Delphi's AI + DePIN Year Ahead 2025

Transcript

[Music] hello everyone and welcome back to another episode of the deli pod this is the last episode in our series of the Year Ahad reports if you haven't listened to the other ones they're all in the feed we did gaming then markets then infrastructure then defi this one is going to be focused on what is the hottest Topic in crypto right now and definitely the most requested report since we asked people what report they wanted to read this is the de and dein one a lot of dpin things kind of have to do with compute so a lot of them fit into the de AI bucket and then of course there are some other stuff like Wireless and energy that we also touch on but

basically just a whole coverage of this sector uh we did do a four-part series on deai that is only a few months ago which is all fre for anybody to read and then this one goes a bit deeper into some of the newer stuff like some of the agent platforms and agents that have been happening since then and some other areas I mean this this space moves so fast so I mean even though we did a whole four-part series two months ago there's a lot that's happened in that two months so maybe we can just start it off here I've got PD pondering durian he kind of has led the whole AI stuff here at Deli and then we've also got Michael rinkle from Market team who helped out on this and then we got Robbie Peterson from the defi team who helped out on

this as well so with that being said uh I think I want to kick it to PD and maybe we'll just start with like you know high level AI where we are uh kind of like macro overview yeah so so first of all I do want to say it was very much a joint effort so I I do want to thank uh Robbie and Michael for for joining the team and uh you know spitting out 130 pages of what is hopefully entertaining content um but yeah I guess I I'll kick off with the macro and then obviously Michael I know this is near and dear to you so so please jump in but yeah I guess all eyes are kind of on the scaling laws and you know I think a lot of conversations in Silicon Valley are hey have we plateaued you know and and I

think the the general consensus if you watch Ilia talk is that you know pre-training there is very much reason to believe that some of those Tailwinds are starting to slow and we might be seeing a plateauing along that Vector um but I'd say you know is definitely not the only tool in the toolkit and so there's a lot of other you know reasons to be bullish even in the event that you know that starts to slow and frankly we're not even sure if that's going to slow down I think you know grock 2 being trained in Memphis is the first 100K h100 cluster that should be you know coming out of production uh from that training run like quite soon um and maybe that kicks the door off the hinges and we're just like full on in an

explosion of capex um but you know in the event that that doesn't happen I do think there's a lot of other reasons that performance will continue um and so I think from synthetic data to reinforcement learning with AI feedback H and then to kind of the new trend which is you know more inference compute like basically letting models stink for longer and we've kind of seen a preview with uh version 01 and and maybe you know it sounds like there might be a big announcement coming out of open AI you know basically as we speak um but yeah and and then you know also agents as well um so that's kind of the the conversation um that that at least we've been paying attention to um and it does seem like there's you know

areas to be a little bit concerned if you're Nvidia or you're you know a a GPU provider that's like really been a beneficiary of pre-training but if you're you know one of these other bottlenecks that's starting to emerge um and I think one that is particularly bullish for crypto in AI is kind of if data becomes a bottleneck free synthetic data being cracked then you know I think there's a lot of room for opportunities uh with those type of plays so yeah I don't know if you guys have any views but happy to uh to to debate in real time yeah I I think you hit the nail on the head there I guess from a markets perspective two of the things that we wrote about in our kind of macro AI section are you know I think a lot of

people are looking at AI right now and they're like wow like you know Nvidia ran up a lot this year paler is up a lot some of these other AI you know you look at crypto all the AI names are up a lot it feels a little frothy it feels almost like a bubble so people are wondering if at some point you know does that pop you know the way that we think about AI you know I I I think AI is like a little bit different than past technology waves so like if you look at you know mobile or um social media kind of being like the two most recent Computing waves a lot of the wealth that was created like within those Cycles was uh centralized so like you had the kind of like app developers the engineers that sat in San Francisco that benefited

and then you had kind of the influencers that were in LA or New York that disproportionately benefited from those waves and I think the interesting thing about AI is that the wealth is a little bit more spread out um and that's mostly a function um of just AI requiring a lot of physical inputs like yes it's like a digital technology but even build it um you got to get the electricity you got to get the chips you got to get all the interconnects all the hardware that goes into like even making AI um so kind of the the wealth that's already been created from this kind of huge wave has been a little bit more dispersed uh the past cycles and I think that's that's that's been reflected in just the like

broader economic performance of the US I think um I think this stock market's also reflecting that um so that's kind of an interesting like like thing that makes this um kind of bubble I guess if you want to call it like feel more sustainable the other thing I think that's really interesting to watch that that we wrote about is the question of like when will nation states really get involved in this race um people have been predicting that like AI will be um kind of this like game theoretic race between uh Nations for decades now um to be honest we've like we've yet to see that um I I think we still kind of think that's coming it's a it's just a matter of time but the US is like the clear

front runner on just about like every metric we have the most you know Frontier Labs the most data centers the most kind of research scientists so we have this huge lead um Europe is kind of still in this like very much like we're going to regulate first and innovate second mindset and then I think the big elephant in the room is China um and everyone's kind of looking to see like when does shiin ping get scaling pilled when do they start building data centers at the like speed and scale that they were building hospitals during Co why do you think it's taken longer than expected to see all the nations get into this Armes I don't yeah I don't have a good answer for that that's something that I thought about I I really don't know I I think part of it is just like

politicians don't fully understand this technology yet and if they do they're a little scared um because similar to crypto AI is like taking power away from these centralized kind of entities and now everybody has chat gbt in their pocket you don't need to go to the government for an answer you can just ask chat gbt um so I think that might be part of it but PD what do you think yeah I guess my my answer is pretty simple is like you know it's just really expensive to even be a part of the race and you know the hyperscalers that are involved like a lot of times their market cap is you know much more sizable than the vast majority of countries um and so you know it it is implausible that the vast majority of countries will be able to

play a race that is so Capital intensive that you know only you know so far five or six companies on Earth that are truly you know nation state scale the vast majority of them are in the US and have been dominant and dominate you know uh market caps of major indices these are just huge companies and they're you know throwing off 1.5% of uh basically GDP and operating cash flow and have you know hundreds of billions on their balance sheet and that's just something that most nation states cannot replicate so yeah I I do think China could you know put up a credible Contender just because they have the resources at a nation state level to do it and the capability and the talent um but outside of those two I think you know

