i hate this i don't ever want to deal with this like i wish these guys would just go away and this m evi v idea is kind of like trying to force me to become one of those guys. Yeah. And I just don't think it's fair, right?

Like I don't want to, if somebody has a great trading idea in Defi, I don't want to go front run them, right? I mean, they put the work in. I don't look at that as my money to make, right? And I think that's like, yeah, if we wanted to go dedicate a ton of engineering time to do that, we could be really good at it, right? But I have different aspirations for where my business is going to be in three to five years. So the hash rate that I operate and own will

will not be participating in these MEB exercises, right? It's just go find somebody else. Yeah. Do you expect that other, because it seems like many miners are also not exploiting it. Do you expect that it will become commonplace among miners to exploit it? I don't think so. And I think this goes back to the decentralized nature of Ethereum mining, right? It's the fact that it's hundreds of thousands of individual miners.

And if I'm one of the biggest ones in North America, like, and I have no interest in doing it, like, there's not really that many close to our size that would have the engineering resources to do so. I think you're more likely to find mining teams that are writing custom software for smaller networks that are going to find a lot more value than they will in picking off defy traders on the Ethereum network. And like these researchers can write about it all they want, and I'm sure that somebody's out there doing it. it's not me and it won't be me. Okay, so I said that was the last question, but I lied. This is the last question. So when I talked to Mustafa, he told me that he thought a lot of Bitcoin hash rate was going to leave China. And actually a lot of it, really material fraction was moving to North America.

What is your view of that for Ethereum? Do you think it's in flux? And, you know, maybe in the next year or so, do you have a view on like the fraction that's going to be North American domiciled? I think that next week, a large portion of it that's in Asia today disappears. Right. I think that the industrial mining operations that bought 4 gig AMD mining cards, a lot of them were domiciled in Eastern Europe and Asia. So I think you're going to see a pretty big distribution change next week. What does that mean going forward?

I don't know. I think it really depends upon what Ethereum price does and how many people buy GPUs or get GPUs for Christmas, right? And, you know, as the number of people mining grows, that distribution becomes more global for sure, right? Just in the nature of how easy it is to access. But I do think that the GPU mining will remain on the smaller scale, right? It's really, really hard to run 50,000 GPUs. I don't think there's that many people doing it. And, you know, especially, for the higher-end compute pieces. But I don't see that many installations around the world at that scale.

And what you may see is you may see service providers who are making it easier for the smaller miner to run their GPUs. But it's not like running Bitmain ASICs where every device is homogenous. These rigs come in all shapes and sizes and people build them in milk crates and on metal shelves or on wooden shelves and with, fans or without fans and, you know, some of the hosts out there when they look at them, they go, I don't want to touch that. So I think that it will remain kind of at the hobbyist level to this kind of small commercial level. And for that, I don't really see the distribution

changing that much, right? It's really based upon kind of what, where the home user comes out or really starts their operation. And I think that historically, that's been pretty global, and I think it will continue to be so. Love it. And for context, we're recording this on December 17th in case you want to place the conversation. Brian, this has been super, super great. How would you recommend that people follow you and Corweave? That's a good question. You don't have a very public presence. I don't. No. It's kind of by design, so I don't say stupid things all the time. I think it will be becoming more public. But we're really trying to just keep our heads.

down and execute in our business and build and do cool stuff. And I typically find that things that I want to say, I'd say to Nick directly and not to the whole world. Well, you got the chance today. Yep. Thank you very much. It's been fun. Yeah. No, thank you for making us your first ever podcast appearance. Is that correct? This is your first? Yeah. Yeah, it is. First of money, I'm sure. Brian, thanks again. Talk to you soon. Thanks, Nick.