Santiago Roel Santos 與 Logan Jastremski 深聊：加密貨幣與 AI 未來大對決

0:00投資者的自我突破與市場變化

0:00感謝大家來參加，今天這場對談我真的很期待。Santiago 是產業裡的重量級人物，根本不用多介紹。今天我們會聊很多關於加密貨幣和 AI 的話題，這就像我們平常的日常對話一樣。Logan 是我最常討論的人之一，他曾經說服我在 Solana 還 20 美元時賣掉所有 ETH，重倉 Solana。他真的很擅長把技術細節轉化成有投資價值的洞見，這其實很難做到。Appreciate everybody coming. Super excited for this chat. Santiago is someone that needs no introduction, really a titan in the industry. So looking forward to chatting all things crypto and all things AI. It's like any other day. Logan and I—Logan is probably the person that I talk to the most. He, for anyone that doesn't know Logan, he was the guy that convinced me to sell all my ETH and go long Solana when Solana was 20 bucks. He's probably the best person in the industry to compile technical into actually investable bites. And that's really hard.

0:48AI 現在已經成為區分投資人能力的最大關鍵，因為加密產業裡有太多雜訊和術語。懂技術是一回事，能把它轉化成投資論述又是另一回事。Logan 幫助我建立了對 AI 投資的看法，不只是在加密領域，這真的很難得。很多早期進入加密的人，包括我自己，都會有冒牌者心態，懷疑自己到底是不是靠運氣。這產業裡很多人自以為很會投資，但只要利率一升就原形畢露。現在市場回報差異很大，遊戲變得更難，必須真的很強才行。And AI has become probably the biggest differentiator in a world where there's a lot of garbage in crypto and a lot of jargon. But it's one thing to understand the technology; it's another to really understand how to make an investment thesis out of that. And Logan, we'll talk about it here, but he has been instrumental in my understanding of how to form an opinion and an investable case in AI. Not just in crypto, and I think that's really rare. Like, like good investor can do it in one industry, and I think a lot of people in crypto that were early, I certainly have it, an imposter syndrome of like, how do you know you're actually good in this game? Because there's a lot of luck. And I think a lot of stuff in this industry is most people think they're really good investors, and then rates go up and they're terrible investors. A lot of the stuff in crypto hasn't really worked out, and I think you're seeing a lot of really concentrated return dispersion, which is good. And the best part about the markets now is you really have to be fucking good now, because the game is just harder.

1:39你曾經跟我說過一句話我很喜歡，就是你沒有義務每天都當同一個人。如果你對某個技術有興趣，也可以同時對其他領域感興趣。投資人常常會想自己有什麼優勢，早期進入加密可能覺得有 edge，但現在已經不是這樣了。我現在完全改變想法，覺得好奇心才是最大的優勢。只要保持好奇，你就能在任何產業找到自己的定位。你本質上是個投資人，剛好投資加密貨幣而已。很多加密投資人不敢跨出舒適圈，覺得自己沒優勢，但其實只要願意學習，任何市場都能闖出一片天。And I think to that point, one thing that I love that you told me is you have no obligation to be the same person as you were yesterday. And if you're interested in one technology, you can also have other interests. Absolutely. Yeah, you know, I think as an investor, you sometimes, like I used to—think too much, like, what is my edge in investing? In crypto, it felt like you had an edge because you were early, and that no longer is the case, even though we reasonably could say that we're early. But I think I've totally removed that thinking and said, okay, my edge is just to be curious. And if I'm curious, then I think you can position yourself to be successful in any industry. So first and foremost, you're an investor that you happen to invest in crypto. But in that order, and I think a lot of my conversations—you're different in that regard—but I think a lot of people that have been crypto investors have resisted venturing out into other asset classes because they don't think they have an edge. And look, the game is really tough. It's more competitive. But I do think that if you're sitting there at your desk saying, I can't. I can't outperform. I can't be successful in traditional markets. I would highly encourage you to remove yourself from that thinking and just apply yourself.

