The Truth About Crypto AI Infrastructure with Nick Emmons and Danny Sursock

Transcript

they're not actually fulfilling needs at every layer of that different stack whether it's compute, data, models, applications, etc. And that's very important. The reality is conventional markets fall very short in giving people the ability to get that exposure. So that's a real need. I don't want to underplay that. And that's really the key of what decentralized networks or crypto enables. I think it's it they are like crypto and decentralized networks are more than anything tools for instantiating tightly scoped hyperoptimized market environments for provisioning different types of resources.

>> That is sort of this new paradigm that we think blockchain can uniquely enable these these uh elements of these models that are coordinated and orchestrated via these markets. So could you guys kind of paint the picture the before and after? How do you verify and create a verification system for trust based on either economics, reputation, cryptographic proofs? >> Welcome back to the rule. This is your home to learn from the biggest guests, the brightest minds at the tip of innovation. Our goal is to help spread the collective knowledge of everyone in our community. We are breaking crypto out of its echo chamber and taking it to the real world.

>> Welcome back to AI Supercycle. This is episode 4, our premier open-source private AI oriented show presented by Near, the leader in AI in crypto. Perhaps Nick from Alloro is set to challenge that, but today we are going to say that Near is the leader in crypto times AI pushing forward with their intent with Zcash recently as well as just broad adoption for agentic transactions and they have a very very cool new private LLM feature set being launched. So I can tell chatb about all my girl problems without Sam Alman knowing. Guys, we've got an amazing lineup. We got Danny from Architect.

We've got Nick from Allora, of course. Rob, guys, welcome to the show. >> Thanks for having us. Really appreciate it guys. >> Yeah, awesome. Um, pleasure to have you guys on. Uh, tons of uh, speculation, all kinds of things that are happening in the space. Before we jump into it, uh, Danny's first time on the show. would love to get a a quick intro and uh tell us a little bit about how you found yourself in uh crypto AI. >> Absolutely. Yeah, it's great to be on the show, guys. Uh my name is Danny Sers. I'm a partner here at Archetype. We're a New York-based earlystage crypto VC. We're on our third fund, just launched that, and we've long had a very big thesis in terms of crypto's ability to play a critical role in the trajectory of AI over the next years,

decades, and beyond. Our view is crypto has a unique ability to impact AI across three frontiers really around access around critical infrastructure and around killer applications. We think we've largely solved that first bucket in a really really important way, but our focus is now on how do we enable truly critical mission critical infrastructure to take AI to the next level and as well as a new field of killer apps that make use of what blockchains are uniquely positioned to do around distribution, coordinization, monization and awesome. Um maybe we can we could start just by opening it up with the current meta, the current landscape, if you will. All of the adoption that we're h we're seeing in terms of real um

growth in the market for teams that are live, teams that are liquid and and launched. They're they're a lot of the boring stuff. It's it's what we would call the return of fundamentals, the payments, the remittances, this kind of financial engineering. and we haven't yet seen the real tangible results of crypto AI. So maybe you guys could could just kind of help us understand how where we are at the stage of development. Is it okay that we haven't seen a lot of real tangible things since the the era of you know replybot AI agents on Twitter? Um generally where where do you guys see us on on the innovation curve?

>> Yeah, Dan, you want to start? >> Happy to. I think it's a great question. A critical framing that we use from our side is understanding the difference between speculative assets and solutions. And that goes to that first bucket of where crypto can have an impact on AI around democratizing access and exposure to the upside of the field. And so our belief is that even though you might look at sort of the capital deployment and the broad market maps of all these different participants that make up what we think of as crypto AI today, the truth is most of that has been launched to solve that first need around giving people the ability to speculate and get exposure to AI.

They're not actually fulfilling needs at every layer of that different stack whether it's compute, data, models, applications, etc. And that's very important. The reality is conventional markets fall very short in giving people the ability to get that exposure. So that's a real need. I don't want to underplay that. But in terms of your question around where the real impact comes into play, part of that is just look, this is hard stuff. AI in the traditional lens is moving at light speed. We're basically playing catch-up in real time to something that is moving so quickly. But the other part is it's taken time to build trust and attract Sier talent from the AI community. I mean, we have amazing talent in crypto, but we need to bring in some of that frontier talent that comes out of AI.