realistically it would be difficult for others assuming it remains you know basically a capex type of type of um competition that that other players will be able to keep up yeah okay so I think that's like pretty good macro overview uh maybe we can get more into the Weeds on some of the crypto AI stuff I know you know we did this whole four-part series before but that was like right before all the agents and everything really started taking off right and so we've seen a lot kind of go on since then I kind of want to have a discussion on like the agents the agent platforms what people kind of like think of all this for me it's kind of like the more I look into it the more I get convinced on

just the platforms or Frameworks and I know this is a very you know more of a risk averse I guess standpoint but I mean you've just seen a lot of these Frameworks and platforms do very well right like virtuals was around like 100 Mil I think when goat launched it hit it tap three bill um AI 16z had like some weird stuff going on early but they seemed to work through that they got close to a bill you've had this new arc framework come out uh and then mean meanwhile you've had like a lot of these agents especially agents on salana I found is very hard for them to crack through like the 10 mil market cap range even if the tech is interesting right I

think that you know there's there's two that we followed a bit right of course like NFA on this these are very low market cap but like rorito and te right these are both unique agents uh the tech is cool but it's like the no one really wants to buy the token I know te it's a bit more nuanced because it's not like officially endorsed by both teams right so there's a bit of that there but then you see like fcoin go to like 1.2 bill right uh which is again just like a pure Meme and I think a lot of people are excited with this meta that we were shifting away from like the pure meme aspect of it and going like a bit more toward the tech play and then you know

you just it's hard to see a lot of these agents even like Luna on virtuals is hasn't really gone anywhere for a while right but you've seen like the game framework do pretty well um you've seen obviously virtuals token do pretty well so it really feels like platforms are the things that people are interested in now but I still do think it makes sense I feel like we we've got Duncan on the podcast just a classic pixs and shovels type guy setus I mean the thing is the thing is with this sector I just think it's going to get very big and so I think pix shovels can be boring if you're going to like an established sector but these pix and shovels were they're not like as cheap anymore but they were pretty cheap early

on right so maybe like the agents go crazy late cycle or something unless unless the Cycle's over now I'm curious here Robbie what you think here because you're the you wrote the the Bible on Moes but like one way that I've been thinking about agents and how and these platforms and how it's all going to like interact is you know some people are starting to compare um or comp these these platforms even some of these agents towards like l1's last cycle and it's like is this the new L1 trade where like you can hit a 10x in size or these things going to run to like tens of billions of dollars and I think it's I think it's definitely possible one one kind of difference that I see though between agents and platforms and then l1s is l1s seem like they develop like a

stronger Moe o over time because like just at a very basic level if I have my money on salana to move it to another L1 I have to like Press buttons and like manually move my funds which is hard it's scary it's a little sticky people don't like doing that these agent platforms seem like they have less of like a lock in than an L1 does they don't necessarily like ACR liquidity over time which like helps them build up a moat um and I also think we're like so early into this trend that to your point sat it's like the market seems to be kind of telling us that like much more is coming um we haven't we barely even cracked the surface of like where this

is going so I think people are very hesitant to like take a big swing on the app layer when you can kind of bet on the uh infrastructure layer and maybe get exposure to like everything at the app layer yeah that's that's one way I've been thinking about it yeah I mean I think kind of on the mod side of things couple thoughts there so one I think it depends on if you're talking about agents or you're talking about the platforms right I think if you're talking strictly about agents I think kind of the Delta between having a moat and just attracting attention is effectively zero um and I think it it really is just an attention game and and kind of that is your vote at the end of the day right now with the platforms I think it's a little bit different I

think frankly at this point no one's really thinking about Moes um and it it's kind of more similar to the agents in the sense that it is just an attention game and yeah no one's doing you know sort of this this uh competitive analysis about these platforms um and trying to figure out you know who who's long-term going to have the most sustainable kind of competitive Advantage here but I think over the long run um I think it will matter um and I think it's increasingly mattering already um and the framework I would kind of apply to this is the same framework I applied to to kind of defi protocols and you know any sort of a crypto app more broadly and that's asking kind of this question um if I had $50 million in subsidies right and I forked this protocol um am I able to

sort of Steal significant market share if the answer is yes then you know you don't have OTE if the answer is no I think you do have OTE I think it it kind of you know it's an over oversimplification to some degree but I think it's a decent framework for analyzing these things and I think for these platforms the way you know I think at this point the answer is um you know I think it's pretty easy to probably steal significant Market chair I think what's going to probably matter more and more is when we enter a paradigm where you have agents interacting with agents on these platforms which we're starting to kind of see a little bit I know PD kind of alluded to this in in in the uh the AI piece but um I think once you have that that's where Network effects

actually start to to um kind of take off and then I think it becomes a lot harder to to sort of replicate these different platforms yeah I think that's a good point right like I still think the the best risk reward is still the platform plays and you know we're still very much in the Carlota Perez like you know speculative Mania and so you know Moes aren't really coming into play as much as people just think about how large The Tam is and these you know couple of players seem like they're in the lead so they get capital and people hope that they can figure it out and carve out a mode over time to me the two modes one as Robbie mentioned is like interoperability within a framework um and so you know if one flame framework basically develops a power law that a

lot of people start using can those agents more easily interact and can they come up with solutions that are you know more interesting than than basically heterogeneous Frameworks I think there'll be a large mesh it it doesn't seem like they're it doesn't seem like a place where there would be a massive power law around one particular framework I do think it'll be a patchwork but that's one potential mode and then also just like developer mind share right so what you're seeing with AI 16z even if they don't have you know any monetization mechanism I do think the amount of excitement um in terms of developer activity and contributions on their GitHub just does seem like something that will eventually be valuable so I almost think about it as

like Facebook you know pre-mobile and a lot I mean this is a big a big stretch it's not Facebook yet it's like super early but you know in terms of eventual monetization if you have you know the eyeballs you know it it tends to work out if you have the developer activity it tends to kind of work out right and so maybe it won't maybe it's different this time but that's kind of the bet people are making this early on in what is hopefully a really really large Market um but I agree it'll be fiercely competitive and you know outside of Agents churning daily I think you're going to start you know a gazillion Frameworks launch as well so I think you take a portfolio approach make several bets in the leading platforms and and be ready to you know evolve those views