2:53AI 與加密的投資機會與產業演進

2:53我們要從哪裡開始聊？要談加密還是 AI 的未來？不如直接聊 alpha 吧，這也是我們每天都在聊的主題。其實我們今年初就對 AI 很感興趣，真正的轉捩點大概是 Opus 4.5 出來之後，AI 商業模式從訂閱制轉向 API 收費，帶來營收大爆發。這對整個 AI 產業來說是個很大的警鐘。大量資本投入到資料中心，股票市場也開始重估 AI 相關公司。我們也開始從零到一思考這個市場的未來，特別關注記憶體這一塊。So where should we start? Should we talk about crypto? Should we talk about the future of AI? Let's just talk about alpha. Let's do it, which is any other conversation you and I have every other day. That is true. So maybe on the AI side, we... You and I got very interested in it early this year. I think the big moment was really around Opus 4.5, where you started to shift primarily from these subscription-based models to really API-based, which really caused this explosion in revenue. And that was a huge wake-up call for, in broad strokes, the AI industry. A lot of CapEx has really gone into the ground, building out these large data centers. And obviously, a lot of the equities markets has started to reprice on the AI side. And so you and I have been spending more time really going zero to one on where that market is going to go, and we got very interested in memory.

3:54你之前在 Empire Pod 也聊過這個話題。我記得你今年初打電話給我，叫我不要看圖表，直接聊 AI 投資重點——記憶體。結果我還是去看了圖，發現價格已經噴上天。這讓我想到早期比特幣從 1 美元漲到 50 美元時，很多人都覺得自己錯過了行情。AI 也是一樣，你可以被圖表嚇到，也可以回到第一性原則，思考真正的趨勢和基本面。其實加密和 AI 有很多物理層面的共通點，比如你要理解 Solana 跟 Ethereum 的吞吐量差異，這會影響價值累積。去中心化資料庫的物理瓶頸也是金融應用很重要的基礎。這些工程思維在 AI 產業也很適用。You've been talking about that on Empire Pod, and we've had some really awesome chats there. Yeah, yeah, I remember you called me earlier this year and said, Look, do not open a chart. I'm going to talk to you about what is an interesting component of the AI trade, and you talk about memory. And of course, the first thing I do as I look at the chart, and it's absolutely parabolic. But it's a good lesson because I think in crypto. You know, I remember, you know, in different moments, like you would have looked at the price of Bitcoin when it ran from a dollar to 50, and that was 2012, and I started looking at crypto, and it was like, oh, I've totally missed this trade. And similarly with AI, I think, you know, you can look at the chart and be totally jaded by it, or you can say from first principles, which—It's like, okay, what is the actual trend here, and what is the fundamental case for something like memory? And there's a lot of parallels. I think you do a good job of, like, maybe we can talk about, like, if you have been in crypto, you've had to understand the physics of it, which I think you're very good at. Like, what is the throughput of something like Solana versus Ethereum? How does that lead to value accrual? What is the actual physical bottleneck of decentralized databases, if you think about financial use cases? And I think a lot of—there's certain parallels. We don't want to reason by analogy, but there are certain parallels that I think are highly applicable to how we think about just viewing the world through that lens.

5:32對現場和線上的朋友來說，加密產業到現在最有趣的問題就是：怎麼讓產業規模化？2017 和 2021 年時，鏈上交易成本超高，大家都在想怎麼擴容。Ethereum 推出智慧合約，然後有 L2、L3、分片等各種設計，還有像 Solana、Aptos、Monad 這種高吞吐量鏈出現。這些都是加密產業成熟的關鍵，讓我們從 NFT、迷因幣這種高流動性資產，走到現在金融市場的極窄價差，甚至鏈上股票市場。這些工程思維其實跟 AI 很像，資料中心也有吞吐量瓶頸，像一個機架裡有 72 顆 GPU，還是會遇到網路瓶頸。你要怎麼讓 token throughput 維持高效？像 Groq 或 Cerebras 每秒 1500 token，但容量有限。用工程角度去看，甚至問 LLM 來拆解問題，就能避開表面投資人常犯的錯誤，找到真正的投資機會。Yeah, 100%. And maybe for—all those in the audience and that eventually will watch this later. I think the interesting part of crypto really up until this point has been, how do you scale the industry? I mean, a lot of you were probably around in 2017 and 2021. It was very expensive to actually transact on a blockchain. So really one of the main use cases was How do you actually scale these different technologies? We started to experiment with a lot of different designs. We had Ethereum come along, make smart contracts. You had L2s, you had L3s, you had different technologies to kind of shard to increase the total throughput. And then you also had kind of high-throughput blockchains like—Solana, Aptos, Monad kind of broadly appear. And I think, to me, that was really an important part for crypto to mature and now get to really what is the next kind of iteration of crypto, which is financial markets. And I think you really had to kind of graduate and scale so that we could go from NFTs to memes, which were even more liquid, to now really very tight spreads with generally one to five bps of execution. And now you're starting to see blockchains move into the equities markets. And so there was a lot of, like, zero to one work that had to be done, but If you kind of look at it more from this engineering lens, I do think it is very applicable to AI. As you look at data centers, there's different throughput issues. One of the reasons why we started getting very interested in memory is because, or even just broadly tokens, as you look at kind of a server within a rack, there's 72 GPUs. They're trying to expand that, but There's a lot of networking issues even within a rack within a data center. How do you keep token throughput very high, where you go from something like a Groq or Cerebras that have 1,500 tokens per second, but very limited capacity. And so I think if you start to look at it more from an engineering lens and For you that are not engineers, just ask some of these very smart LLMs. They'll help you kind of parse these things apart. You can start removing a lot of the things that I'd say surface-level investors do, and you can really get to the heart of how these things will eventually scale and, I think, make some interesting investment decisions from that.