And it's taken time to explain to them the value proposition of crypto rails and blockchains to give them time to to really internalize us as a credible frontier for the development of the future of AI. We're seeing a turning point right now in terms of some of the most interesting development happening at a distributed sense, both in terms of model development, around data collection, to your point, around early form factors of killer apps. I mean, the agentic field of of Q4ish of last year was not terribly interesting, but at least it was different. And so, all of that is maturing in a way where yes, we've solved bucket number one, give people access to the upside of AI in a liquid form. Now, we're moving into let's make a real impact at the infrastructure and tooling layer, and let's drive forward net new applications that are intelligent and onchain.

>> Yeah, I I I agree a lot with with what Danny said as well. I mean like I I think like a core piece of this is that this infrastructure is just difficult to build. I think like we're essentially having to rebuild the entire AI stack in a kind of decentralized form factor and that's just a non-trivial issue. I think um we obviously saw some some sort of early networks come to market in in some of the I guess like more straightforward problem spaces within the AI stack. I think like GPU marketplaces have been around for a long time now and stuff like that. I think we've seen some really interesting progress in the other layers of the stack like around like data labeling, data provisioning.

Obviously, Grass has grown to be really big, other protocols like that. I think um we're still seeing some of the infrastructure be built out and and be brought to market around uh like the execution layer of the AI stack, which is like the the kind of suite of protocols or networks that enable models to basically run in decentralized environments. Uh I think we'll start to see a lot of those start to come online, gain adoption in Q4, Q1, beyond. Um and uh yeah, the infrastructure just hasn't been laid yet. I think um agents are probably the most like immediately viable kind of sector within AI. I don't think that's what agents were Q4, Q1 of of last year and early this year. I think uh like we've approached agent development as an industry uh in a a

kind of misdirected way. I think we focused on one half of the agent puzzle, which is capital formation. Obviously, tokens are really effective at at forming capital. If you create an agent that essentially acts as as sort of an autonomous Ko or an autonomous attention seeker with a token attached to it, doesn't matter what it's doing, it's going to form capital. But once it has that capital, these agents don't do anything interesting with them. And so I think there's a a new wave of of projects, companies in the space that are building agents in a like a more productive way that can actually both form capital and then do something productive with it. That's why I think like DeFi uh is going to to be this this kind of uh like de facto uh environment

in which this kind of nextgen of of agents is going to operate. We've seen some early stuff around this, but I think that becomes the the kind of main application space or user space uh within the the cryptoi space. So, it's these things are just coming. I think the these kind of cycles to build these things out of of uh launching these things is just longer than sometimes people in the crypto space are happy with. We've gotten used to a bunch of infrastructure existing, it being trivial to just fork a bunch of stuff, making small nuances, and then resulting in this like super fast feedback loop. I think uh AI and crypto is just a fundamentally new category that needs that initial buildout period basically.

>> Habachi provable trade per with speed and security. Earn rewards. Visit hibbachi.xyz. Trading on fire. The wait is over. The Talis testnet is live. Be part of the revolution. Join now at testnet.talis.network. So Nick, on that on that frame of thought, right, we've seen the broader kind of onchain landscape and just general like blockchain infrastructure. We've gone from like, you know, you you you could call like chain centric where users are using a chain. Um, you know, then, you know, you could say that we went through like this intentric framework where was like you intend to, you know, do a crosschain transaction and then you have these solvers filling these transactions. You can even now say

that we're moving to like app ccentric where if I just want to use poly market, I don't even care what chain or what intent. I'm just I'm I'm bypassing all that. I'm just going to polyomarker. I'm just going to um you know, I'm just going to unis swap or whatever it is. >> You've you've kind of thought deeply about this objective centric AI model and like frame of thought with regards to how we should approach blockchain development when it comes to AI. Is that like a evolution of a previous thought process in in in a similar kind of model that the broader crypto landscape went through? Or maybe you could just kind of walk us through like what the the different like centric models that we've had so far in AI development in crypto, why you think objective centric is the

current and maybe even what's next. >> Yeah, for sure. I think it's a great question. Um I I guess just to start the like to define objective centric I think um it's it's useful to sort of like examine the paradigm of how we interact with AI generally today not just within the crypto space and then we can get into sort of like what the crypto space uniquely enables in in AI paradigms. I think if any of us here want to interact with AI if any developer or company wants to integrate AI into something they're building they have to do so in a very model ccentric paradigm they have to kind of survey the entire universe of AI models that exist. uh assess which is the best model for their given use case and then lock into that model and models are always changing. use cases and

