Why the AI Business Model Is Cracking and How Crypto Could Help Fix It

[SPEAKER_02] But, generally, I view it as Big Brother fear-mongering, and I think it's super annoying. The idea that we're letting some AI lab dictate what we're able to ask and why is a really slippery slope. Because you just don't want to be told what you're able to research or look at or ask about, right? Like, do you want to be in China and you're not allowed to ask about Tiananmen Square? Like, no. Like, people don't want that. [SPEAKER_01] Hi, everyone. Welcome to Unchained, your no-hype resource for all things crypto. I'm your host, Laura Shin. Thanks for joining this live stream. Before we dive into today's show, let's hear a word from the sponsors that make this show possible. This episode is brought to you by CAPE, America's privacy-first mobile carrier. Same premium service you'd expect from any other carrier, but designed so your number, your location, and your data actually stay yours. [SPEAKER_01] Get 33% off six months at cape.co. Fidelity has been investing in blockchain since 2014. They're not wondering if digital assets will shape the future. They're hiring the talent to help ensure they do. Explore opportunities today at crypto.fidelitycareers.com. Fidelity is an equal opportunity employer. Quick plug before we keep going. If you want crypto news without the hype, subscribe to The Unchained Daily. [SPEAKER_01] It's our free morning newsletter. The day's most important stories explained clearly in a few minutes. Sign up at unchainedcrypto.com slash newsletters. Today's guest is Tommy Shaughnessy, founding partner at Delphi Ventures and co-founder of Delphi Digital. Welcome, Tommy. [SPEAKER_02] Laura, thanks so much for having me on. I've been listening to your show forever, and it's great to be on. [SPEAKER_01] Thanks. Yeah, excited to have you. You've had some really eye-opening tweets recently about the AI business model, and those spark some obvious questions ahead of these mega IPOs that we're about to see, including SpaceX tomorrow. But before we talk about all that, I just want to start with a question for the crypto markets. We started the year at about a $3 trillion global crypto market cap. It's now at about $2.2 trillion. And over $600 billion has left over the last 30 days. [SPEAKER_01] And there's so many theories out there. But I was curious why you thought that was, and also if you felt like some of it maybe has to do with these, well, first of all, just AI generally, but also these blockbuster IPOs that are coming up. [SPEAKER_02] Yeah, when you say the numbers like that, now I know why I'm crying on the shower so often. It makes a lot of sense. Yeah, I mean, crypto markets have been extremely brutal, right? I think it's a mix of things. I think headline for the overall trend has just been outflows to AI deals. Like people have been killing themselves to get into second, third, fourth layer SBBs for OpenAI, Anthropic, and a whole host of names below them. [SPEAKER_02] And then on the not-pull side, you just have people pushed away from crypto because of, you know, fears about quantum, fears about strategy, and things like that. So it's been rough sort of on both sides. [SPEAKER_01] And what do you think will happen in the crypto markets these next few months with these big IPOs starting with SpaceX tomorrow? [SPEAKER_02] It is interesting. I think SpaceX does well out of the gate. It's, I mean, it's, my partner Pierce published an incredible report on SpaceX, and I read through it, and it's just such a fascinating story. I mean, he built his thesis around the cost per kilogram to ship something into space, and what you get as that decreases, like all the new technologies and capabilities in space. So I don't know. I think it does really well. [SPEAKER_02] I've been watching Polymarket, and it looks like consensus is above a $2 trillion value, 2 to 2.5 closing market cap, and it's going on at 1.75. So I think it does good for the first couple of days. I'd be a little bit more wary as time goes on, though. [SPEAKER_01] Okay, okay. But then do you expect the crypto markets to be choppy these next few months with OpenAI and Anthropic coming? [SPEAKER_02] Yeah, it's such a hard thing to reason about, right? Because on one hand, you have people who were never able to access these IPOs, right? Index funds, you know, maybe me and you that couldn't jump in, or just mass retail. And they want to buy AGI stocks, right? They want to tell their friends they're in it. They want to be part of the story. And they use these products all day, and they're exciting, right? And they're more exciting than owning Bitcoin right now. It's just clear. It's hard to argue that. [SPEAKER_02] The problem, though, I think, is not on the short term, right? Like, the floats are small. I think the retail interest is there. And I think the index interest, the pass the funds, that's enormous amounts of capital that will be either forced to flow in or want to flow into these names. I think the problem could be medium term, a little bit after