On The Brink with Castle Island - David Vorick (Nebulous) – Building Decentralized Cloud Storage (EP.57)

Episode Date: March 25, 2020

David Vorick, the founder of Nebulous and the blockchain Sia, joins the show to discuss Sia and its latest layer 2, Skynet. We also cover David's experience as an ASIC manufacturer, and what he learned about blockchain security models and the merits of ASICs versus GPUs. In this episode: How Sia is different from other decentralized storage platforms Different models for financing protocol development The story behind Nebulous' settlement with the SEC How Skynet represents the completion of the original vision for Sia David's experience manufacturing ASICs for Sia and Decred – and his clash with Bitmain David's proposal to launch new blockchains with 'protectionist' mining in place

What's up everyone? Welcome back to On the Brink with Castle Island. I'm Nick Carter. I hope you're all staying safe and healthy and socially isolated for now. Keeping up our more frequent cadence of episode releases today, we have David Vork, the founder of Nebulus and one of the creators of the SIA blockchain. David's also a member of the Bitcoin technical community and has been so for a long time. I've known David for a number of years now, and we've had some really fascinating discussions about Bitcoin security model and, of course, the big ASIC versus GPU debate. So I thought it was about time to bring him in. So the main topic of conversation today is SIA, which is a decentralized cloud storage

system. It uses a blockchain to effectively mediate contracts between renters and providers, but the data is not directly stored on chain. And the SIA folks have just created a new solution called Skynet, which vastly simplifies the experience of using SIA. There's a few decentralized storage projects out there. SIA definitely has the most traction and the most data on the platform currently. So people sometimes write off the category because file coin has taken a really long time to ship. But Sia works. It's in production. So we get the download from David. He explains the nuts and bolts to us. We also cover a topic that I personally am super interested in,

which is blockchain security models and what the difference is between GPUs and ASICs in terms of aligning minor incentives and ensuring that they behave appropriately. We cover this idea of if you were to launch a new chain, how would you ensure that there's security at inception, how would you balance that with a fair distribution of coins, and could you potentially explore a method whereby the dev team actually creates A6 and gives early miners protectionist access to the first six months or a year of supply? To my knowledge, this hasn't ever been done in production yet, but it's a pretty interesting of novel approach to a proof work launch, which attempts to strike a middle ground between fairness

of issuance and security, as well as not violating any securities laws. So this conversation was a long time coming. I'm really glad I was finally able to get David in the studio, and I hope you enjoy it as well. 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.

of Britain's ailing economy with a new round of quantitative easing. You print a couple trillion dollars and all of a sudden people started to worry. So out of this worry, we have something called a Bitcoin. David, welcome to On the Brink. Thank you for appearing on the show. Absolutely. Great to be here. So David is a stalwart of the Boston crypto scene.

You're the founder of Nebulaus. You've manufactured A6. You've done all kinds of stuff. actually. And you're, I would say, part of the kind of Bitcoin technical community as well. Thank you. Yes, it's been, it's been upon eight years. There's so much we could talk about, but maybe we'll just talk about SIA to start. So SIA is the project that you're probably best known for, which you develop under Nebulous, your company. Sia is one of the several kind of data, I don't know how you describe it, maybe data storage projects.

that uses blockchain, although they're all sort of quite heterogeneous. They're well, and some of them don't exist yet, the popular ones. So SIA works, right? Yeah, so SIA is a, we bill it as a decentralized cloud storage platform. And increasingly we're kind of moving towards like a cloud 2.0 idea as the way that we build SIA. And so it's a place where you can upload data, download data, you can generate links that you can share with your friends. and they can access it. And the whole idea is that like we invert the way the cloud works.

So today when you put data on the cloud, it's like Amazon owns it or Apple owns it. And they can make decisions about who else can see it. Or if you're writing a review for a platform, that review is only available on that platform. Other platforms can't take your content and make use of it. And so it's really like I think not the optimal way for a certain. society to work. I think, you know, if I write a review on a product, it's because I want the world to know about it. And so it's like, it's not that I want Amazon to be able to tell the world about it. I want everyone to be able to share in that. And so SkyNet is really a push to, uh, take the data

and put it back in the user's control and put it back into into user ownership, um, so that anyone can, uh, anyone can participate and use the, you know, the wealth of information that we're building as a society. And you've been actually working on this for about six years now, right? Yeah, that's correct. I think work started in 2013 at HackMIT, I think it was the first MIT hackathon. And since then, been more or less full-time working on it and had to finish up some school. But then we went straight in, me and my co-founder started a company 2014 and have been building ever since.

So it's been a long burn, I think. Did it turn out to be more complex than you thought to build this network? Yeah, very much. When we got started, I told Luke, my co-founder, that we would have all of Sia up and running by the end of winter break. Which here we are six years later. I think when we launched Skynet, February 18th, 2020, that's when I would say we got we got to the point that I thought we would get to by the end of winter break.

So I instead of two weeks, it was six years. But we did make it. We built the whole thing. That's like that reminds me when you're like installing files or something and the estimates change rapidly. It's like two hours or, you know, six weeks or whatever. Yeah. It felt a little bit like that as we started to realize how much goes into building a disdemeanor. distributed and decentralized system.

So today there's about 750 terabytes of data being stored on SIA, correct? That's correct. Is that the most of any of the decentralized cloud storage platforms? Yeah, I believe second place is RWeave in the 300 gigabytes mark. So that would be gigabytes, not terabytes. So three full orders of magnitude or 1,000 decks behind the SIA network. Filecoin is obviously the one that people. are familiar with IPFS works, but Filecoin is still, you know, not launched. And we'll see if

they're able to. They are certainly very well funded. Yep. And IPFS, I would highlight, is not really a storage system. It's more like a sharing thing, but whoever's doing the sharing needs to remain online. If they disappear, the content becomes inaccessible. And so that's really where Saya, like, draws the line as being a storage system. is that like if you put content on SIA to share with your friends and then you turn your computer off, your friends can still get the file you shared with them. They don't need you to stick around. Right.