8:18AI 與加密產業的交集與分歧

8:18大家來這裡應該都想知道加密和 AI 的未來。最近有個現象很明顯，就是大型比特幣礦工都轉型去做 AI，因為算力是最稀缺的資源，他們要把算力用在 AI 上，回報率更高。像 Core Scientific、CoreWeave、Crusoe 這些公司都在做這件事。現在大部分資金和注意力都流向 AI，這是合理的。以我自己經營 Inversion 的經驗，原本想當控股公司，買企業、導入技術，最早重心在加密，像穩定幣、資產代幣化，我們花了九個月研究，最後發現加密雖然是好技術，但創造的價值有限。AI 則很具體、很早期，感覺潛力很大。Yeah. Let's talk about the, like—I guess you guys came here to think about what is the future of crypto, if there is one, and what is the future of AI. A couple of things that are top of mind. One, some of the larger Bitcoin miners have totally transitioned to being focused on AI, and that just tells you the opportunity cost for them is, you know, compute is the scarcest resource, and they're going to apply that to intelligence, and that has been AI, if you look at. Core Scientific, you know, CoreWeave, you know, Crusoe, all of them now. If you're provisioning compute, which is the scarcest resource, you're going to go and do that for AI because there's just much higher ROI. So that's, like, top of mind in terms of how I think most of the flows, most of the attention is certainly going in AI, and rightfully so. I mean, it's incredibly—I'll tell you, from our perspective in Version, I started it thinking, hey, as a holding company, let's go buy businesses, let's implement technology. Most of that was initially the focus was crypto, like stablecoins and tokenization. We spent like nine months in that. The conclusion of that was crypto is a good piece of technology, but it doesn't lead to that much value generation. I'm in the business of—Like releasing value, not just proving to the world that we can implement stablecoins. I think we're far along in that journey. But AI is very, at least from my perspective and the way I run my business, it's very tangible, and it feels very early.

9:46大家應該都看過那張圖，A16Z 做的，全球用 AI 的人還不到 5%。大部分人對 AI 的印象還停留在聊天機器人，不是像 Claude Code 這種工具。AI 剛開始是聊天機器人，像有個永遠不會說你笨的家教，讓學習變得有趣。這種互動對教育和自學很有幫助，未來如果有勞動力被取代，也能用 AI 重新訓練。然後進階到工程師，像 Claude Code，Anthropic 是史上成長最快的公司之一，年營收破億美金，連 Salesforce 都還沒做到。這種成長速度很驚人，投資人都會想機會成本，AI 現在真的還很早期。I think we've all seen that chart of, like, I think A16Z put it out. It's like, you know, there's still less than 5% of the world's population is using AI. If you were to survey people what AI means to them, most of them talk about being a chatbot, not like Claude Code. So, like, the way I think about it is, you know, you talk about Opus 5. Like, first it was a chatbot, which is like everyone had a tutor that would never tell you you're stupid and would make it, I think it's the first piece of technology that has made learning quite fun. And the interaction, the dialogue that you have, yes, it could lead to psychosis, but it's actually quite, quite novel when you think about applying that to education and just self-learning. And if there's going to be labor displacement, you want to retrain people. I think, like, you know, that is hugely impactful. Then you had, like, an engineer, and for someone that is not technical, I mean, that's just the Claude code is—Number one, Anthropic is the fastest largest, like the fastest scaling company to ever go to like 100 million of recurring revenue. I mean, it took Salesforce. Like Salesforce is not even there, and it's like 10 years in. It's kind of crazy when you think about that. Like, we were talking backstage about, as an investor, I think we always think about like what is the relative opportunity cost, and At least from my perspective, I think we're very early in that transition.