the IPO window, where stuff starts to get a little tricky. [SPEAKER_01] Okay. So, yeah. Let's now talk about the AI IPOs and just generally that whole business model. We're going to circle back to crypto at the end. But I want to talk about your post that went viral recently. And it's been vindicated by other analyses that have come out from the likes of Citadel Securities. And your post was titled, The Most Basic Way AI Could Blow Up. Explain your thesis. [SPEAKER_02] Yeah, it is wild getting 2 million views off a tweet. I just can't understand how many people that is when you think about it. I mean, my thesis is kind of straightforward, right? I mean, there's an incredible associate within Delphi Ventures. His name is Rawson. He's exceptional on the AI side. And we have an API spend for Anthropic at our company, right? And he kept hitting this wall on spend, which I'm really happy about, right? It means he's killing it with analysis. [SPEAKER_02] But it was weird to me because when you get a ChatGPT or Anthropik subscription, you get a ton of usage. You get, you know, it's hard to hit a cap on that usage, right? So, once you shift from a subscription to an API model, you hit a wall really quick on spend. The subscriptions are subsidized by about 40x versus what you get at the API. So, $200 a month gets you about 8 grand worth of API spend. [SPEAKER_02] Now, what I think is happening is businesses are transitioning to these API spends and they're hitting a wall. They just don't have enough money to spend. Uber cut their spending. Microsoft cut their spending. A lot of companies are realizing that, hey, this is just too expensive. And we're used to what we got on the Chachi GPT or OpenAI or Anthropik subscription, but it's just too expensive for the API side. [SPEAKER_02] So, I think what's happening is these companies are starting to question, where can I get my AI usage without paying a zillion dollars? And open source inference providers provide this solution. You can go to Base 10, OpenRouter, Venice, NUS API portal and get your same intelligence for one to one 100th the cost. You're not getting frontier intelligence. [SPEAKER_02] You're getting something 80 to 90% is good, but 1% the cost, if not more. And I think that that has the potential to negatively impact this whole flywheel, right? Long term, I think we get AGI and very bullish. But short term, what happens is if people churn off, revenue slows. If revenue slows, these companies' metrics and ability to raise capital slows because they got to go to the debt market and say, hey, we might not get this revenue back. [SPEAKER_02] We may or may not. It's already pretty exhausted at this point. And that could negatively impact the flywheel. So, I don't know. Everything's been priced to perfection recently and it's been up and to the right. But I think these enterprises are starting to tell their employees you can't token max anymore. And I think that that could really negatively impact the current flywheel. [SPEAKER_01] So then what do you think that means for the upcoming IPOs? You know, they're sort of like a low float, high FTV sort of structure to start. And I guess earlier in the episode, you did say you feel like they will do well out of the gate. But do you feel like they're going to crash at a certain point due to the competition from the open source models? [SPEAKER_01] And like generally, what do you think, you know, should happen or could happen in terms of their revenue? [SPEAKER_02] Yeah. I mean, it's I think they do well out of the gate for a couple of reasons. The first one is the floats are pretty low. You know, for for some of these, it's, you know, call it sub 10 percent when historical IPOs have been 10 to 20, something like that. And there is a lot of passive capital index funds, folks that want access. I think they do well out of the gate. [SPEAKER_02] The part where it gets tricky is historically people have aped in these companies because the stories and the incredible technology with limited to no financials. Right. When you're a public company, you now have to start disclosing everything. You know, what are your payback periods? What are the margins? What is the growth? What are subs like the whole nine yards are now for the public to to debate? [SPEAKER_02] And I think that people are going to see if my thesis on the enterprise is getting more cost conservative pan out that could negatively impact the flywheel here. It's it's to be seen. I mean, there are a lot of counters to my thesis, like businesses will pay up and they will go to the API and they will use it indefinitely. And there will be no cost concerns. [SPEAKER_02] We're not really seeing that from the citadels of the world who posted like a similar thesis and and others. But the metrics do point to that story. Right. Anthropics revenue historically has been like hyper growth. So for that to slow down will be tough. But I think that there will be some slowdown because I think the stuff is just really expensive. [SPEAKER_01] Yeah. Side note, the Citadel piece was called