Yeah. I would describe Saya's effectively two-sided network, you know, where the market occurs on chain. And you have, you know, professional file stores. Do you have a name for them? Yes, we call them hosts. Hosts. And then obviously, you know, people that desire stores. Actually, it's kind of, you envision it more as a B2B to C model, right?

You don't directly believe the most end users would specifically use SIA per se, right? Yeah, kind of a lot like how AWS works, where it's like AWS, you know, most people when they use the internet, they use AWS for 50 to 80% of their browsing. Every time you're on Imjure, you're using AWS for a long time. It's no longer the case, but for a long time, every time you use Dropbox, you were using AWS. And but the end user never knows because it's B to B to C. And so we think SIA is going to be the same way. And your developers and your entrepreneurs are going to be aware that they're using the SIA infrastructure. But their users are just like using applications, using something more friendly and user-oriented.

And there actually are third parties that are building on SIA. Yeah. We have a comfortable number of products, including commercial. commercial and funded projects that are building on top of the SIE network. And so, you know, we normally take a pretty skeptical view of tokens around here. That's our, you know, that's, we have a, you know, a well advertised, you know, distaste for, for spurious tokens. So how would you defend the, the existence of the SIE coin in the system? What's the purpose of it?

Yeah, so the SIEC coin exists for two reasons. The easy sort of cop-out answer is that for our consensus system, we needed a way to enforce file contracts. And that really, in order for the marketplace to work, that needs to be automated, there can't be any approvals from like a PayPal or we can't be waiting for, you know, visa to settle a transaction. So that needed a blockchain. And at the time that SIA launched, so SIA launched before Ethereum launched. Most people don't know this. So at the time that SIE launched, the only option was to make your own chain because file contracts didn't work on Bitcoin, at least not in the way that we needed them to. And so that's like the easy answer.

I think the more involved answer is that when it comes to governance, which is something I think about a lot and I care about a lot, I think you really want every application to be on its own blockchain. because what that means is that every single user of that blockchain is unified around the same idea. And so if there's a governance crisis, something happens, you need an upgrade, there's a bug, and you have to hard fork to fix it. The impacted community is also the only community that you have to convince to adopt upgrade. Whereas if you're like on Ethereum and, you know, Maker has an issue, Maker has to convince the Auger people and the CryptoKitties people and, you know, the Uniswap people. So they have this whole campaign to go in order to get a change through in Ethereum. Whereas on Saya, it's storage people telling other storage people that the storage system needs to be upgraded. It's much cleaner.

That's actually one of probably the better defenses I've heard of kind of a many chain world. Yeah, we actually saw this issue with Bitcoin. You probably remember this. It's actually concerns another file storage company or project where I believe, it was the counterparty tokens, which were the multi-sig was discouraged or made non-standard by unnamed Bitcoin developer. So Bitcoin's always had this tension with like non-Bitcoin uses of the chain. So maybe today there would have been a way to do something like this on Bitcoin.

But if you frank, Bitcoin, it sort of disempowers non-Bitcoin uses of Bitcoin. It might be maybe for the better, but that's kind of the way it's always been. Yeah, and I think I would argue that it's for the better. Definitely the goals of Bitcoin and the goals of SIA are pretty departed. The scalability requirements are different. The risk tolerance is different. So, like, for example, you know, Bitcoin is like, at least personally, it's what I think about if, like, the U.S. government shuts down. Bitcoin's going to be the money system that, like, keeps people alive.

and so the the amount of care we have to have around a system that's like trying to serve that use case is extremely high whereas like with SIA you know people's data is at risk but you know oftentimes they have backups anyway non-Saya backups and the amount of dollars at stake per user are you know hundreds maybe maybe maybe tens

instead of tens of thousands and so I think just just those types of profiles are so different, it doesn't make sense that the two use cases would be under the same governance. And interestingly, you didn't really buy into this fad of having like a monstrously large pre-mine. So I have to congratulate you for that. I mean, SIE was effectively a fair launch, as it's called. Yeah. And sometimes, you know, I've wondered if I think pre-mines I don't like, I'm confident I don't like pre-mines, but dev fees. I've sometimes wondered if dev fees are okay. SIA has no dev fee and no plans to implement a dev fee.

But I think over the years, I've relaxed my view that dev fees are strictly bad. Well, I think they weigh on the social scalability of the system because it becomes like this very pitched debate. What do you use the protocol derived rewards for? So like Zcash is having this debate. Yeah. And then you get into the situation where like, when the dev fee runs out and the dev team wasn't properly prepared, you know, should you bail

the dev team out? Yeah. I mean, and then it becomes, again, a big governance headache. And I think one thing that I've learned running size is like you really just want to avoid governance headaches, decentralized systems don't have much wiggle room for governance. And when there's when there's a fight, there's a lot of fallout. And it's also not an unforeseeable thing that whoever is the benefit. beneficiary of the protocol finance reward would seek to enlarge that and become dependent on it.

Reminds me of this concept of aid dependency that we see in international development. Yeah, it's a real mess a lot of the time. I always thought that the way you had segmented it was actually quite elegant, not to pander or anything. But so SIA has the effectively the utility token, which is in many ways of pure utility token because, you know, you have no incentive to encourage people to treat it as an investment asset because, you know, you're not conflating the investment use case and the utility use case like many other tokens do, right? They conflate them. So it's like part equity, part, you know, token frequent flyer mile kind of thing. So you disentangled those two things. Yeah, we very intentionally, our company does not own any material amount of SIA coins, which means as a business entity, there's nothing for us to gain by SIA coin going way up in value.

Yeah, and in fact, your incentive is aligned with maximizing the usage of the system, specifically the contracts, because the asset you do have is a number of SIA funds. which is another token, so there's two tokens. So whenever a SIA coin or a SIA contract is, I guess, consummated, a small fraction of that, which is about 4%, accrues to the holders of 10,000 SIA funds. So you can think of the SIEFund as like kind of the dedicated claim on cash flows, a financial instrument that has a claim on the usage of the system. And Scy Coin is just simply the vehicle for making transactions.