circumstances around use cases are changing. And so this this like pretty heavy lift of of doing this analysis and then uh model lockin is is like a meaningful source of inefficiency in how we interact with AI. And it also results in the kind of market structure we have in AI today where this where's this kind of like hyper consolidation around singledigit number of of enterprises that have built the models that the majority of of AI consumers use. And when you when you exist in a a market that has such consolidation, you are are essentially excluding massive amounts of resources, kind of hard resources such as data, compute, things like that, softer resources, for lack of a better term, like talent, other models, etc.

And so it creates this really inefficient market structure. And so the idea of objective centric AI is instead of the consumer or the user of AI doing this work of looking for the best model locking in doing this continuously and ad nauseium is you create environments leveraging decentralized networks not too dissimilar to what we've done with intents and other systems in in kind of the the blockchain or crypto space and then enable users to just specify what they want AI to do and what they define AI doing that thing well is and then allow these kind of efficient market environments to produce the best output as a function of that that kind of objective. And so you remove all of the onus or all of the overhead from users

in needing to know anything about the models they're interacting with. Just allow them to specify what they want and then let this efficient market of models produce that. And this is like very analogous to to transactions and intents like you were saying like like before intents the only way any of us could interact with blockchains was by determining what we wanted to do and then walking back a specific set of functions or state transitions we wanted to to make on chain to achieve that effect. This is a ton of overhead. It's suboptimal often we're not the best at path finding and so it results in this again consolidation of DeFi protocols consolidation of uh of kind of environments in which we interact with.

When intents came along, it essentially offloads all of that path finding, all of that discovery to a specialized set of market participants and solvers such that users could just specify what they want to happen. This specialized market of of solvers could then figure out the best path for it. Results in way less overhead for the users. It results in way more optimal path finding for each transaction as it's going through as a fun as a function of it being coordinated by these competitive markets. And that's really the key of what decentralized networks or crypto enables. I think it's it they are like crypto and decentralized networks are more than anything tools for instantiating tightly scoped hyperoptimized market environments for provisioning different types of

resources whether that's transaction paths whether that's model outputs whether that's energy in this in the case of Bitcoin compute trust whatever it is. And so that's the real meaningful unlock in crypto AAI, I think, and and where we're going to see a lot of the most tangible benefit in in starting to uh refactor AI in a kind of decentralized form factor. >> Anything to add there, Danny? >> I think Nick said it incredibly well and it sort of alludes to one thing that we also really believe strongly here at Architect, which is there's too much emphasis in almost this lazy take around blockchains purely offering decentralization to AI. And our view is blockchains offer so much more in terms of composability and coordination, the ability to scale critical resource markets uh that AI needs to to go to the

next level in terms of giving agents the tooling that the almost like exoskeleton that they need to be functional. And so there's this incredible amount of dynamism that blockchains bring to the table. And to Nick's point, the important thing is to give people the freedom to sort of think through and express a whole range of scenarios and use cases that they want and then use blockchains to actually enable and scale those which kind of brings us to the to this you know paper that you've drafted and and put out around the benchmarks. So I think we've established here the frame of thought for how um we should be thinking about basically the intersection of these two industries on chain and kind of you know analogized it

to this intent kind of world and it you know Nick in your in Allora's design there are actually quite literally solvers in intents used in this model but you know if we're going to analyze Allora or we're going to analyze some of these networks I I don't think that we can look at something like plasma and be like, oh, look at the stable coin TVL and the amount of yield on DeFi and like think about how that, you know, ties into the token valuation of a of a given token or what that network should be worth. Danny, you seem to think that we should have new benchmarks for AI when it comes to this intersection. Maybe you could just give us the, you know, 30,000 foot view on what this looks like and then we can drill into it.