because I do think they'll change over the next six months I think um like AI 16z versus virtuals is interesting because they're both kind of these platform plays but they're also very different uh like virtuals is this place where you go and you launch agents and then you know you pair the new agent token with virtuals you have to pay some virtuals to launch it there's of that burn it gets the fee income all this stuff it's very much like this coordinated Launchpad for agents ai6 Z was it's it started as like this Dow right and it like this hedge fund essentially that you were just buying at a crazy premium to nav and it was like hey why are people paying like 100x the

premium to nav on this Dow right but now you see like the Eliza framework has been going after absolutely insane right and so now you can kind of like start to work in you know other things they don't have the monetization side figured out um but their their framework is obviously like the leading framework right now it's the one that's getting the most interest out there and so I don't know I think these are two just two like interesting platform plays which are also very different and kind of like the construction and the actual value cural to them um I think that's why virtuals probably trade I think I think you can argue that a6z Tech is better than virtuals I don't

know maybe that's a hot take or maybe they're around the same let's say they're around the same okay but virtual still has what 3x the market cap or 4X of ai6 Z because it's like it's more coordinated it's simpler it like everything kind of flows elegantly and it's also like the the everyone's looking for what's like the base token right and it's like it's it's not eth it's viral right and so that kind of goes into it too I think yeah I I agree it feels like virtuals is very much like constructed from the bottom up like very organized like a pretty pretty intricate design with like decent incentives baked in and then the other platforms like ai6 and Ziri bro like you know basically hit

a jackpot in terms of you know kind of meme coin and attention um and it wasn't really originally supposed to be you know monetized in this format right you started with an agent with Ziri bro and then you're basically like holy I have a $400 million valuation I need to justify this like what's the best way to do that like oh actually probably the long tail and launching my framework um you know is probably a best the best way to do value acrel and then AI 16z is kind of similar in that it was kind of an open source framework they were monetizing it via like meme coins and their own agents um initially which you know is kind of a weird way to monetize open source software

um and I think you know at some point they will hopefully make it a little bit more official and and bake in some sort of value acru but but maybe not right but yeah I do think it's it's kind of like the sporadic chaotic Bottoms Up Cambrian explosion versus like a more elegantly designed system from the beginning and and frankly we'll we'll see which one WIS yeah I also think um the chaotic thing like I put out this tweet this morning that'll piss off a bunch of people you know October 13th is kind of the day I consider this crypto AI agent narrative to begin that's that's a few days after goat launch but it's the day that everybody started talking about it right I don't even think you guys wrote about bit tenser in

this report uh if you did I don't really remember seeing it but it's down 35% since that day okay while you've seen Go I mean go came back I guess far coin is the terminal uh token now um but you've seen like virtuals go up a t T you've seen A6 zbro all this stuff right I think that whole chaotic thing is just you know you've also seen me like a lot of these meme coins have not done too well since then either you've had some of the big top dogs do well but a lot of the mid to smaller ones have totally sold off now right and so I think like I think there's going to be a lot of money made in this sector and a lot of money lost right which I don't think is a hot take it's like just what's going to

happen um because this is all moving so fast right like if you just if you just held your tile bag over the past two months you were early to crypto aai and you lost 35% instead of making like 10 20x right and if you're not like completely plugged in to this sector today I can say yeah virtuals is the easy set and forget buy but it's also over two bill now that's not like cheap and maybe something better comes along for this sector it's just so new it's like everyone just kind of thrown at the wall and I think the fact that far coin is now we 3x goat it's kind of flipping a lot of everyone was like oh

goat's the Bitcoin of this crypto ey narrative yeah it's just a meme coin but it's like burst it's set off this whole wave no way it's going to be like flipped or anything like this a lot of these preconceived notions you know I've kind of been thrown out the window so that's why I find the sector interesting though is because it's I know we have these internal chats and there's like some people that have just been w waiting for like their agent to uh do well and it just hasn't hasn't worked out for them unfortunately Rinko and I are in that bucket but but yeah cus if you'd uh hopefully our our listenership read dai3 and dai4 and uh you know ditched the towb for uh some more creative Endeavors so fingers crossed it

it continues yeah I think on the market share point like I see market share within this AI sector being extremely fluid um for the foreseeable future and like this is something that we see even within kind of web 2 uh macro AI you know I think from 23 to 24 um open AI lost just under 20% of market share mostly to anthropic so the space is incredibly fluid I think the pace of innovation um kind of makes this like much more of a Traders Market than kind of like a just a trending Market that you could buy and hold over long periods of time and you know one of the big questions that talk about in this piece is like what's going to be the next big

breakthrough and I think people are are kind of asking that on the macro AI side but also like um onchain crypto side you know right now the agents are cool but they're still kind of like toys you know what I mean like the most that these things can do is like order pizza for you you know that made headlines a week or so ago um they're kind of they've moved over to these these tea enclaves which I know C you guys probably talked about in the INF report so we're starting to get you know more capabilities and more like credible like autonomous agents um but we're we're still not quite at the point where like at least in my opinion these things are like legitimate like instantiations of

intelligence that are like taking their own actions in the real world um there's usually some sort of human somehow in the loop to me that just that just it seems like market share is going to G to you know continue to be super fluid until we get deeper into this trend yeah I I 100% agree um but kind of back to the the macro discussion like I I do think one of the reasons I'm bullish is even in the event that we don't see any more capabilities emerge which I think is highly unlikely or even if they just start to slow down I just think there's so much white space for how to use these tools that haven't even begun to be considered right so it's like literally top of the first inning in just figuring out how to productize this new um you know generation of software right and so

and I've been encouraged by what we've seen in web 3 and the amount of creativity because you basically just take the Baseline intelligence and the tools and you get it in the hands of a bunch of weirdos and a bunch of really smart people that might not work at you know uh Mega corpse and they do cool and interesting things and they do a lot of experimenting um and I think you know even in the event that capabilities don't have a step function increase which I think they will I think there's a ton of room to run with basically just what we have already okay maybe we can move on to something Rinko brings up all the time there's very strong opinions on each side which is worldcoin uh and like World chain I know we wrote about this a bit in the report you know this