11:06我們現在正處於 AI 的關鍵轉折點，越來越多人開始用 AI agent，等於有一堆數位員工隨時待命。AI 的普及甚至比加密還早期，大家對 AI 的理解還很有限，爭議性也很高，像資料中心、AI 的負面觀感等。坦白說，我現在對加密的未來反而沒那麼有興趣。很多人說加密和 AI 有交集，但我覺得其實沒有那麼明顯。And we can talk about the specific numbers, like, you know, and I think we're entering a really interesting inflection point of AI where people start using it with agents and, you know, basically have all these employees at their disposal. It feels earlier than crypto in many ways, like the understanding of AI and what that means and—It's controversial in the same way that crypto has been controversial, like data centers and people have, like, a negative perception of AI. So that all bodes really interesting, in my opinion, of I'm candidly, like, less interested in the future of crypto. And I think there are probably interesting things that we should dispel. Like people talk about, like, there's this intersection of crypto and AI, and I'm not sure it's there, to be honest.

11:50加密產業的現實與商業模式

11:50其實加密產業一直到最近才有比較明確的商業模式。2021 年大家都在各種實驗，Web2、Web3，但 Web3 其實沒真的做出產品或營收。你常在 Empire 談這個問題：到底機會在哪？更本質的問題是，哪些項目真的有營收、可以用商業角度評估？我認為加密其實是新一代金融科技和資本市場的進化，現在是全球舞台。真正能賺錢的加密項目，已經不是單純發 token 賣 token，而是做出有持續性的生意。大多數還是金融相關，像交易所、穩定幣、永續合約等都做得不錯，因為加密本質就是金融基礎設施。Yeah, and I think maybe to your broad point, I think the hard part with crypto, really up until more recently, has been we had a lot of experimentation in 2021, and in large part the industry said it was Web2 and then Web3. But Web3 never really materialized from, like, a product or revenue perspective. And you chat again on this quite a bit at Empire. It's like, where is the opportunity? But I think the more kind of meta question is, where are the revenues being generated that you can actually underwrite from a pure business standpoint? And when you look at crypto, it really, in my mind, is this kind of next evolution of fintech, the next evolution of capital markets. And really that's happening at a global stage. So if you look at the opportunities within crypto that are making revenue and kind of don't just launch a token and sell the token, but actually make more durable businesses. And large part, that's finance. You see exchanges doing fairly well. You see stablecoins doing well. Perpetual products obviously doing well. But that's because crypto is really more of a financial infrastructure than anything else.

13:07AI 這邊也開始有很多公司賺到錢。投資人要思考的重點是：什麼是能長久經營的生意？我自己在加密創投這幾年，投了超過 160 個項目，幾乎每週都在投。創辦人素質越來越高，很多華爾街的人跳進來，像 Solana、Hyperliquid 這些基礎設施都很有趣。金融市場很大，聚焦這一塊沒錯，但 AI 的機會更大。像 Anthropic、OpenAI 都已經做到年營收破億美金，企業採用還超級早期。投資時我可以喜歡技術、討厭價格。很多會議上的討論都是因為大家只被允許投加密，所以只會比較不同鏈、不同項目。但如果讓大家選，要買 3000 億美金的 Ethereum，還是同價值的 AI 實驗室？十個有九個都會選 AI，這就說明了相對潛力在哪裡。And I think on the flip side with AI, you are seeing, obviously, these businesses start to make a good amount of money as well. And from that standpoint, I think the opportunity cost is really what are the durable businesses? I mean, great question. I find it, as someone that has been very active on the venture side in crypto, I've noticed like a couple of trends. Like I went from the last five years investing in like over 160 projects in crypto, like every other week I was investing in one or two projects, I think. The quality of founders was interesting. You started to see more people leave traditional Wall Street jobs, say, hey, there's something real here. There's like infrastructure, whether it's Solana or something like Hyperliquid. There's super interesting primitives. Don't get me wrong. I mean, financial markets are massive. I think it's a good thing that we're narrowing that scope. AI just captures way more opportunity, in my opinion. But in a testament to that system, Anthropic and OpenAI both have, like, reached basically 100 million ARR. And you're still at a point where, like, you're not even fully—you're basically super early in the enterprise adoption. And so, again, like, in investing, it's, it's, I don't have to. I can love the technology, I can hate the price, and I think. A lot of the conversations that happen in these conferences are people that are, their sole mandate is to invest in crypto. So, of course they're going to compare one crypto project to the other, or one L1 to the other. But, you know, I think, like, if you actually, like, a lot of the times it's like, would you rather own, you know, for instance, like Ethereum at like 300 billion or, you know, an AI... Lab at, like, the same valuation, I think most people would say, yeah, I would definitely buy the AI. You know what I mean? Like, nine out of ten times or ten out of ten times, privately they'll tell you, for sure I would buy the same asset, you know. And I think that that really just sharpens the conversation of where you actually see the relative potential.