>> I think you're absolutely spot on and we sort of frame it as like there's benchmarks and there's evals. The benchmarks broadly speaking are what I was alluding to earlier which is you know give people democratized access to the exposure of AI's upside build critical infrastructure that drives forward the future of foundational work in AI and unlock net new killer apps but across every single part of the stack the evals what you're alluding to differ tremendously and what we've historically done a pretty poor job of is really thinking through what those evals are from the perspective of the stakeholders at any part of that stack and building around them. we typically end up operating in these siloed echo chambers where we build things that nobody really cares for in the real world. And so at any different part of that stack, it'll

be different. So for compute, for example, it'll be everything from, you know, latency and uptime of GPUs, the cost per token and the ability to generate them fast. It'll be around uh how can you scale up for you know latency insensitive long duration uh inference workflows. So it's all of these different things that matter at every different part of the stack. Likewise at data, you know, it isn't as simple as just let's throw tokens to get people to supply their private data. You have to think through the market structures around what data looks like. How do you build post-processing and really make the data from raw to useful? Who you're selling into? How do you think about economics posts selling into and sort of creating a durable ongoing return of value to the providers of that data. And so the point that we were making in the article was simply that at

every part of that stack, there's a set of evals that are very specific and guide whether we're building useful things, things that will actually create and capture value. And it's very very important that we as an ecosystem of participants are very clear about what those are and are building towards them versus sort of like the highle blind uh approaches that I think we've historically taken in many other parts of the stack. Yeah, it it's usually the hot shiny thing gets all the attention, gets all the the technical people to kind of flock around it and we don't really zoom out and think, okay, well, is this thing really going to be useful? Um, and and so when you're applying like Dan, maybe maybe we could go through just a fun exercise here like if you're applying those evals to Aurora, right? And and

you're looking at Aurora and you're looking, okay, how do we know this thing is really useful? How do you look at it? What are you looking for? what are the questions that you're asking of Allora the network and then we can kind of you know come to an understanding and then use these general rules to apply to to you know other projects as well. I think it's a great question and there's a few right and so first off you know maybe to to the point Nick was making around the ability to to give people the freedom to express a range of things they want to execute on chain and actually solving for you know it's not too dissimilar to thinking about how bridges work today right the things that are going to matter to those people are the ability to first off express all these different

things that they're creatively coming up with and then actually solve and execute on in a way that is reliable that is costefficient that uh unlocks net new outcomes that they historically couldn't whether that's opport opportunities to trade and be participants in a very sophisticated ecosystem that was previously off limits to them, whether it's around the alpha that they're generating and sort of their actual returns, whether it's around the time that it frees up to them. And so, it's around understanding what these people are actually looking for and then being able to create the evals around that to say yes, in fact, we are building systems that are delivering on that value proposition. And likewise, around the ideas of, you know, broader swarms of intelligence, right? one of the big value props that we have in crypto around can we sort of scale horizontally to drive more intelligent overall

systems. Again, the the approach there should be very simple, right? It should be around first off looking at the net incremental contribution from any one of those participants and then looking at overall is the system getting smarter as a result of that horizontal expansion and you can use similar evals that you would uh in a traditional industry where okay maybe it's around a specific use case of the overall you swarm intelligence that we're using in a a a specific enterprise context or a retail context or as a gaming or enter entertainment piece. Um, but in any one of those scenarios, it's just about understanding what people are looking for. And I think Nick and Allora have been somewhat uniquely positioned over the last couple years to think very granularly about like what actually matters here and then how do we meet those needs versus sort

of like highle pie in the sky ideas. >> Blockchain shortcuts, templates, and widgets launch without friction for builders by builders. Visit enso.build. The future runs on intelligence, creativity engineering health money productivity. AI promised utopia, but delivered only hype. Blackbox models and zero accountability. With tons of tools claiming to be the best, how do you know what's real? Introducing Recallfind. Top models ranked by transparent reputation from researchers, users, and developers. Test AI skills, curate favorites, and compete to see what works.

Don't guess what's good, recall it. Relay, low fees, lightning speed, and the power of multi-chain at your fingertips. Instant multi-chain payments. Choose any token on any chain. 75 plus network supported. More than 5 million users, 8 billion plus volume processed, 70 million plus transactions, fast, reliable, everywhere. Start swapping today. Only on Relay. Nick. And so, you know, one of the things that I think that we've been pushing on in this show that is probably one of the evals that we think about is, you know, we have the general kind of privacy and user ownership when using LLMs and, you know, the organic nature

of of data sets, you know, prior to using an LLM, these types of things. But one of the bigger concepts is also about verifiability. Um and and so I think one of the evals that Danny, Robbie, and myself would be looking for is how do we guarantee this level of trust when these AI outputs are probabilistic by nature? How do you verify and create a verification system for trust based on either economics, reputation, cryptographic proofs? Like what is the what is the stack that you think is the best format to guarantee trust for AI outputs knowing that they're probabistic? How does Laura approach this? How do you guys see this uh problem being solved?