valuation has always been super insane to me uh every time Rinko brings up maybe longing it I look at the FTV and the unlocks and I'm like what like what are you talking about I need that I need that young young 20s energy back but uh you know you guys actually did make a decent case for it you know the potential for like worldcoin maybe to flip like open AI at some point you know I'm I'm still not planning on getting my eyeball scan but I know some of you guys here have done it it only Burns for a couple days yeah but you you close your eyes and you see Sam 's face yeah but I know I mean you guys so maybe we can talk into jump into this world coin quickly and like

also there's like a stable coin angle you kind of put here after too right yeah so I guess in this section as a whole like and and we can jump into worldcoin but basically just wanted to hit like four SE uh segments of Dai that I thought were particularly crucial to the outcome and the relative market share of web 3 meets AI versus just whatever AI ends up doing on its own and kind of those four I guess pillars ended up primarily being identity payments um decentralized training and data right um and so I think just with the capabilities that AI is continually generating um you're just going to see you know a ton of synthetic intelligences on the internet and we're

already seeing it like you know you look at our reply tweets for the deli year ahe heads and it's like all Bots um and so I I just think like as they get better and better there'll just be more demand for proof of humanness and it just seems like there's three broad approaches right so there's like the state-backed Biometrics almost like you know Adar in India um and then there's kind of like the specialized uh cryptographic Biometrics which is the world coin and World Playbook and then there's kind of the patchwork you know government ID plus ZK TLS plus social consensus um and it just seems to me like in the event that AI gets really good which seems likely um just from everything that I'm seeing from a trend

perspective then you would need really good Tech to prove that someone's a human um and the only person that's investing heavily in doing that right now through like you know boots on the ground scaled Hardware is World coin and so it's like very expensive very painful but in the event that that is the right Playbook then that's a very real mode for the next generation of the internet when a lot of other stuff is breaking and I think identity and payments are a very good Bas layer platform from which to launch you know a lot of other applications you know in the event that that World plays out even if worldcoin is 27 billion right now that Tam is a lot bigger in my mind but there's a lot lot of risk to that execution road map I do think when you talk about like Moes

this is like a real one that if they get it is basically I'm not going to nothing's impossible but basically impossible for like someone else to take them over then because of all the hard work because of all the deployments the fdv itself almost becomes a moat because it allows them to pay you know these tokens are unlocking and they're paying for real world operators you know in Africa that are scanning people's eyeballs and if you don't have that same valuation you literally can't afford to pay those same people to go out and scan yeah I actually think that's a great take um and when I was writing kind of the that Moes piece that's something that kind of came up is like number go up and being able to craft sort of a strong narrative that subsequently allows kind of number to go up um is

kind of a remote because that kind of kicks off the flywheel where you have more incentives to kind of give out and it's easier to boot trap in network um but yeah I I 100% agree I think you know what worldcoin is trying to do at scale is I think the ultimate mode um it could actually be a very strong wedge to kind of building out sort of this ecosystem around sort of that identity layer that PD alluded to um I think that isn't obvious today but I think kind of as we head towards this sort of agentic economy right I think it will become very obvious that that there's a need for that I also think one of the things that kind of went under the radar is sort of their world app um kind of ecosystem they're kind of building this app store kind of on top of world Point

um no one really talked about it but I do think that could be pretty consequential um and important and I think you know in crypto at least we kind of lack this app store um and this is something I kind of wrote about in in the defi year head report but we kind of don't have this this canonical like App Store where you access um sort of crypto apps and I do think whoever can kind of win over that market whether it is you know Apple it easily could be you know I probably don't see them going that route and I think you know Apple and Google kind of have been blocking a lot of crypto apps from getting on their app store right I think the wall layer is probably the best Contender I think at this point so metamask is kind of doing

this with metamask swaps um you're seeing backpack is now integrated blanks um so they're kind of creating like a monolithic user experience akin to sort of Tik Tok or you know even like X where you're accessing not apps but you're accessing blinks so and then indirectly interfacing with the apps um and then worldcoin you know I think probably has at least at this point um or at least at scale I guess um would have the best kind of wedge to building this out where they can prove that these are humans and these aren't agents right um so when you're accessing their apps on their app store you know you're interfacing directly with humans so yeah I think I think it's kind of a call option on a very asymmetric out-of- the money call option on sort of you know these agents becoming very mainstream heading towards

the agent economy and therefore for more demand for proof of personhood yeah I I think on the world coin uh Point like my my kind of one liner on like why I find it interesting why why I think it's um kind of worth following is like if you think people will find Value in distinguishing between a human and AI online then we need some sort of way to separate those those two buckets and and right now it's like relatively obvious right like you guys were just talking about like under our Deli tweet there were all these Bots like you could still kind of tell the difference between a bot and a human but if you play this trend out a little bit longer it's going to become increasingly harder and harder to to tell the difference online um so

then it just comes down to like is there value in telling the difference between the two I think there is people didn't really understand that right there wasn't like an obvious example of like oh that we we need to distinguish the difference that is until like goat actually happened I think goat was like the first big example and it was kind of flipped actually goat was the first time people wanted um to prove that something was an AI not a human right previously we wanted is this guy a human not a bot yeah remember the the spelling mistake of course it it dumped goat's market cap so like goat um truth terminal in case you're not at to speed is like this agent made by Andy and uh whatever he endorsed this goat mcoin and everything

and then one day truth terminal tweeted something it had a spelling mistake in it and people aren't used to seeing spelling mistakes and like chat gbt and stuff right um and so everyone thought that it was just Andy writing these tweets at this point right and so you kind of had this real life turning test maybe on like a lower level of it but everyone was trying to figure out like oh wait is Andy just tweeting these things out or is it actually the agent doing this and then it kind of LED people to understand oh wait so he's using this like jailbreak model okay so spelling mistakes are actually possible here because of the data the data that's being trained on and so you kind of saw

everyone had to figure this out in real time CED was very nervous net worth went down about 50% no I went to bed I I slept through the whole thing I woke up and it was over and uh but I looked at the chart when I got up and I think it Wicked down to like 7 cents or something uh it was crazy lots of lots of panic on the timeline but no I mean yeah I on worldcoin yeah I do think that obviously they have a lot of regulatory issu in Europe it's really hard everyone seems to like want them gone uh but that's anything AI in Europe I think that's anything in general in Europe these days I don't know they definitely want to to to keep following yeah and I I think it's like a it's a call option I