Yep. And, you know, I think you were probably one of the very first projects to do this to have a kind of a dual token model. Yeah. So honestly, I think that a lot of ideas that happened in 2013 just get overlooked and forgotten because they were on random altcoin announcement. posts in BitcoinTalk.org. So I don't remember us being the first, although now I struggle to point at a project that was before us because I think most of those projects died and disappeared. Yeah, there's none remaining at least.

Yeah. But certainly when we made it, I was actually looking very much at the stock market and modeling it after the way dividends work because it seemed like a good model. overall to be using. And of course, this kind of bit us later. At the time I was doing this, I had no idea about any financial regulation. I was a computer scientist, not a business major. I guess it's funny and I don't know, ironic is the word, but you, you know, you realized that you had to align incentives properly. So you created two classes of instruments. And then, you know, as it turned out, the sci-fonds ended up resembling equity maybe a little bit too much. And which

good, right? I mean, that meant that it was the thing worth owning, you know, that it was an investable asset. But it also meant to you caught the eye of the SEC after a while. Yes. And then eventually they came after us. They find us, they find us more than the amount of money we raised in the sale we conducted. But I guess that's just what happens. Yeah, that struck me as a little strange because like the SEC has, to me, like if you look at their pattern of enforcement, they've had this pre-July 2017 era and then the post-July 2017 era with the Dow guidance being the thing that is the demarcation, right? Yeah.

And so some projects that they were pretty harsh with is because they had sold a token after the SEC made it very clear that, you know, they didn't like token sales, right? Which was ostensibly the conclusion of the July 2017 guidance, but yours was three, you know, three full years, the token sale that you effectively did out of your dorm room for the... Yeah, we did, we did do it out of our dorm room. We were still in college at the time. And again, neither of us having any business experience or taken any business classes. So it just... And you raised, uh, what was it like, 100 grand, 150 grand? I think 125 grand total. For, you sold, um, something like 15% of the, of the SIF funds?

About 12. Yeah. And, uh, and, uh, And that was to finance the build out of this network. Yep. And then in 20, it was last year that you finally settled up with the SEC, right? Yes, in 2019, which is, so the statute of limitations is five years. But occasionally the SEC will request companies sign a tolling agreement, which extends the statute of limitations. And, of course, that's never fun to do.

Yeah. But the guidance from our legal counsel was that if you don't sign the tolling agreement, the SEC will act with much higher aggression because they have to move fast. And usually, usually that's worse for you than just giving the SEC more time to act carefully. So we did, we did sign the tolling agreement. We extended our statute of limitations for them. But I guess what is, you were able to weather this because, you know, you at that point were a well-funded company. You'd grown. And the size of the raise was very small. And the fine, even though it was large relative to the size of the raise was also small in absolute terms yeah i mean you know uh there's

two hundred thousand dollars of cc fines i forget the exact number um and then and then roughly an equivalent amount in legal fees um associated so overall like it it definitely stung yeah um we didn't you know it's not like uh you know it's not like a tuesday a normal tuesday um to write over that much money. But at the, and then really like the tax on our energy, two of the company executives spent, you know, an enormous number of hours over, you know, the six months leading up to the settlement just on that when we could have been focused on, you know, the SIA network and growing our business. So I do, I do think the costs are material, but clearly non-fatal. So we made it through that without, you know, having to do any layoffs or anything like that. Yeah. And I would say that's

definitely the exception. I mean, virtually all of the other kind of token or blockchain products that settled up with the SEC. Most of them became defunct after that, especially because their primary token was then deemed a security. In your case, it was just the sciophon, which was deemed to security. Yeah. So I guess that means they can't circulate on the crypto exchange infrastructure right now. Right. Which itself is annoying. That one's not, well, it's not, it's not, necessarily the SEC's fault directly. It's definitely not the settlement's fault, but more just like it's been very difficult for the crypto industry as a whole to get infrastructure alive for trading security tokens. Well, that is, that actually does come back to the SEC. Yeah. There's some sort of key

points of clarification that they haven't granted to the would-be custodians of these things. Yeah, which I think, I think is a shame. I think we're leaving a lot of, uh, interesting. opportunities and innovations on the table by not facilitating these markets. Yeah, I think there's something like 40 outstanding applications to be security token like trading venues or custodians. So they've not exactly been that accommodating from that side of things. I didn't want to dwell too much on the settlement, but I do want to point out that that happened in like the same month as the EOS settlement.

Happened on the same day. They were announced side by side. So the juxtaposition was just exquisite. Because on the one hand, we have a relatively small project where the token sale happened in 2014, and the raise was six figures. And it happened, you know, well before the SEC issued their guidance on ICOs. And on the other hand, we have the largest ever ICO that raised $4 billion, allegedly. and settled for a tiny minuscule fraction of the total raise.

They said it was a $50 million settlement or $20 million. I think it was $24 million. $24 million settlement and had really very little in the way of punitive measures that they suffered. Obviously, Block 1 and EOS were basically completely unscathed by that. And their token sale had happened continued long after this July 2017 line as well. Yeah. So that one stung a little bit. Just like, why is it not fair?

Yeah, I think after that point I really sunk into me that they're not in the business of fairness. Right. Yep. My mental model of the SEC is like some sort of predator on the savannah, and they would much rather go after like a wildebeest that's like sick or dying or something or a young one as opposed to like the rhino that's going to be really hard to take down. Right. Yep. And I think that's the case. I think the SEC was, you know, would not, would not look forward to a hundred million dollar legal battle with EOS, which is certainly, certainly, you know, something they could budget to defend their ICO. Yeah. But I think you actually shared some of the same council. Yes, we did. So on the topic of science, so some people think that the data which is stored through the SIA network,

is stored on the blockchain. But that wouldn't make any sense, right? That's not how it works. That's correct. That's not how it works. Basically, what happens is you form a contract on the blockchain that says, you know, this person is supposed to hold this data. Here's the hash of the data.