>> Yeah, I I think it's really use case dependent. I think some use cases require different levels of of sort of verifiability than others like like some things will require like like uh like ZK verifiability of of uh an output's fidelity and some will require sort of lower fidelity means of verifiability. how how we've sort of approached the problem is uh in in creating a kind of environment that produces I guess like economic veriability in in that we haven't like like concerned ourselves as much with this problem of of sort of like decentralized execution of models which I think it goes handinhand with like like more of these kind of hard forms of verifiability cryptographic etc uh of these models that sort of lives

outside of the scope of the network what we've done instead is we've created an environment where the the kind of mechanism design of the network uh ensures that the the sort of best responses are being emitted to the the ultimate consumers of these outputs as possible. People are are rewarded if they're producing good responses as a function of the rest of the the sort of network participants uh and their relative reputation and they're punished otherwise. And so I think in that you create uh you you sort of subvert the need for some of this hard verifiability when consuming AI. AI is a nondeterministic form of compute. It's very difficult to verify. There's a lot of uh like great projects working on that. I think for us it's been more about how do we create an environment

where uh like one's uh returns are are based solely on on how effectively they're able to produce good responses and I I guess like trustworthy responses in this kind of environment. So people are emitting inferences on Allora. Those things are being all aggregated together and then the rewards for for those uh individual kind of model contributors is being distributed based on how how useful each of their model outputs was. We don't even care where people are running models on the network. People are running closed source models on the network. You know, the best forms of AI today are still closed source. And so trying to tackle this problem of sort of hard verifiability and also encompassing these kind of closed source models is is uh uh still I think like a fairly

difficult challenge. >> And I think to Nick's point actually eval many many scenarios are a good way that you can sort of solve for the same thing that verifiability would. In other words like as long as you're delivering on what users expect, you know, let's say it's a an agentic trading protocol, right? as long as it is unlocking the performance and the other use cases that people expect of it. It doesn't matter if you're actually running the model that people expect you're running on the back end on the exact piece of hardware that you're claiming or in fact you've locked a quant in a basement who's doing raw calculations by hand and then putting the trades on your behalf. As long as it's working, people aren't going to care. Where they start to care is are you starting to deviate from the value proposition that they promised

them. And then the questions come into play like are you cheating by using lower grade hardware or a less performant model and things like that. Now, of course, that's totally separate from much higher stakes use cases around, you know, let's say medicine or private data and things like that where verifiability does become a lot more relevant. But I do think for a lot of more general purpose use cases, you don't necessarily need hard verifiability. You could approximate it using the right eval over time. H >> how do you guys anticipate that we're going to blend some of the principles that are are alive and well in crypto and and without becoming maximalist and decentralization or that kind of ideology? there's plenty of web two style AI centric uh principles that we can also apply to as we're kind of

building out this sector that's blended of the two and you know we kind of get caught up in our ideology in the crypto world but there there's and you know to Nick's point a lot of these good models are are closed source still they they've built this without uh UX detriments super easy to use these kinds of things what what are some of the principles in in web 2 AI centric SF development What are some of the principles and the ideas that are shining through now as I think we're realizing in crypto world look we don't have to be decentralization maxis hyperlid is is is you know capturing a lot of market share people are kind of coming to this realization that it's this practical era we're building for the real user and we're kind of leaving

aside some of those you know the core ideologies they're they're still making it but like we're not maximalists about it. what how do we blend these two and and what are some of those principles that we can draw out? >> Yeah, I I I think and Danny, I'm cur curious to hear your answer too, but I I think we've done like a pretty bad job of this to date. Uh for one, I to what Danny said earlier, I agree. I think it it's been very tricky to get some of the AI industry into the crypto AI industry. I think like crypto has has a somewhat troubled past obviously and so that that's sort of marred people's perception of the space and that's contributed to it. But I think in some ways like ideology is sort of held on this pedestal as as I think this this