don't think it's a cheap call option you know 20 27 billion for this road map is pretty aggressive but it is one of the few like potentially transformative projects in crypto at the moment based on the world that I forecast you know five to seven years out um and so that alone just makes it worth keeping an eye on because that's you know a multi hundred trillion dollar I mean 100 billion doll opportunity at the very least yeah I do think that the it's very rare that an asset in the space goes over 100 bill there's not a lot and it's definitely one of those that you could see it like if you told me in 5 years from now worldcoin was half a Trill I would say like maybe yeah I guess I can see it I mean it it has the most

plausible chance of like solving um the Cy problem which I see as like one of the biggest problems in all of Technology right now like crypto struggles you know aird drops are kind of screwed because of the C problem social media is like a bot Fest because of the cwall problem like this is a problem that like affects almost every industry you know if a crypto project if anyone can like solve that kind of Bas layer problem I I I think that's going to be like enormously valuable um and there's all sorts of approaches which PD wrote about but in my opinion I I think worldcoin has um kind of the best odds of solving that in this like super

intelligent AI feature yeah maybe we can move on to decentralized training I know that this is this is a topic that I think nobody expected decentralized training to be able to compete like a few months ago I think PD you were also very much in this Camp too um my My Views have evolved you know strong opinions loosely held you gotta got to gather new data yeah no you got to be able to this is the thing we had Deli we're not Maxis of anything you know we don't dig our heels into the ground on opinions um you got to you got to change when the facts come out in in a space like this so maybe let let's jump into this you wrote some sections on here un like kind of what's going on in decentralized training yes so I think it's just important to set the record

straight um in my in my first report you know came out fairly skeptical just given the bandwidth needs of you know a lot of these large training runs like effectively what you do is you build a superc computer of you know massive clusters of tens of thousands of gpus and you wire them together you know as tightly as possible with you know expensive interconnect um so they can talk to each other um and you know it seemed very unlikely to me based on my understanding of how these large models were being constructed that a decentralized approach would be anywhere near as capable or performant as what the large Tech platforms have been doing right um and and so you know I was quite surprised when uh new research came out

with their results on you know a decentralized training run that just used orders of magnitude less bandwidth right um and I think it was like 800 times less the bandwidth than a traditional approach uh to to basically large model training um and it seems like from their research you know there's no reason why this shouldn't be scaled up and so to date you know I I I want to be clear like the largest model that's been trained by new was only I believe 15 billion parameters which is you know still very very far away from The Cutting Edge in terms of the most performant models but you know they basically did used a decentralized approach to mock the performance of soda level models at you know that tier and

that parameter count and So based on what they wrote this year and you know what prime intellect did you know two weeks later and then what nose followed up with another month later it does seem like there's a realistic path towards low band with you know training of large models which I think is a very big deal and it potentially you know transforms the Paradigm from hey let's just build massive superclusters you know in one particular area that's like hooked up to a nuclear plant or you know uh has tons of energy that just gets shoveled into it towards like hey you know is there a way to string together multiple data centers um to do you know much larger training runs across excess capacity um

and so yeah I think it's very much worth paying attention to and crypto does have a role to potentially try and gather a lot of that compute and aggregate it but it was one of the biggest bottlenecks in crypto AI like being a real thing in my opinion and not just relying on Zuckerberg and what he's doing at uh Facebook and open sourcing his models um so it does seem like you know based on some of the early results that there is a credible path towards uh decentralized approaches to training you know Cutting Edge level models um and I think that's really exciting so I it won't just be you know at uh Microsoft supercluster level tomorrow and these super clusters aren't just like staying still they're also like sprinting and wiring together

ever larger clusters but I do think it's a very real uh direction that research will head in and you'll you'll explore both simultaneously right we will you know head towards hundreds of thousands of h100s and B2 super clusters but you'll also start to see people you know working on uh heterogeneous um decentralized methods which I think would be a big deal yeah and I think this kind of leads us into dpin a lot of the dpin stuff has like an AI angle to it especially when you talk about like decentralized training my perspective on dpin I definitely think this is a sector that's going to continue to grow the incentives definitely make sense as someone who is deploying Hardware it's just like really hard to come up

with a good long thesis for a lot of them cuz they're just so early in the stage right I think like helium h&t is the only token that is relatively close to fully distributed and so it's also the biggest market cap but you know helium I'm looking at it right now it's like 6 660 and it's kind of been there for the past half a year it had like a nice rally in the beginning of the year and then it's kind of died off but yeah there's definitely a lot of like very interesting things going on in deepend so maybe we can uh we can get into that I know we focused on wireless and energy in this uh in this report yeah happy to dive in here yeah so I think PD said it best um the way I think the best way to

view a lot of these deep ends is they are very sort of deep out of the money call options I would say some are you know deeper out of the money than than others I think Helen's probably um I would say like the best valued uh call option but but at the end of the day you know they're you know most of them on paper at least if you're kind of doing some financial analysis and and kind of modeling these things out they're baking in some very ambitious assumptions that said I think some of these projects could potentially make sure on a lot of these assumptions and and I think the big thing is you know these are they're addressing massive massive markets um so I think that's kind of how you have to see them is you know when when the upside is so asymmetric um then I think

they kind of get priced accordingly so I guess we could kind of maybe start with DUI and then you know if we have time we could also do decentralized energy so I think you know kind of broadly you can split the you know sort of Telecom Wireless industry into kind of three subcategories so the most common one I guess is Wi-Fi um this is one we actually didn't discuss too much um and the reason is kind of simple Wi-Fi is like already very ubiquitous especially in um you know more developed areas it's more of an issue in in you know more developing kind of countries um and there's a handful of projects that are kind of addressing that that need but but just generally you know the Wi-Fi markets are relatively you know ubiquitous and efficient and accessible so I would say the bigger opportunities

in two other sub sort of uh verticals one is mobile um so this is kind of cellular data so you know when you're you know if you look at your phone in the top left corner you know you kind of Lte that's kind of what mobile wireless is and um that's kind of done by having all this infrastructure kind of the these towers and um you know and it's it's all kind of distributed and it's handled typically I think it's Verizon Comcast and T-Mobile are kind of three you know main carriers and they kind of have this Collective sort of oligopoly on um sort of the Telecom industry and the reason is it's a very intuitively it's a very Capital intensive industry um and if you look on paper like these look like incredible businesses and I think this is kind of the best lens