Or more accurately, here's the Merkel root of the data. And then if they can prove that they're holding the data after, you know, this amount of time has expired, they'll get paid for holding it. it. And if they can't prove that they're still holding the data, they'll get slashed. Okay. So there actually is kind of an enforcement mechanism if they misbehave. Yes. Yep. But I feel like when I've heard you talk about crypto economic mechanisms, you're not that big a fan of slashing. Yeah. So in the case of proof of stake specifically,

slashing is a challenge because you are slashing the consensus operators. But slashing Consensus operators requires getting proof of misbehavior into the consensus system. And so you have this like the watchdogs are watching themselves. Yeah. And it's it's not a great alignment. But on SIA, the hosts and the consensus operators are two completely distinct groups. You have the minors and the ASIC minors and the hosts, they're not the same. And so to ask the ASIC miners to hold the hosts accountable seems a lot more reasonable than

asking the hosts to hold the hosts accountable. So effectively you have a separation of powers. Yes. And I guess the problem in the context of staking is the validators could just omit the proof that they're slashing if they wanted, if they were able to collude. Yes. Once in a proof of stake system, once collusion reaches a certain threshold, they completely control what everyone else sees and therefore are immune to slashing because they just

won't permit a slashing transaction onto the network. So there's no way to tell on them effectively. Yeah, that's, I feel like slashing is treated as like a silver bullet in staking as like the way that it works. But I don't know if I've ever really seen it work at scale. So, and so you actually use payment channels on SIA, which is quite interesting. SIA is probably the most sophisticated payment channel network in the world. world. Now that lightning is growing, I think we, that title is at risk and may may have been

handed off to the Lightning Network at this point. But for a long time, certainly from like 2015 to 2018, it was. And so basically we have today 300 hosts on the network. Each of them have maybe 100 channels open with various users. Some of them have more than, you know, some of them might have a thousand channels open with users. And then users do a lot. bunch of tiny payments. So in terms of raw transactions per day, I think the SIA network is unquestionably out in the lead. I think on a good day, we'll have as many as 5 million payment channel transactions go

through the SIA network. But fewer actual settlements, right? Yep. On-chain transactions. So a typical payment channel life cycle might be, you know, a channel opens than it does between 1 and 5 million transactions over the course of about three months, and then it gets refreshed, which is effectively on-chain settlement. What was the reasoning for using payment channels?

Yes, completely scalability. So one thing that we really care about on the SIE network is that everything runs on money. We think if you're providing resources of any kind, you should be paid for providing those resources. And then just because of the way the trust mechanics work, we want to, you know, we don't want the renters to have to trust the host to send them a whole, you know, gigabyte of data at once. So instead, the renters pay four megabytes at a time, you know, fractions, fractions of fractions of a penny in one go. And so the trust is very lightweight. And then we just do lots of transactions in order to move data around. So the way that

you achieved trust minimization is by packetizing the individual payments into really small chunks. Yes, and that's also why the transaction number is massive, because it's all very tiny packets. And that only works if you can compress many, many payments into fewer number of on-chain transactions. Yes. And did you come up with this design before the Lightning Network paper was written? I'm trying to remember the chronology. That was, wasn't that in 2015, maybe, if I'm, I'm misremembering? I think the lightning paper came out in 2014.

I want to, I want to say we already had this design idea in place. I think the paper, the paper was published before we had implemented any of it. Even after it came out, I think we still, for our system, we felt the Pacti's payments were a lot better. And again, this is, this is where I struggle to point back. into like the primordial soup of ideas that we were drawing from because I know you know there were lots of ideas of like hub and spoke and I'm certain I was not the first person to think of packetized payments I'm almost certain I drew that from someone else well you could have got it

from TCP IP or something right in terms of the way that that data is sent on online I do think it was it was from like Greg Maxwell or something honestly but again it's like there's so many ideas in in crypto, which just like can be traced back to like Greg Maxwell saying something on IRC. Yeah. And so I think a lot of those side ideas draw from that sort of like soup. Payment channels were already at that point, not understood, but that was an established kind of thing. Yeah. And I think there was, I think actually Mike Hearn had already implemented payment channels in Bitcoin J.

but I think we were the first system to use them heavily. I think for the most part they were theoretically interesting and never found a practical use case that gained traction until we did it. And so then compared to IPFS, IPFS is different in that like there's no hosting built in. Like you have to find out how to host your files or if you are not going to serve up the data in IPFS,

you need to establish some relationship with the third party to do that for you. But that's not a codified thing, right? That's correct. So IPFS has no means for guaranteeing the data sticks around. And I think the long-term vision of IPFS was Filecoin would do this. And so what the route that protocol labs went was that they built the file sharing network first,

and they built the file storage network later. We did it the other way. We felt that the file storage network would have the most severe constraints and be the most difficult to build correctly. So we built that one first so that we could get that right and then we put the file sharing network on top of it, which is now called Skynet. So for a long time, SIA had no file sharing capabilities

as just personal data and data storage. But today, we've grown past that to basically having full- IPFS capabilities, but also with decentralized painting effectively. So Skynet is the apotheosis of your idea. You finally achieved the quote-unquote original vision. Yes. I think this is the skynet is the completion of the original vision that I had in 2013. And then so given where you are in terms of progress, are you,

do you think, are you happy with the amount of data on the network or do you wish that it was orders of magnitude more? Yeah, I definitely wish it was orders of magnitude more. I think it will be. I think that, you know, there are a couple of big things holding us back. One of them being just adoption of crypto has been a lot slower than I thought it would be. But then another thing is that these systems haven't really been super easy to use. And so something we've really been focused on over the past four months, especially with Skynet is like, it needs to be super easy. Like developers. need to be able to write an app in an hour flat that works.

And so we put a lot of effort into making Skynet that simple. That's why we have the web portal. It's like you don't have to run a Cyanode to play around with it. Just use a web portal. And I think that's actually already paying dividends. Skynet is growing at a phenomenal rate. On launch day, we had 7,000 files total, go up onto Skynet over the public portals.