like I don't know kind of nebulous sometimes useless thing. I I think we operate in a like a market economy we we uh make decisions based on like a very core set of of of like uh functions. So calculus is very simple about when we make decisions. Am I going to get more value out of using this thing versus that thing? Does this environment uniquely enable me to tap into new types of of uh experiences, markets, etc. than this other thing? I think that's where crypto has shined. It's not been in in this kind of uh like religious devotion to a set of ideologies. It's been in unlocking kind of net new venues and net new verticals for people to play with. I think that was the success of like the early uh AMMs like Uniswap. It it opened

the the floodgates for sort of longtail asset classes that traditional finance couldn't accommodate. We've seen this with the memecoin cycle over the past couple of years. Obviously, prediction markets have sort of tapped into this more efficient and autonomous coordination of market participants and brought over users not because of an ideology necessarily, but because of a like a fundamentally kind of uh like new new uh feature that's born from decentralization. And so I don't know if I have a great answer of what principles we should borrow from like the the I guess like centralized AI space because I think what determines winners and losers in in this space and any space is is much more based on sort of fundamentals and and like what you unlock for your consumer base more than

it is based on on sort of any specific ideology that that's attempted to be followed because often ideologies get left behind in the face of those fundamentals anyway. >> Yeah. Danny, then maybe a more like direct way. Is there areas where you think decentralization is going to actually change AI outcomes versus just like slow things down for the sake of decentralization? Like is there a kind of a black and white here that you've carved out? >> It's a really good question like and to your point around like what we can learn from web 2 as part of this other question, right? Like we most align in theory with like the open source AI community that exists in SF and around the world. the ethos that the mission I think is very similar. The problem is we haven't approached getting there in the

same way. Right? In the open source web 2 AI community, it's still about the efficiency of the systems and how you're going about things, making sure that things work uh they are functionally advancing that mission. Whereas in crypto, I think too often we've overindexed on decentralization at the expense of functional systems, tools and products. And so for us the reorientation has to be okay we can move towards you know open- source you know decentralized uh uh form factors but the mechanisms that get us there have to be very practical. They have to be informed by economic considerations by technical considerations etc. And it also alludes to another point that we've made repeatedly which is there's a big difference between decentralized and distributed and very often in this space we use the two interchangeably when we

shouldn't be. A lot of the value prop that crypto brings to the table in how it can impact AI comes from more distributed form factors than true decentralization in like the old school protocol concept. And so for example, like if we're talking about like data collection at the edge, right? You're more than likely going to have an operator, a centralized operator who's opinionated about the types of data that they need to collect, the ways that you create an application to incentivize different scenarios and reward them. the post-processing thinking about enterprise sales and how you create economic value around that but the actual supply side of contribution is very distributed. Likewise, if you think about like a reinforcement like network uh reinforcement learning like network in crypto, right? Like the value prop there can absolutely come from

horizontal uh distributed setup where you have environmental and verifier diversity that leads to better uh evolution of more robust interesting agentic models. Um, but at the same time, you absolutely need a centralized team that is focused on the curation of those environments and thinking about tweaking things on the algorithmic side, what they're actually going to use these models to, how you bring it all together in a way that's coherent. And so the TLDDR is distributed uh value proposition from crypto makes a ton of sense. The decentralized value proposition of crypto can make sense in certain areas, for example, in economic ownership and governance and things like that. But forcing it at those other parts of the stack is what sets us behind. where the open source AI community in web 2 is like, "Oh, crypto, you could actually make an impact here."

They take a look and they were like, "None of this works. We're gonna go back to doing what we do today." >> In order to bridge that gap, we have to actually think through building things that are functionally guiding us in the direction of the mission that we do share with the open source web 2I community. >> So, could you guys give us a bit of an education here on on the way that these models are currently curated and and bringing together some of these distributed elements of the stack and different parts of inference? There's there's kind of a before and after picture here. The before is the way that these these larger models are orchestrating the these these models right now. And the after is is basically turning all of those distributed mechanisms into one coordinated market and and having all these submarkets and