through which you can understand like why is De why does deepin have a role here and um what are kind of the structural advantages of a deepin alternative so if you look at like Verizon for example um you know on paper they have 59% um gross margins you know generating I think it's like over 130 billion in in Revenue um and that's all on you know about 100 billion of productive assets um and these These are the assets distributed you you know kind of all over the world so on paper these look like really really good businesses right um but I think and the EV3 guys wrote a report um in 2022 they kind of highlighted why this isn't actually the case and why these aren't very good businesses so when you kind of get into the nuances um the first big cost is Spectrum um so for those not familiar

spectrum is kind of the different range of of kind of waves that these these networks kind of operate on different frequencies and the government kind of has this Monopoly on spectrum and they get to kind of determine how this spectrum is allocated so the way this works is um there Spectrum auctions and basically the EV3 guys again put this better than I could of but they said kind of through these auction mechanisms um where you have carriers kind of bidding for spectrum um the government effectively grants themselves sort of a preferred Equity claim on Telecom cash flow so this is actually like a lot of money so Verizon has I think one it's like a 155 billion in Spectrum licenses on their balance sheet in other words they've spent about you know over 100 billion just on Spectrum so that's kind

of like one of the biggest costs um that kind of erodes uh the margins of these these telecom companies and then you have like you know operational costs obviously um you know there's very high capex for for a lot of these networks and then there's like a lot of kind of it requires a lot of Leverage too um so this is all financed through Deb markets traditionally that's finan at like 3.7% something like that you know with the rates being much it's about you know 5% so long story short like after accounting for all of this um you go from generating about like 50 cents on the Dollar on for every productive asset and then after these expenses it's about only 4 cents left over for shareholders um and these are like typically you know dividend paying companies so what actually happens is these telecom

companies actually have to borrow money to actually pay dividend so it's like not very sustainable and these are just fundamentally like very flawed businesses so I think so what is does this have to do with deep end how can kind of deepend solve this right so deep end has like inherently the structural advantages over this model given its kind of distributed um nature right so the first is kind of capital efficiencies kind of how you could describe this so that's that's where these deep in networks instead of you know having to put up all this money up front and borrowing all this money uh to distribute this INF infrastructure right they're kind of tapping into the collective you know economics of everyone kind of distrib out the world and you know obviously doing this through token incentives initially with the hope that you know the network kind

of sustains and then can kind of pay for itself right so that's kind of the first thing the second one which I would argue is probably the most important um and we'll kind of see how this ends up playing out but it's sort of like an arbitra like a spectrum Arbitrage where unlike you know the big telecom companies that have to pay these like very high Spectrum costs to the government um the deep in networks to date like helium have not paid for the Spectrum and that's because they're actually leveraging um what are called unlicensed Spectrum band so the the government the FCC released um I forget the exact range but a certain range of spectrum that kind of anyone can use um and you don't actually have to pay for and that's kind of what helium's operating within is is that certain Spectrum um you know bands so that kind

of gives them this very very strong structural Advantage where they don't have to where they can kind of circumvent the Spectrum auctions um and save a lot of money and then there's other oper like the operational costs are also outsourced to to people operating these networks and then again they're they're not tapping into debt markets because instead you know people are actually paying for this so needless to say this gives you you know very strong struct structural advantage over traditional telecoms so helium's actually competing and this is unique to to most Deens they're competing on price and I think this is a really important Nuance where it's like I think it's like $60 to $90 a month for for you know like someone like Verizon you know to to get access to to mobile you know bandwidth it's actually only $20 a month for helium so anyone can get an unlimited

plan right now for just $20 um which is pretty incredible and then last thing I would add kind of on healing more specifically um is this is this is a huge structural Advantage they can actually sort of steer the incentives where they kind of need it to be to more efficiently match supply and demand so I'll give you an example so I think it's Miami that has like a ton of you know activity and what they've done is they've incentivized a lot of hotpots specifically in Miami where there's a ton of concentrated um demand for um you know mobile data right and as a result they've been able to generate like a lot of Revenue specifically within that area right and that's unique to them because they're very granular um sort of coverage um and they kind of steer these incentives accordingly right so I think

that's another structural advantage that they have that allows them to compete on cost so that's kind of yeah the mobile sector um I'll kind of pause there in case anyone wants to hop in you know with like questions or comments and then we could maybe briefly touch on um fixed bandwidth which I think is probably a more interesting and um possibly even bigger opportunity yeah I mean I I I guess I I just appreciate their go to market I think is kind of clever in the fact that you know you don't need to boot bootstrap the entire network but they're kind of providing a little bit of an Arbitrage opportunity and partnering with you know the number three Kelco um who's trying to gain share but they're also doing that while they simultaneously bootstrap their own network um so it's not like Hey we're g

to uh basically um compete on performance out of the gate because that's like super expensive and you know really hard to build up Network effects and cover the you know entire us um and so I do think it's quite smart that they've basically gone this partnership route while they simultaneously fill out their gaps and then I think they can command more and more economics over time uh as they build out their own network coverage yeah for sure yeah and so for context what PD is kind of alluding to their is um so they've partnered with T-Mobile um to kind of provide coverage in areas that maybe helium doesn't um have coverage but then the other side of that which is really interesting is there's also like other carriers are now all floating data to

helium specifically in areas where where there's very high usage so this is another Revenue stream for them in addition to the $20 a month that they're that they're getting paid and some people would argue that this will actually end up being kind of the principal Revenue stream for for helium kind of going forward so um just to kind of wrap things up with helium you know in terms of kind of the token and and what's being priced in um I think it was trading you know at the time we wrote this it was at I think a $ 1.9 billion fdv if you kind of model that out what they're effectively pricing in is by 2030 their pricing in that helium will grow 21x um which sounds again that sounds very ambitious any Project R 21x is that's pretty incredible that said that would only still be like