tokenomics. And before I read it, I assumed it was about crypto economics. And clearly it was. Yeah, it was just about AI tokens, people, not how we define tokens. But anyway, well, so, OK, they're they're trying to shift to this model where people are going to pay for or businesses really are going to pay for tokens. [SPEAKER_01] But there is, you know, going to be this leakage, as you called it, to the open source model. So is there going to be some kind of like striation in the market or, you know, like how how are you seeing? First of all, these basically frontier labs companies, like how are they going to fund themselves if so much, you know, activity is going to go to the open source models? [SPEAKER_02] Yeah. I mean, it's it's an it's an interesting world we live in, because. Because when you have a when you have something that has pretty close substitution for such a low cost, it kind of begs the question, like, is everything right? Right. Like, it's not like you can't access AI through only through the frontier labs. Like there is an avenue where you can access this intelligence for 90 to 99 percent cheaper. [SPEAKER_02] Like you can easily port over and it's very easy to port over nowadays. I'm a huge, you know, news investor and fan of Hermes agent. And that's the agent that I use my daily driver. And I hit one key on my keyboard and I can switch models in an instant. I can go from open AI to Anthropic to open router to Venice and then choose my model. It's very easy for most people to switch. And so the cost is ridiculously cheaper. [SPEAKER_02] We've already sort of established that. The second thing is the data privacy and data retention key points that push enterprises to these enterprise plans is no longer a monopoly offering. Open router allows private models. Venice has really strong privacy guarantees. They offer models on private data centers. They offer models with end to end encryption run in TEs. [SPEAKER_02] So the idea that you can't get privacy as an enterprise, too, is kind of slowly being shipped away as well. So, again, I really just think it's a timing mismatch in my opinion. I don't have access to their financials. I don't have that data to dig through. So I have to go through what I have. But what I'm seeing is just a really cheap substitution for really good quality on one hand. [SPEAKER_02] And on the other side, I'm seeing enterprises all over the place hit a wall on spend. And when you put those two together, it kind of makes sense that usage is shifting. So that's sort of where I think is happening short to midterm. Long term, I am 100% bullish. We got AGI robots on these mass data centers. [SPEAKER_01] Yeah, but let's focus on the business model question because, you know, what you said there makes so much sense. But it almost feels like a tragedy of the commons type of situation where everybody wants the models to get better. But that requires data, but they don't want to give their own data and they don't want to pay. [SPEAKER_01] Like there's just so many weird little things where the way the incentives are currently set up, people are it's trending in the direction where we won't actually get what we want. You know, whereas like there's it seems like there could be a way to, you know, fund this type of stuff that would allow everyone to get what they want. [SPEAKER_01] But but what so just with the way things are trending currently, you know, what do you feel like will be the way that these frontier labs companies try to get the revenue they need to fund their GPUs, to train their models, etc? [SPEAKER_02] Yeah, I mean, Laura, that's that's like a really good question, right? Like that's like the crux of this is like, how do we get that AGI frontier capability we all want? But we also need to make sure users can pay for it. It's accessible to everybody and we need to pay for it, but it's getting expensive. But we also want this end result. I think that's like the best question and honestly what everybody should be asking themselves to answer you. [SPEAKER_02] I think it is a bit of I think it's multifaceted. So on one hand, I think these businesses will open and anthropic will continue to make absurd amounts of money like through chat, through the subscription mall, through the API that will continue indefinitely. My my concern is the rate of that growth. And if people will shift off near term to alternatives, they're extremely cheap. And I do think that that is happening. [SPEAKER_02] Open router has shown massive usage, just raised a 50 million dollar round. Venice's usage that Eric Orhe shares is off the charts. Base 10, you know, pick your inference provider together. AI fireworks, you know, whatever. They're all showing mass usage. [SPEAKER_02] The other thing, though, I think is I really think that these frontier AI labs need to have direct ownership of the massive new creations they create to rationalize the spend. If it's a new medicine, they should own. I don't know what it is. Ten percent of it. I don't know if it's a new robot. [SPEAKER_02] If it's a new