and Nick maybe you can speak to the markets piece that is sort of this new paradigm that we think blockchain can uniquely enable these these uh elements of these models that are coordinated and orchestrated via these markets. So could you guys kind of paint the picture the before and after here? >> Sure. Danny, you want to start off or do you want me? >> You start. I mean, you've been building this for years. >> Sure. Yeah. I I mean like when we think about like uh the the state of like the market structure of AI today, I I think it it looks like a lot of traditional markets, right? Right. Like there is a lot of overhead that goes into sort of bringing models to market to sort of establishing uh some meaningful level of market share in the AI market. And we can talk about just the model sector as

well, but like if I if I am a a a company or an entrepreneur trying to bring a model to market today, there's a lot of of of kind of like expensive non-trivial steps to get from model development or model conception to value capture. I have to stand up a business. I have to acquire funding. I have to uh set up a bunch of administrative stuff. I have to uh productize my model often. Um I have to find distribution PMF. There's tons of overhead. It's this long indirect path from sort of the the initial seed being planted that I want to come to market with with a model or I've had some useful innovation in modeling to capturing value from it.

What decentralized networks enable us to do is is shrink that path to to the most direct and and sort of shortest path possible. It allows us to go from model development or model conception to value capture just by pointing that model at a at a given network. That's that's kind of in the scope of what we're building. This could be many different types of resources obviously, but like decentralized networks turn any industry into that kind of uh before and after. Bitcoin did this with energy, right? If if I had access if we invest had access to energy prior to Bitcoin, it was still like a fairly long and direct path to turning that energy into value. Sell it to a grid, stand up an energy company, etc., etc. Bitcoin creates the shortest path possible of turning excess energy

into value just by pointing it at the network. And so we can start to break free of this this like material inefficiency that exists in the trad AI industry. This these long indirect paths from from sort of creation to value capture by wrapping the the market environment in this kind of autonomous and coordinated decentralized network. And that's really the the most like impactful aspect of bringing decentralized networks to any industry. And and that applies very directly to like the before and after of bring decentralized networks to AI as well in my opinion. Danny, do do you see this as a common uh principle or building technique market bringing this sort of market structure to the AI landscape as a whole? Do you anticipate more projects to follow suit?

>> Absolutely. I mean to Nick's point the whole mission really that all these projects have is like the ability to take from the ground up you know a frontier model uh to market in web web 2 is restricted to a very very small set of teams that can aggregate the capital the technical resources the talent etc needed to actually do that what crypto seeks to do and what it I think has the biggest potential to do is use the ability to aggregate capital and aggregate people and resources and input and data around the world in a way that creates an alternate stack, a supply chain that helps people do that and it does so in a composable way. I think most of that has been promise more so

than actually delivered solutions. But I think we're starting to get a lot closer in having bits of that stack come together in ways that actually, you know, materially make an impact. So like even around just like compute itself, right? The compute substrate we have in crypto for anything beyond just sort of like spot or on demand inference has never been reliable enough to be a real part of the conversation. But there's a lot of projects that have actually made material strides in the last couple years in terms of building these frameworks where not just inference but increasingly you know reinforcement learning and ultimately like pre-training can be done in a distributed context. And so ultimately the goal is to replicate the endtoend supply chain stack that is very closed off and controlled in web 2 AI.

Uh but it's really really hard because we have to do it at every layer of that stack and every one of those independently is extremely difficult. Doing it in a very coherent end to-end supply chain way is even harder. So then >> how do you think the just one one quick followup Danny how how how do you think the capital formation evolves? Do do you think it's M&A and bigger AI companies coming in and buying some of these like more modular players or or or do they team up or how do you think this evolves to become you know significantly impactful? >> It's a great question. I think in web two we absolutely will see a lot more consolidation. I mean at every layer of that stack over the last couple years we've just had an explosion in the number of participants and attempts to

build companies of them. Most of them are just struggling to reach scale and breakout. And so it just makes sense that you'll see consolidation. Um in crypto I think where we have the advantage is blockchains give you the ability to hit sort of like sustainable economic scale with a much smaller base of participants. And so I think what's interesting about crypto is not just markets that give you the tools and the resources from a sort of a technical side but also like the capital to be a like a sufficient and a sustainable player at a much smaller scale. In web 2 doesn't work in crypto there's a feasible path where a very uh open and diverse ecosystem of much smaller players around this stack can sustain itself over time because of what blockchains enable at an economic level.