Today I think there's something like 120,000 files, so we're less than a month after that. And we're already doing more activity per day than our massive launch day. So I do think that 700 number is going to start growing very fast. Yeah, you can see on the chart on Sia stats, you can see the acceleration. Yeah. So you have 700 terabytes being stored The actual chain state

How many gigabytes is the blockchain? So I think right now it's about 30 gigabytes So we are So it's about the tenth A tenth the size of Bitcoin Yeah and we're actually working on A technology called UtriXO Something that Tadj Jaija invented for Bitcoin

As it happens Saya is also a UT XO model. And so most of the work Tadge is done translates very well to SIA. And so that would allow us to bring the chain size down substantially. And then another thing that we can do, because we're a storage network, is we can actually move the chain state from your disk to Skynet and just let you pull the chain state off of Skynet. So because the data is not, you know, literally being stored on chain, you can have an enormous ratio between the total. like data stored through the system and then like the actual you know chain state that you

have to ingest or on a node yeah and so I think you know a lot of a lot of things like sync time and you know disk storage and all this stuff we will be able to optimize substantially over the next few years we're still not quite focused on that yet we're still focused on the usability aspects of of skynet and making sure that you know the renter system is is really clean and easy for users. But we're very close to refocusing, I think. So what do you make of this kind of much more naive approach

where people will just chuck data in the Bitcoin op return and they want to take advantage of the nature of Bitcoin is this very transparent and highly available online database that you can in theory use, you can use it for all sorts of different purposes? and then other other blockchains, I would say, you know, Bitcoin Cash and BSV took this idea and to its sort of logical conclusion and they began to advertise themselves as a way to store lots of arbitrary data perpetually in a very available way.

It seems like they basically seem to think that there's a free launch here. Yeah. I mean, it's always tricky with a live system like Bitcoin because, I can I can like postulate what the social contract might be but the truth is that every user who's using Bitcoin has their own reasons that they're using it and really like Bitcoin has to match that. When you're using something like Bitcoin as a storage platform you have to remember especially because it's for free a lot of the ways that people are using Bitcoin for storage are not consensus critical so we could in theory rewrite

Bitcoin Core to just ignore that data, not store it, never serve it, and instead just provide, you know, whatever, whatever Merkel branch, and then the nodes can still fully validate because it's not actually not store set. Not store, not store operturn data, right? Yeah. Yeah, totally. And then suddenly your full, you know, your maximum availability solution is not, is not maximally available. And the users who are omitting your operturns are unaffected, right? They have no reason to keep it around. So I think we have to keep in mind that kind of stuff is possible. So what you're saying is there's no alignment between node operators and people that are putting this externality on the nodes of making nodes full of potentially junk data. Yeah. And I think if it becomes a significant

issue, we'll see work in the direction of finding ways to prune that junk data. And that junk data will lose its reliability guarantees. Right. So effectively, the system, which is being advertised as immutable, were it to become too burdensome on nodes would lose the immutability. Effectively, node operators out of their self-interest would simply just not store the arbitrary data. Right. And this would, most importantly, would not impact the safety of the Bitcoin financial system. And so, you know, the financial system would still have this enormous redundancy. And so this obviously applies to the clones of Bitcoin, which make this part of their value proposition. Certainly.

Yeah. And, you know, in Bitcoin, I wouldn't say it's a problem. So it's only about three gigabytes of data in the offer return versus 300 gigabytes total. Yeah. Actually, a bit less. So it's even less than 1%. And if I'm going to like kind of just project a little bit, what I understand the social contract to be, which again is like, there's no formal social contract. there's no terms of service but my my interpretation of it is that like when you run a bitcoin node the

bitcoin node says i will never cost more than x to run you're going to need to add this much storage per year it's going to consume this much bandwidth this much CPU and so you have these you have these resource cost ideas and my opinion is that as long as however you're using the bitcoin blockchain stays within those resource costs no full node is going to be mad they are they already signed up to they're already subscribed to that sort of resource requirement. And so if you're outbidding other people for chain space, like more power to you. The way I see it is like full nodes are interested in cultivating UTX O set

and they're not interested in the other stuff. And if you try and make them do a whole bunch of other stuff, they'll find a way to resist that. And operative outputs are unspendable. So they're not part of the, well, I guess they are UTXOs, but they're not spendable. So maybe they're not even UTXOs, I don't know. Although what's interesting is people used to embed arbitrary data in Bitcoin with these kind of malformed addresses. So if you look at there's in some early blocks, there's like Ben Bernanke-Askey art.

And that's encoded in addresses as opposed to being in operands. So people did find creative ways to put like human readable information into Bitcoin. And I think you always will be able to. I mean, even at the Bitcoin core developer level, you have people like Tadge proposing, you know, hey, in our signatures, we have this random value that we need to choose. What if instead of a random value is like storage? And typically they have, typically they're doing like an atomic swap or something. They're like doing something fancier than just storing ASCII art.

But if you can do it for an atomic swamp, why not also do it for Asky Art? So one thing that you've done, which is quite a dramatic thing to have done, you became an ASIC manufacturer for a time. So you had a subsidiary of nebulus called Obelisk, and you were a U.S.-based manufacturer of ASICs for both DECD and for SIA. Yeah. And what a crazy decision that was. Yeah, that was quite a journey.

I mean, you actually sold like, what was it, north of $20 million worth of basics. Yeah, we sold almost $30 million worth of basics. And then from what I recall is quite a turbulent time. Like Bitmain also made A6 for SIA, correct? Yep. And so did it in our silicon. Both of them also made A6 for Decred. So you announced that you're making A6.

It was kind of like a Kickstarter style. And, and, but you're, you're pipped effectively by, by Bitmain who fired up their infrastructure to make some before you get to market, right? Yep. That's correct. Um, which I guess is like, there's a real lesson in just how anarchic and wild the, uh, the mining markets are, right? Yeah. Um, for sure. And so now you're, you're out of the business of, uh, of making A-Sex, I guess.

Yeah. Um, I do want to clarify one point. the nearly 30 million in revenue that we made ended up as a net loss for the company. So I think a lot of people are like, oh, they made a ton of money and they're like bad cats now. And then like that never happened. Making A6 is super expensive and the number of things that we had to do to ship on time. And even then we were a little bit late, definitely sapped all the revenue. So was the motivation to become an AIC manufacturer, was that because you just wanted

sire to be as secure as possible, you know, because A6 have this nice property of incentive alignment, or was it you just thought it'd be a fun experiment? It was definitely not just for a fun experiment. There were two primary motivations. One, and a big one was like, when we kicked off Obelisk, this was around 2017, when the Bitmain drama in Bitcoin and ASIC boost was reaching a peak. And so Bitmain at that point controlled basically every altcoin and Bitcoin. and it just looked like we were going to be living in a bit main world.