0.005% of the entire Telecom market and if kind of network effects hit hit escape velocity at some point I don't think that's unreasonable at all um and I would say at a minimum that's probably um out of all the deep in projects that that you could sort of place a beted on I would say that's probably pricing in at least the most realistic expectations Al be you know a lot of people would argue not very realistic yeah yeah I do think one thing on helium I do think it yeah the the I do think the valuation is not it's like not insane right and an industry with a lot of insane valuations for a lot of like dumb I've been looking at Helium for like a year um I'm probably going to NFA I'm probably going to buy to some point I definitely got more bullish on them there's a period of

time where it seemed like they're kind of dead uh but they really came back with this mobile thing and have executed very well on this yeah let's go into fixed bandwidth and then also I want to touch on dble zero yeah so fixed bandwidth it's kind of like so this is like the actual sort of back hul that kind of connects the rest of the the internet so it's kind of broken into to three tiers so tier one is like this is like the backbone infrastructure so think like sort of Submarine cables um these are typically like Fiber Optic Cables kind of running um under the sea and then you have sort of regional fiber networks that connect to the actual data centers and then the last is sort of tier three which could be described as kind of the The Last Mile um and this

Bridges the G from the actual to the distribution points um and kind of to the individual buildings so like historically this kind of to topology has resulted in like very sort of monopolistic a very monopolistic Market structure um because again it's it's and this is kind of the common denominator you'll see them on like all deep in networks is they're kind of these inherent monopolies because they're very Capital intensive um and it's kind of resulted in these winter take most at least markets um and then if we can kind of distribute this from like a h and spoke to more distributed model um through token incentives and kind of tapping into you know people then can kind of disrupt this these these uh monopolies in theory right so this is no exception to the rule so I think the important Nuance here is it's the last

mile that costs the most um that's the most Capital intensive um and as a result I think 51% of Americans um only have access to one provider so obviously this gives them a lot of you know pricing power um and it's it's very expensive as a result so andrina is kind of the the I guess Pioneer in this category in the fixed bandwidth category and this is a project I'm extremely extremely bullish on I would say probably even more bullish on um than helium um so what they're doing is there's kind of a few key trends that have kind of allowed them to build out what they're building so first as I alluded to earlier FCC released the certain range of spectrum that they can kind of operate on and the second is like what's what's unique about Andreina is there's actually like technological Tailwinds as well so they're using

what's called beam forming and point to multi-point technology so fixed bandwidth intuitively it's it's fixed and it's always like these physical networks what they're doing is they're building out almost like these satellite networks that can kind of communicate so this allows you to deliver that last mile without with while circumventing sort of having to build out these fiber optic cables that are really expensive right so today they already have 10,000 subscribers across 10 different states um and they're already basically offering uh gigabit service at 30 to 50% cheaper than incumbent so they already have this structural Advantage which is uh really really unique and this is before even integrating a lot of the the the deepin elements um so to date all they've been doing is really just partnering with like different like real estate companies you know buildings that

have like where they can kind of place these their hotpots and they haven't even tapped into kind of token incentives right so they're now building out Dawn which is basically their dpin kind of protocol which is an extension to kind of andrina as a whole and kind of what Dawn's building out is they're using token incentives to one basically provide they call proof of back hole so you can actually download I have it downloaded right now actually but the the dawn extension um and you you can kind of start farming uh Dawn I don't think the incentives are going to be great for the extension but it basically just shows kind of this proof of of of kind of coverage and and back all um so they can kind of have this network and kind of map out you know kind of the different internet connections um which

kind of helps augment the the entire network as a whole and the idea is like at scale basically you could provide you know fixed bandwidth but just way cheaper basically using one this technology and then two harnessing kind of token incentives to eventually get people to in their at their house for example put up their own um kind of hot spot and yeah they get scale this could be a very effective model and I think it's interesting because it's at the intersection of um you know one de using deepend incentives but most importantly there's actually like a technological advancement here um and the founder Neil he's a very very sharp guy he he like comes from a background where he was like building these systems out I think he was a pred Dropout so he's also a really really sharp guy and then I guess

one thing I would just add here is the biggest threat to this model is probably nural link but I would say the structural advantage of something like Nur link or starlink or sorry excuse me uh starlink appreciate that PD but I would say like the the advantage that that uh Andrea would have is is kind of physics favors proximity so starlink it's it's probably going to be tough to compete just given that these satellites are are very you know far from where they're ultimately kind of connecting to and then also starlinks like inherently do more expensive at this point relative to to urina so we'll kind of see how that plays out but I would say that's probably the biggest threat to this I thought Robbie was like living 10 years in the future but um I I I do I do want to I do want to hear about dark fiber I

know Lov dark fiber yeah this is the uh this goes in line with the high throughput blockchain thesis that I kind of wrote about too right yeah I would be curious what you guys think of this as well um and I apologize if I'm uh rambling here but we'll try to get through this one quickly so double zeros I alluded to this earlier but you have these like Fiber Optic Cables these undersea cables right traditionally like the way the internet works is you're kind of routing these different um you know data packets sort of across across the globe right but the way it's done right now is like super inefficient um and I think most importantly it's optimized for for cost and not performance right so the way the route that you're packet ultimately takes um is the cheapest route not necessarily the most performant and the

the fastest route so why does this matter well I guess particular for like distributed systems like blockchains this is actually like really consequential and really important like you want again for more performing chains um you want it to be optimized for Speed not necessarily cost um and there's other like demands from you know um you know AI training or content delivery networks and other kind of bandwidth intensive applications that would also benefit from from speed over cost but I think the applications most interesting kind of in the in the blockchain context so this model would basically allow and and I'll kind of first explain maybe you know how what's the deep inside of this and and how does it work so there's a lot of like hft firms for example who have these private uh fiber optic networks that they basically built themselves um kind of

under the sea and again these are obviously optimized for latency these are hft firms but a lot of these fiber optic networks don't actually get fully used um it's very rare they're even operating at like you know 60 70% capacity so you have a lot of idle capacity that idle bandwidth that they're just kind of not using so what double zero is basically trying to do is use cryptoeconomic incentives um to incentivize um these people to basically add like contribute their excess bandwidth um to to the double zero Network and the idea is at scale you build out this massive Network um where people can kind of pay for this this premium bandwidth that's that's a lot quicker right um so you kind of have like you know the normal routing system and then you kind of like this Express highway system um

that does things just a lot faster and a little more expensive and the applications are specific I guess in the blockchain context um so this could be like faster transaction processing um this could be you know more consistent validator performance um so things like this but I'm curious that I don't know if you guys wrote about this in in infro or not but yeah curious if anyone else has kind of a take here I just mentioned it because they was announced I think the day before in for went out but yeah this is something that people have been hinting at too since breakpoint especially mert uh they're basically building a new internet it's just like a very elegant design I think this whole like you know uh increas band with reduced latency thing there are just physical limits to you know how short you can get Block times just because of