>> Yeah. So it sounds like the answer to this the question of you know which AI primitive or which of of the AI primitives will actually create value on chain sounds like you know we talked about agents we talked a little bit about compute but it really sounds like this coordination uh kind of problem of models of of data of incentives you guys you know seem to agree that that coordination like network if you will is really the core driver of the of value onchain for AI >> so it's uh it's when I think of it from the the infrastructure side of things I think coordination is a really big part of it but there's also the advantage that crypto has on sort of the

application frontier which is you know we're the only place that has spent the last half decade thinking through what is like digitally native rails for the creation the distribution the monetization the composability of content and assets and IP look like and so when we think about what is AI ultimately going to get used for I think crypto also has a huge advantage on that front everything from the absolute end layer for apps but also even for the agentic stuff around payment mechanisms around reputation based on cryptographic primitives for these things and new capital credit markets around uh AI in the wild and so yes I think coordination is probably really valuable on the infrastructure side but I think everything else that crypto spent years refining and developing and stress testing that's where we have a huge

advantage in terms of actually productizing AI and coming up with net new apps and use cases. Yeah, I I think that's a great point. I think um like uh the the like the value capture of AI, the real promise of bringing AI to crypto in my opinion is is in like moving towards this like utopian state of society where uh like we we achieve hyper financialization where every possible thing in society can be turned into a financial market. We've laid a lot of the foundation in that in crypto and DeFi itself. And now we have these like more efficient autonomous uh and hyper composable financial building blocks that can be combined to create ever more and more uh expressive building blocks. But the people using those the market

participants are still like quite inefficient. And so there is a a fairly low ceiling still on the level of financialization that can that society can realize. And AI is the most deflationary technology to ever exist. agents can participate in market sectors, process data sets and make make perspectives and decisions in domains that are incredibly longtail, incredibly niche or esoteric. And so I think a lot of the value is going to be captured in the like rapid expansion of the the kind of applicability of markets in everyday society. I think AI agents are going to be the the deacto user of prediction markets. They're actually going to make prediction markets like hyper useful across everything in society. they're going to start

introducing markets more into like our governance processes and uh from from local to federal, whatever it is. You know, I think like a lot of the value capture comes from just like unlocking this this incredibly proficient net new market participant and what that means for uh embedding markets into more and more of our daily lives basically. >> Yeah. Yeah. And I think that trend is accelerating and as you've as you've mentioned AI kind of p like pours fuel on the fire in a way um and kind of enables a lot of that and I think one of the examples right is the the use case for crypto the agentic transactions and all of the work that's came out the last couple weeks just on X42 and Coinbase

promoting it and we had Jeremy Aair on the show a couple weeks ago talking about it. Um, it's just been it's been a large large growth here that I think people are really starting to pay attention to. Um, guys, we're coming up on time, but Nick, I'd be remiss to not fulfill the main community question here, which is what is the state of development in Allora? Uh, I know, you know, mainet's looking ready for Q4 here. What's the timeline? What are you guys building? What can we expect uh closing? Yeah, I I uh I unfortunately can't share a specific date, but uh everything is ready and so uh it's uh it should be quite soon uh is is what I can say basically. But uh but yeah, I mean everything's ready. The network's good

to go. Um uh we're we're in the kind of like uh we're working on some sort of final migration steps and uh yeah, really excited with how things have been coming along and we have a bunch of exciting stuff planned for uh the the coming weeks and and beyond basically. Can you can you spill any like day one launch partners? Any any anything that our community can expect? >> Um I I don't know if I can uh do that. I I will say we have some uh and we'll release some more more stuff on this soon, but I I I will say we have some really exciting agent related stuff coming out around uh around that. So uh we spent a lot of time building out a bunch of stuff to start enabling millions and millions of of DeFi agents

to start spawning and tapping into this Oh, >> millions and millions of DeFi >> coming to a blockchain near you. I hope you have high TPS. >> Nick, Danny, guys, thanks for your time. Thanks for the insights. Uh, it's been very educational um and just an overall pleasure, guys. Look forward to the mainet. Nick, I I know you're busy, so thanks for sharing some time with us and great to meet you, Danny. >> Thank you, guys. Really appreciate it. It was a pleasure. >> Thanks, guys. All righty, guys.