Our understanding of how mining works, the game theory behind proof of work, all of that was a lot earlier than it is today. And so there's also a big sense of like nobody knows what they're doing. Like the mining was this giant black box that we had been trying to get a sense of how it worked. And it was like someone just has to break in to the black box. So we did it first and foremost to protect Sia. We didn't want Saya to be a bit main monopoly coin. And then secondarily, we did it just to understand the ASIC world better.

And overall, both goals were achieved. I think now that we know everything that we know, in hindsight, I don't think we needed to create obelisk in order to ensure that Sia would not be a Bitmain cryptocurrency. But at the time, we didn't know, and you have to remember that, like, things, things that seem obvious today were just completely off the radar and would have felt very bizarre being said in 2017. It is interesting how far we've come.

I think a lot of people were super indexed on ASIC boost as the reason why BitMain was blocking Segwit originally. That was kind of like the thinking man's theory for why they blocked. blocked it. My pet theory is that effectively, Bitmain just wanted to demonstrate that they had a veto power over changes to Bitcoin, just as a desire to remind everyone that they had a say in governance, as opposed to any specific quibble. That's kind of like with my benefit of hindsight, that's my theory about it. Interesting. I think I'm still on the, they were protecting ASIC boost side of the But certainly, like, Bitmain proved itself, especially in 2018, to be a very emotional company.

So I think these emotional theories align well with other things that we've seen out of Bitmain. Yeah, because my view on that is like it was an ultimately financially very costly thing to do. So you'd need like a darn good reason, basically. And I think we estimated ASIC boost to be worth to Bitmain well over $100 million a year. really? Yeah. So it was quite material. It was quite material to protect ASIC boost.

And so there's also been this long running debate, which you've taken one side of quite vociferously over whether GPUs or ASICs are better for the health of a network. I think this debate, like, it's kind of almost been settled, actually. So you're on the side of ASICs being better for incentive alignment. Do you want to briefly go into why you think that's the case? Yeah, and I think there are like two key killer points for GPUs. So the first is that ASICs are better, even if we could have a GPU mind coin. But the second is that you just can't have a GPU mind coin.

Specialization is too effective and engineers are too good at specializing things. And so any GPU algorithm will eventually be ASICT and it will be ASICT to a high degree of efficiency, just because you're playing with the laws of physics. Engineers are very good with physics, and they'll figure out something you didn't expect. But going back to the first thing is that you really want your consensus builders to have skin in the game. You want them to care about screwing up,

and you want them to make sure that they're running consensus correctly. And then if they don't run consensus correctly, you want to have some tool to punish them for it. And if it's all, if the whole world is GPU mind and, you know, the SIA consensus screws up. You know,

someone does a double spend or something. And so we go and we try to punish the consensus builders. We change the algorithm or whatever. The GPUs just hopped to a new coin and they notice a tiny dip in revenue. So it's not that big of a deal. The loss of revenue is basically equivalent to the market cap of SIA over the market cap.

in this example of all the GPU mine coins. Yep, that's correct. That's small. Right. Especially with the other big GPU fish in the ocean. But with A6, if SIA has one machine that mines on it and that machine can only do SIA, then your consensus builders are very incentive aligned with keeping the SIA community happy. If they do, you know, if they start failing at their job and the scia community gets unhappy,

this fork represents a massive material loss to the consensus builders who are essentially sitting on a cash cow. And so, you know, we like that. We like that the consensus builders of the SIA network have to care about what the SIA team thinks and what the SIA community thinks. And they have to look out for forks, look out for governance crisis. And they kind of have to have to mediate everything that's going on. that's part of their job. And they care about the exchange rate of the coin as well, right? Yes. Because their ASIC, the value of that ASIC is just the discounted set of cash flows occurring to it over the depreciation, the useful lifetime of the ASIC.

That's true. However, I think that if you migrate to a fee dominated world, which ASIA actually does not have to, but I think there are a lot of, there's a lot of discussion going on about, like what happens if chains were primarily funded by fees instead of inflation? I think in that world it actually the miners don't care so much about the coin price because the demand for on-chain fees is, well, in theory, independent of the coin price. Yeah, I guess it depends on-transactors are thinking of fees in like Bitcoin terms or in like dollar terms. Right. In the early days, fees were just like homogenous. They were like, I think there was a time where fees were just 0.01.

BTC was the default fee. Yeah, I think that's not the case now. Yeah, so which side actually, because you kind of noticed one of these people that reasons a lot about security of blockchains. What do you make of this argument that Bitcoin is going to need to add inflation or something? term. Yeah. So right now I think I would lean towards the side of Bitcoin probably doesn't need it. I currently expect that a fee market will develop and it will be large enough to sustain Bitcoin. I'm 100% on the side that we shouldn't have this uncertainty. It's not something we should be depending on or betting on. So I think that, you know, if I could go back in time and convince

Satoshi to put you know on infinite inflation into Bitcoin I would push really hard for that despite that being a big part of the early draw for Bitcoin Really? So just a stable rate of issuance in perpetuity? Yeah, or maybe

You know it starts high It starts at 50 bitcoins per block drops down to like five Bitcoins per block Over time and then once it's at five Bitcoins per block it just stays at five rather than going to zero that would be if I could go back in time. I think I would certainly argue as much as I could for that.

Now that we're here today, I think it's too much of the Bitcoin like social contract, too many expectations of node runners or that Bitcoin is fixed at $21 million and that the issuance will run out and that Bitcoin will transition to a fee market. So I think that ship has sailed and it's not a fight worth attempting to have. and if you if you want infinite issuance you should just really make another coin and try it on that yeah or just adopt any of the other more inflationary coins yeah right i mean i i see bitcoin is something like fundamentally new and that like it's it's truly hard hard money uh you know it's it's got this supply inelasticity um you know and and and to most people that's like very contrary to what they want and what their intuition is and uh so they they rebel against this idea and maybe they create inflation by creating lots of altcoins

or they create outcoins with lots of inflation. So Bitcoin kind of offends the sensibilities of people in that way. So I think it's useful that we have at least one that's like that. Yeah. We have plenty of inflationary systems, you know. That's fair. Dollars, one example. So back to the topic of ASICs.