technology, if you have ridiculous frontier intelligence creating these things, OpenAI cannot be the beneficiary of a $200 a month subscription. Right. That just isn't fair to them. It doesn't make sense. [SPEAKER_01] OK, so in a moment, we're going to actually talk about one other factor, which I'm going to call the China factor. But first, we're going to take a quick word from the sponsors who make this show possible. If you hold crypto on your phone, your biggest vulnerability isn't your wallet. It's your carrier. AT&T, Verizon and T-Mobile have been breached again and again. And SIM swaps are still one of the easiest ways for attackers to drain accounts. That's where CAPE comes in. America's privacy first mobile carrier. [SPEAKER_01] Same premium service, but CAPE rotates the identifier on your SIM every 24 hours, deletes your call and text metadata after a day and protects against SIM swaps with a 24 word recovery phrase that only you control. You also get two middle to end encrypted secondary numbers for banking and signups. So you stop handing your real number to every app that asks. [SPEAKER_01] Go to cape.co slash unchained and use code unchained for 33% off your first six months. Fidelity has been researching and investing in blockchain since 2014, long before it was a headline. And they're hiring crypto and DeFi professionals to join their team and discover what's next in finance. Fidelity is looking for people with fresh perspectives from different backgrounds, whether it's tech, UX, or product design. [SPEAKER_01] Whether you're crypto savvy or crypto curious, as long as you have the passion to make a real impact at a company striving to make finance accessible to all. Explore crypto careers at Fidelity today and make the decision that could change your future for the better. Visit crypto.fidelitycareers.com to learn more. That's crypto.fidelitycareers.com. Fidelity is an equal opportunity employer. Before we dive back in, a quick word about something we make ourselves. [SPEAKER_01] If you like the way we cover crypto on this show, clear-eyed, no shilling, just what actually matters, you'll like the Unchained Daily. It's our free newsletter that lands in your inbox every weekday with the most important stories in crypto written by journalists, not marketers. No jargon, no hype, no number-go-up cheerleading. Just the news you need to stay ahead in the time it takes to drink your coffee. Subscribe for free at UnchainedCrypto.com slash newsletters. [SPEAKER_01] Back to my conversation with Tommy. In various tweets, you also talked about the open-source Chinese models and the potentiality that they'll end up going closed-sourced. So I was wondering how you thought the moves of the different Chinese models would affect the revenues at these U.S. companies, the frontier AI labs. [SPEAKER_02] Yeah, another great question. The Chinese models are all open-source, right? DeepSea, Kimi, Minimax, GLM, and they're incredible. I really like GLM 5.1. People like all these models for different reasons, and they fine-tune the hell out of them for all sorts of reasons. The thing that is interesting is eventually I think the Chinese models may go closed-source. [SPEAKER_02] I don't think that people are expecting that, but there were kind of some rumblings here, right? Like Alibaba released the model a couple of months ago, and they open-sourced it, and the next day the main dude was fired. GLM, I think, kept 5.1 or 4.7, one of those, private for a little while before fully open-sourcing it, right? So it feels like there is some rumblings there, but nothing has really changed. [SPEAKER_02] The reason I think it could go closed-source is because these companies are spending tons of money to train these models, right? And they're really not getting any value flow from it. They're not getting the data because the models are run on inference providers like Venice, Open Air, things like that. So they're not getting revenue from that. They're not getting the data flow either. [SPEAKER_02] When you use the model to train the model, like that's a huge training loop of information that they can use that they're not getting. So I think eventually if they continue spending obscene amounts of money, they might go closed-source. Their current strategy to open-source makes a lot of sense. If you don't have the frontier intelligence, you want to impact your competition by giving a close and free substitute, right? It impacts margins. It impacts their funding. That's what we discussed earlier. [SPEAKER_02] If China does go closed-source, though, the models will have to be really good. And they're not there yet. Because if everything's closed-source, you're going to pay for the best version and you're just going to use OpenAI or Anthropic. [SPEAKER_01] Oh, okay. So even if they go closed-source, you don't feel like it will affect the competition amongst the frontier labs here in the U.S.? [SPEAKER_02] I think if they go closed-source and they're not better than the frontier labs here, it helps OpenAI and Anthropic. Because you no longer