like sending a packet from you know New York to Tokyo takes a certain amount of time and I think the way it works today is it's just like super random right so if you're sending a packet from New York to Tokyo it might actually go to like Chicago first and then go to I don't know London and then it goes to Tokyo right and so that again just adds more latency and the whole point here is if you're willing to pay for that direct line to Tokyo uh you can get it right and it's just just like there's like uh light fiber and dark fiber right and so light fiber is the stuff that's being used all the time and then dark fiber is essentially the fiber that's just sitting there and it's not being used and so they want to light it up that's

like the tldr let's light up the dark fiber right this is very you know this whole High throughput blockchain thesis the whole thing is like reducing latency you know I don't think I don't think l2s compete with these decentralized l1s like salana s uh like mon when it comes out hyper liquid if they're able to to do this because of like their single SE uh sequence or architecture right the Holy Grail is getting like multiple concurrent proposers on This Global distributed Network and if you are able to you know one of the ways to do this is yeah you need to protect the actual blockchain itself but you need to fix the whole networking like bandwidth layer too and so that's where Z comes in and it's just fixing the internet for this specific use case and that makes me

pretty bullish on actually getting a decentralized NASDAQ one day right it's like a very ambitious Target the the Salona manlets care so much about this that they're literally building a new you know this is from Austin from salana Foundation there's some fire dancer contributors some others I don't know it makes me really bold up on the long-term potential of blockchains being able to be actual like price Discovery for a lot of these assets because the way it works today is that price Discovery still predominantly happens in C on centralized Avenues right uh centralized venues and so actually being able to get Global price Discovery with multiple concurrent proposers where we have like leaders in like Tokyo New York London

all at the same time that's kind of the Holy Grail and that's also why I don't think people like to compare Salon against ethereum a lot they're two very different things like if the world Demands a global decentralized price Discovery Network that is not something ethereum's competing in right that is something that Salon is competing in something that maybe s is competing in again some of these newer ones coming out it's like a very specific thesis in like very specific product and this is kind of like the lever that I think can Propel the industry there so yeah I'm pretty bullish on the idea one one thing that I've kind of been thinking about with double zero you you guys both hit on it is um the the crypto internet is

is different than like the the normal web 2 whatever the traditional internet um so like when I'm loading up a YouTube video or a Netflix video Netflix or YouTube doesn't necessarily care if they serve me that video in 400 milliseconds or 500 milliseconds they just want to make sure that it's under a second and I don't have any like lag time right but there's no like cost to them or to me if it's 400 or 500 that's kind of how I think about the public internet whereas in crypto there there actually is a real cost if your if your transaction takes 400 versus 500 milliseconds and being able to to model out um whether it's going to take you know whether it's

going to take this or that becomes really really valuable and right now like over these kind of traditional public rails um you have it's really hard to figure out is it going to take 400 or 500 milliseconds and I think kind of the fundamental problem that double zero is trying to solve is give people um more certainty over the time it's going to take to get uh a piece of data from one part of the world to another um and that I I think the really cool thing is that actually wasn't a huge problem you know was a problem for like hft people um but it wasn't like that big of a problem before crypto came around and kind of like embedded um financialization into this like very

base L layer of the internet um so yeah I'm I'm super bullish on double zero as well yeah it's very well said and just to underscore that point because I do think it's really important the internet today is like the way data is routed these data packets is strictly optimize for cost um and what double zero is trying to do is basically you know bifurcate the system into things that are optimized for cost and things that are optimized for uh performance right um and I think at scale and this is kind of the the broader Vision I guess stepping back for a second of of all these deepend networks is it's kind of similar to people who are familiar with Enron this is kind of and Enron like ended obviously very very poorly but and that's why like you know you don't want to yeah exactly you don't really want to

conflate the two just narrative wise but but I do think Enron was directionally correct in their Vision maybe this contrarian take maybe it's not but um where effectively what what they were trying to do is create sort of markets like and this was before tokenization um but it would kind of be like securitizing um everything from bandwidth compute energy right um and what double zero is doing is basically you know you're tokenizing performant bandwidth right um so you're getting even more granular than you could in my opinion this is what crypto is good at right it's good at kind of two things one is incentives and two is creating more efficient markets right and I think this is if this is like the vision for for crypto I'd be very happy if we kind of get this to to scale and the EV3 guys have a very elegant way of putting this

is and I think this becomes also very true as we head increasingly towards sort of an agentic economy but um maybe the currency of today is sort of you know dollars the currencies of the future are kind of going to be you know compute bandwidth storage and energy um and deepin is basically giving you sort of the the substrate to basically build um markets um where you can tokenize these things and have efficient markets for these but yeah and maybe to put it differently like you know an AI agent doesn't care about dollars they don't dollars they don't demand dollars right um so what becomes sort of the most scarce and and demanded resources in this agentic economy it's it's those things that that I kind of listed so that is kind of what makes me also very bullish on deep end it's not just

basically building out these distributed networks um and disrupting these monopolies but it's also actually what what those networks allow is actually having markets built on top of these where we now have more efficient price Discovery um for all these things um and I think at the end of the day that actually benefits more than anyone it benefits you know us the kind of end consumers yeah I thought I thought that was well said and one thing we didn't cover that's covered in the report so you should go read it is basically how you know markets for data are impacting this equation if indeed data is the new bottleneck in AI so I think there's a some juicy segments in there around how data can be used in embodied intelligence and unlocking private data sets uh but in terms of unliking markets that seems to be you know a big theme

for 2025 is how do you unlock markets for data and it seems like crypto does have a role both across da and dpin um so I think that's also worth worth uh reading through in our report yeah definitely uh go check out this report it's like I don't know two-hour read but it's an easy read it's very engaging um you learn a ton I thought everybody here did a great job today on this pod I think this was probably our best discussion that we've had on the year heads so again appreciate all the guys coming out here taking the time you know different mix of time zones here so appreciate everyone making it work and uh yeah once again if you're a listener here go ahead

check it out on the portal you can get one1 trial for $199 that gives you all the reports we've made and like everything else on the portal and yeah thanks everyone have a nice holidays and we'll see you all in the new year [Music]