So you actually have talked about this. You had a blog post on, on a proposal to work with developers of a coin and give them kind of a temporary monopoly over hash rate in the form of A6, which were tuned to a specific hash function that only the launch team would have prior knowledge of, which is a very clever idea, I thought. and the idea was you don't want the network to be monopolized by some unknown group of ASIC manufacturers right away, right? Yeah, so I think the term I would use here is protectionism. It's like when, so if you don't do this protectionist

like dev-controlled launch, what you get is you get bitmanned. And if it's not bitmane as some other company is going to rush your coin make hardware for it going to get you know a complete monopoly over the mining hardware and it's this this unknown quantity and so now you have uh you have some unknown entity with a6 and and there's because they exist it's not a green field anymore so it's much harder to justify development nobody knows how the miners work and so and so you get this like just this black box miner sitting on top of you and maybe they're good maybe they're evil but it's this unknown entity and i think much better is to bootstrap your coin in a protected way, you know, have the developers make something

that's open source, accessible. Everyone can kind of see how the base model worked. And then for like, because that costs a lot of money, give them, you know, protectionist access to say the first six months of block rewards. And then, and then after that, open the, open the field to everyone. And so it's kind of like, you know, I think like raising a child, you don't, you don't send a four-year-old out to the store to buy groceries all by themselves. You really need to protect them. And like, yes, it's, it's, you know, it's not the true nature of decentralization. It's not fully decentralized, but it's also only to get things bootstrapped. And then once, once you know the coin has legs, it's old enough to like kind of survive on its own. You've got this nice open source starter

kit that companies can use to launch from. Then you can open competition. And I guess, I find this idea elegant because, first of all, it requires and it supposes proof of work, which I think we are both fans of. So it also allows the protocol team to potentially monetize their launch on a one-time basis. The developers sell, they could sell the A6s at a premium. And so monetize the launch, but without extracting a continuous flow of funds from the protocol, which as we talked about before, like leads to all these difficulties and arguments and like social scalability problems. So I think that's good. And, you know, like maybe a one-time, you know, payoff is a good endowment for it to support developers, you know.

Also, the early advantage decays over time, right? So whether the developers retain the A6 or whether they sell them or give them away, those A6 depreciate. And if the coin is worth something, other companies would just make other A6. And difficulty rises. And so the initial A6 lose their advantage after a period of time. It also does something else that's really nice. It's something that Grin kind of ran into, which is that it guarantees that the coin has A6. And so Grin had more than enough attention to get A6.

But they split their block rewards so many ways. There were so many paper tigers and uncertainties. and all these manufacturers playing mind games with each other and attempting to fake each other out, that what actually happened was nothing got funded at all. And so Grin is now without A6, and there are no A6 on the horizon because the algorithm is extremely difficult to make A6 around.

You need an expert team, and that costs money and costs time and requires certainty that Grin kind of sacrificed. And so as a new coin, if you're not doing this protectionist thing, you're kind of gambling that someone's going to care enough about your coin. And for many coins, that's not a very big gamble. For many coins, it is a very big gamble.

It's not a guarantee that someone's going to drop $10 million in private to R&D in ASIC. And so you can kind of remove that uncertainty with this game. If investors are a lot more confident funding in ASIC, if they know that there will not be competition for the first six months. And the risk is, if it's a GPU, coin then you have the situation where miners just exploit it they have no incentive to be good stewards of the project right yeah and you get into this like you know the double spend problem where the GPU if especially if you're a smaller GPU mine coin you have to worry about nice

hash attacks where someone spins up a ton of hash rate double spends your chain and then makes a ton of money and we this we see these it happens on a fairly frequent basis um happens on vert coin Bitcoin Gold, Ethereum Classic, several other smaller coins. So this is a real problem. It's a real thing that happens. So having A6, the point is having A6 is something that virtually guarantees, I would say, that the miners don't misbehave. Right. And I mean, you also have to be willing to follow through with that.

So if you have A6 and your miners misbehave, I think you're kind of compelled to fork, break all those A6s and kind of demonstrate to the miners that. You are serious about them needing to be good stewards of your community. So you need to be able to credibly threaten that you would brick the miners, the A6 of the miners. Yeah, I think it's very important that you have a credible threat of removing value from them. And there's also a securities angle to this. So one of the best things to me is that this allows the developer team to disentangle the investment asset and then the transactional. you know, unit. Kind of the way that you, you had the same intuition back in the day about disentangling the two.

But so the ASIC is like a financial product. You know, it's a claim on some accumulated set of cash flows for a period of time. And then the coin can just be issued as a fair launch. So that means there's no senior age, really, with the coin, right? The coin is issued whatever the cost is. You know, so leaving aside the initial six-month period and perpetuate. the coin is then issued at market price and nobody gets a discount, which to me is a very nice property. And then the ASIC itself is the thing which is considered being an investment asset.

So I don't even, I don't know if any securities lawyers have waited on this, but you're of the opinion that the ASIC itself actually wouldn't be a security, right? Yeah, we're fairly confident. And again, I'm not a lawyer myself, but that this would fail to how we test, which means it's not a security. Because the key element there is that the dev team is selling the ASIC, not running it themselves. And so something becomes a security when you're not putting any work into it. The original case was like these orange trees where the company selling the orange trees also grew them and took care of them and did all the work. But on ASICs, you have to run it yourself. You have to find hosting for it.