have a substitute to use away from OpenAI and Anthropic. And two, OpenAI and Anthropic could charge whatever the hell they want because there's no alternative. And if everything's closed, you're just going to use the best. [SPEAKER_01] Okay. So I did also want to ask about some news that just came out. The Wall Street Journal reported that OpenAI is considering drastic price cuts in its competition for users with Anthropic. So how do you see the price wars between especially those two companies in particular playing out? [SPEAKER_02] Yeah. I think that OpenAI is lowering prices directly because of what we're seeing on enterprises spending too much money. I think that just simplest answer is usually right. I think that there is a lot of enterprises shifting. And if we're seeing it publicly, it means they're seeing it privately way more than we are, right? The second thing is, I think if Anthropic is probably going to IPO first, I'd say. [SPEAKER_02] So you want to lower prices and eat into their growth and margins as much as you can before they go public. You want to make their numbers look as bad as you can. So I think that's probably the second reason. [SPEAKER_01] Oh, okay. Yeah, that makes sense. Well, one other way that they either compete or that basically the competition between them could be affected goes back to some of the kind of paternalistic features that you mentioned earlier. You know, people have been complaining about how the latest releases will kind of like play cop or try to stop them from working on something. [SPEAKER_01] There was a viral tweet that went around where someone asked Fable about mitochondria and whether it's the powerhouse of the cell. Like flag the message saying like, you know, due to security features, like whatever. Anyway, like do you, it's such an innocent question. [SPEAKER_01] But anyway, do you also see them kind of competing in this way where this kind of big brother behavior kind of also changes and, you know, is another sort of turf war between them? [SPEAKER_02] Yeah, it's it. There's so many different stakeholders at play. Right. If you take the most the best view of anthropic in the world and you say this model is actually dangerous, like if we give this to people and they ask how to build a crazy computer virus that shuts down the entire world without being noticed and it could do it like that actually is a concern. And I get that. And we should we should protect against that. But generally, I view it as big brother fear mongering. [SPEAKER_02] And I think it's super annoying. The idea that we're letting some AI lab dictate what we're able to ask and why is a really slippery slope, because you just don't you don't want your you don't want to be told what you're able to research or look at or ask about. Right. Like, do you want to be in China? You're not allowed to ask about Tiananmen Square. Like, no, like it like people don't don't want that. The other thing, too, is. I just feel like. [SPEAKER_02] Having open source AI is the ultimate solution to the safety issues that everybody talks about. Right. Like, I just I don't view it as we should trust 10 people in a room as godlike as they can be. And I'm sure they're incredible people, but it doesn't matter. [SPEAKER_02] Because open sourcing and accelerating and giving everybody this technology at scale to bulletproof it and test it and audit it back and forth and battle test it is the same reason why Linux is the most secure operating system and beats Windows all the time. Right. Like you want this stuff open source. You want it to. Sorry. You want to get people up to speed. And that's the best way to protect the world. I'm very much in the accelerationist camp. I'm in the Beth J. Zos camp. [SPEAKER_02] But I understand why people might not be. [SPEAKER_01] So I actually this is the perfect segue to where I was kind of taking this conversation, which is, you know, the problems that we're looking at here. They are around, you know, open source versus closed source. Closed source. They're around censorship resistance. There's issues here with kind of fundraising or, you know, crowdfunding could be a solution. And those things are all things that crypto excels in. [SPEAKER_01] So, you know, reading about all these different issues, it goes back to that tragedy of the commons framing that I had before where I couldn't help but think, you know, crypto could really help with, you know, capital formation, creating decentralized systems like, you know, something that's open source, something that is censorship resistant. It basically feels like crypto could kind of crowdfund the development of those types of AI tools. [SPEAKER_01] And, you know, I know that there's already a bunch of different crypto projects that are kind of attacking this angle. So, you know, do you see that as being a possibility someday that it could actually solve some of these problems that we've been chatting about? [SPEAKER_02] Yeah, I mean, I definitely think it can be right. I think that a couple of I think a couple of crypto AI overlaps have not come into fruition. Right. Like decentralized training is a is a phenomenal idea in theory. Right. Like let's use devices all around the world. They're not being used anyway. And let's train models. Right. And it's something I used to be really bullish on. Right. I'm really not so much anymore. That's one aspect that I'm happy to see it at this point. Right. [SPEAKER_02] It's just too cost intensive and it's just too hard. Right. Google released DeLoco, which was a phenomenal training mechanism for around the world. And many projects did big training runs. Prime Intellect, News, BitTensor, Submet or two and a couple of others, which were great. And they released Decouple DeLoco, which is even better a couple weeks ago. But it's just really hard. But I think there's other areas where crypto and AI overlap. And at the core, it's at the ethos of open source. Right. [SPEAKER_02] And that's where it thrives. And I think the easiest way for I currently think the easiest way for that overlap to occur is one, exactly what you mentioned on capital formation. Right. Like, it'd be very easy to do an IPO or sorry, an ICO for a major open source AI company and build an incredible model, even if the model is being built centralized. [SPEAKER_02] If we could all own it and secure it and deploy it and that company makes money, which is tough because you have an open source model, how you make money? It is difficult. But I think that's the way to go. The other one is on the app side and the harness side. I mean, news research with Hermes has been an absolutely stellar release. And it's not a centralized lab. [SPEAKER_02] On any given day and for the past week and month, news Hermes agent is 3 to 4x the usage of OpenClaw and OpenRouter. It's an open source harness, not built by a lab. 150,000 GitHub stars and thousands of contributors continually improving Hermes agent. It's my daily driver on Telegram, on my desktop. It's huge. [SPEAKER_02] And once you tap into the open source global community, it puts projects on a flywheel and a velocity that's just really hard to catch. And people argue, like, well, that's not true. Like, the labs are so strong. But, like, it is true because Hermes agent is larger than OpenClaw, right? And that's, like, a key data point. And OpenClaw is incredible. But when you get the open source community involved, it just takes on a new life. [SPEAKER_01] Super interesting. So, last question. You know, you already mentioned news research. But, you know, as I also talked about, there's already a bunch of crypto players in this space. And I was wondering if you had opinions on kind of which ones could tackle which pieces of the various problems that we discussed. With, you know, Venice AI, which you've also mentioned being another contender, or BitTensor, as you also mentioned. [SPEAKER_02] Yeah. I mean, some of these we own through the Fonder, or I might personally, just as a disclosure, my CCO would want. I think Venice is doing an incredible job, right? I'm a huge Eric Voorhees fan. I think he's just good for humanity in general. But giving people open source access to these models at scale with privacy is unique, right? You just mentioned Fable, like, blocking people's requests. That's exactly what Venice is built for, right? [SPEAKER_02] To not block your question. I really like Grass. We invested in them quite a while ago. They're just an incredible company in getting, you know, using 3 million user nodes around the world to get real data for AI labs, right? It's a key Deepin use case that just sort of makes sense. We're really excited for Ambient. [SPEAKER_02] It's a proof of useful work AI project that will offer one hyper-optimized model really well. So there's a lot of areas here, like whether it's Deepin or Capital Formation or the app side, that are starting to take shape. But it's a very different world than what you and I saw, you know, 12 to 18 to 24 months ago, right? Where it was just like these AI reply bots with tokens. That is completely dead. [SPEAKER_02] And the only ones that will be able to survive are real infra projects, I think. [SPEAKER_01] All right. Well, Tommy, it has been such a pleasure chatting with you. Thank you so much for coming on Unchained. [SPEAKER_02] Laura, I love your show and you've just given the world like an awesome medium to just hear from really good founders and things like that. So I appreciate you doing this for so many years and it's a pleasure coming on. [SPEAKER_01] Thank you. Thank you. This weekend will actually be the 10-year anniversary. Damn. [SPEAKER_02] Wow. [SPEAKER_01] I know. I can't believe it. [SPEAKER_02] Time flies. [SPEAKER_01] Yeah. Seriously. Time flies when you're having fun. Exactly. All right. Well, everyone, thank you so much for tuning into this live stream and we will catch you next week. Bye for now. Nothing you hear on Unchained is investment advice. This show is for informational and entertainment purposes only. And my guest and I may hold assets discussed on the show. For more disclosures, visit unchainedcrypto.com.