You have to do all the maintenance. and because you have to put work into it and your investment success is dependent on your own material efforts, it's not a security in the eyes of the SEC. So I would suggest, of course, I don't have a proper legal opinion on this. That's fairly compelling to me. I think even if it is deemed a security, I think it's still an attractive model. Yeah, regardless. Because it does free the, the, like,

actual native unit of the cryptic currency from the burden of potentially being declared a security and then you can't trade on any exchanges. The biggest issue with this is that GPs have the very nice property of hobbyists being able to mine a small amount of the coin if they want, right? So lots and lots of individual hobbyists that had GPs for gaming mine Ethereum and that was their on-ramp into Ethereum, right? And so I think that's like a pretty valid complaint that you're totally narrowing the set of people that can actually get access to the coin in a permissionless way. Yeah. I'd agree with that. I think in a perfect world, A6 would be super accessible and you could just like take any set of, you know, $200 and poof,

turn it into a fully competitive at-scale ASIC. And of course, that's, that's not the reality. So I think A6 are not a perfect solution, especially because of how difficult it is to make a competitive one, you end up with these manufacturing gatekeepers that in the ideal world these manufacturers don't exist. But I still think that given all the other options, this is the best thing that we know right now. Yeah, it's really weird because for this Goldilocks zone of this sort of ideal launch method, you want A6 to, be much better than GPUs for the equivalent task, right? Because you want that incentive alignment. But once you have that, you don't want there to be a way to massively improve on the

other, on the existing A6, because that implies empowering like a single entity. Yep. But there's really no way to tune it, such that both of those things are true, right? As far as we know, I'm hoping at some point we figure out some neat tricks. but certainly I couldn't tell you what they are or where they would come from. But to me, like, proof of work is still, you know, people say it's obsolete or whatever. I think we still have a lot to learn about proof work. I think it's one of the most powerful tools in our arsenal, especially when it comes to distribution, you know.

There's so many folks that have gotten secondhand A6 in Venezuela and use it to mine Bitcoin, and that was their on-ramp to Bitcoin. And it was permissionless, they didn't have to, you know, use an exchange and docs themselves. Like, that's a really nice property. Taking electricity, which is available globally in little pools of strand and electricity, and turning it into, like, this currency. Like, that's one of the best things about Bitcoin. Yeah, I agree.

That's kind of magical in a way. Yeah, that's why we're really lucky about the circumstances of Bitcoin's launch, that A6 didn't emerge for a long time. do you think that there is a alternative is there's a way to tune an algorithm such that accessible asics can be built like ASICs that don't require such high fixed costs maybe yeah i've um i've done a moderate amount of research into it i think ultimately it wasn't as groundbreaking as i would like something to be that i'm working on i mean frankly frankly i think skynet is like just incredibly world shifting. And so that's got all my attention. But when I was looking at ways to design proof of work algorithms to make, you know,

increase fairness, what I landed on, super surprisingly, is memory hard, which for a long time was discouraged. But I think it's my preferred way to go. I think Grin almost, the type of chip that you make for Grin as an ASIC is in the right direction, with the exception that the algorithm itself is very complicated, and so you need a lot of theoretical knowledge to make an optimal chip,

which you don't want. That's a bad property, but like, Grin is super S-RAM heavy. S-RAM, as it were, is cheapest at the 28 nanometer level, which is a lot more accessible than 16, 14, 12, 10, 7 nanometer. And so it's like, and then where seven nanometers is really effective is an energy consumption, but grin is so, like memory consumes very little energy and consumes a lot of dye area. And when you talk about purchasing price, a dye area is, is dollars. Like when you buy, you know, an M20 or an M20S, you know,

60% of the money you spend is just on die area alone. And then everything, you know, all the rest of the stuff is the other 40%. So including margin. And so for the Grin miner that we made, the economics were more like, you know, 90% or 95% of the money you were paying was for diary. So we had a machine that looked very similar in size and energy consumption to the M20s. But instead of being $2,000, it was $20,000. And that's just because it had the same number of chips and the chips were five times as big each. And I think that this is potentially a way to level the playing field.

It makes it much more about doing chip areas, as possible, which I think is an easier challenge than doing energy efficiency as much as possible. And the intuition is that a broader set of entities can compete at this challenge so you don't have this total concentration in terms of who's manufacturing these things. Yeah. And there's a, there's like a fundamental design tradeoff that I think is a lot easier as well. It's just DSMC, when they go down. to seven nanometers, they're mostly concerned with the physics of making seven nanometers work because they're trying to save power. Because, you know, that's, they're trying to save power and

they're trying to improve speed because that's what their clients want. But with cheaper S-RAM, if it doesn't make a cost sense to drop down to 16 nanometers, you don't have these other other issues pushing you. And so you can just stay, you know, stay at 16 nanometers and focus on making that cheaper. So that's kind of where I landed is you probably want a super simple algorithm. Like it should, you know, it's something, you know, a sophomore electrical engineer could implement themselves in varilog. And then you want it to be very memory heavy. And I never, I never ended up with such an algorithm. I actually think that hash is like pretty close. You'd probably start with that hash and find ways to make it better. So that's, that's kind of the direction I would

push. And you'd want, you would want to aim for an algorithm that consumes maybe between 20 and 50 megabytes of RAM. So that's another place where Grin really made life hell for ASIC designers is they want a gigabyte of RAM or a gigabyte of S RAM, which is too much. And that puts you squarely into like world class engineer territory, whereas 50 megabytes is like, you know, most chip teams could figure out how to safely get 50 megabytes onto a 28 nanometer chip. Well, I'm still hopeful that someone comes along and designs both the hash function and a model for a launch, which kind of satisfies these qualities that we seem to like.

Me as well. But that does require that there's a coin worth making. A lot of Bitcoiners would reject the premise here. So, David, this has been really fascinating. Where would you direct people to follow you, follow your work? stay up to date on SIA. Yeah. I mean,

so the easiest thing is our website, which is saya.com. And if you're interested in Skynet specifically, that's Sia sky.net. And then if you want to get involved in like the day to day, I would send you to our Discord, which is discord.g.g slash SIA, and then if you want to follow me personally,

I think most of my interesting content comes out on medium. So I have a medium account that you can follow. I blog a bunch. Blog.com. It's a good way to get to my medium account. And then I have a Twitter handle. I'm not super active on Twitter,

but occasionally I poke my head out and say something interesting. And David has written some absolute bangers of medium articles, so specifically on proof of work, on A6, extremely good stuff. So thanks so much for coming on the show.

Awesome. Great to be here.