15:11加密金融產品與 AI 的普及化

15:11過去大家都說加密是價值儲存工具，但其實很多資產的價值儲存說法都很牽強。因為如果你用現金流折現法來算，根本不合理。所以我認為要把加密當作金融產品來評估，能產生營收的才有意思。AI 這邊則像水一樣，會被整合進越來越多領域。兩者都是很重要的技術，值得同時學習。現在的加密市場跟 2017、2021 年完全不同，設計路徑已經探索得差不多，現在聚焦在金融基礎設施是好事。Yeah, and I think, like, crypto in the past was—Store of value, kind of, in large part, I think the store of value for most assets is kind of a farce. But you told yourself that these were store of values because if you tried to do a discounted cash flow based off revenues, these things made absolutely zero sense. So I do think if you separate them from crypto as finance and start underwriting them as financial products that do have the capability to make revenue, that is very interesting. But also on the AI side, it is becoming almost, as most people put it, kind of water, and it's being integrated into more and more things over time. That is also going to be a very important technology, and I think it's okay to also learn about both. And so. Yeah, it's an interesting time. Crypto markets are definitely evolving. It's much different than it was in 2017 or even 2021. And I think we've kind of explored the design maze, so to speak. And I think it's a good thing that crypto is really landing on kind of the financial rail side.

16:19回到你剛剛問的，現在加密和 AI 有交集嗎？我在 Frictionless Capital 找投資標的時發現，真正有產品市場契合的公司，像 Nous Research，原本做去中心化訓練，後來發現難以規模化，就轉去做 Hermes agent，現在在 OpenRouter 上表現很好。Together AI 也是，從分散式推論轉型做雲端。這些原本在加密和 AI 交集的公司，最後都專注在 AI 本身。現在這個交集其實很小。大家常說 agent 需要穩定幣來交易，其實不太對。即時結算這個特性，只有在交易時才真的有用，因為要降低對手風險。如果只是訂機票，根本不需要即時結算。穩定幣是很棒的基礎元件，但不代表 agent 一定會用，因為智慧合約風險很高，像北韓駭客就很會攻擊加密系統。AI 讓這些系統更容易被破解。信用卡有可逆性、chargeback 等特性，這些都是重要的基礎設施。很多人以為技術進步就會改變人類行為，但事實並非如此。我自己也曾經因為違背人性而賠錢。穩定幣流通量要成長 100 倍，可能還是得靠金融應用。But maybe to your question earlier, is there an overlap between crypto and AI? Because I think it's a good question. One thing that I have broadly observed trying to find these different companies at Frictionless Capital to invest in is, in large part, the companies that have achieved product-market fit, if you look at Nous Research, for example, they originally started out doing decentralized training. But over time, they found that it was hard to scale that business and then pivoted to the Hermes agent, which is now, I believe, the top agent harness on OpenRouter and doing fairly well. Another example is Together AI, originally started doing distributed inference, but has really pivoted to an EO cloud. And so I think, at least for the products, again, if you look not at the token, but ones that are actually producing revenue within the crypto AI field, in large part those businesses that started at the intersection have moved completely to AI alone. And so I think, to me, trying to be a student of the capital markets, you kind of have to see where the founders are going and—The overlap today looks relatively minimal. One of the things you'll constantly hear is people say that agents need stablecoins to transact, which is largely untrue. And one of the things I've changed my mind on recently is this idea that, like, instant settlement, like for what exactly? Because I think you hear, like, the properties of decentralized finance, stablecoins, instant settlement, and yes, you reduce counterparty risk, but, like, I think it—The nuance here is for what. If you're doing trading, yes, because you want to minimize counterparty risk. If I'm trading with you or you or you, I want to do instant settlement because you can, you can, there's finality. You know, you don't have to do T plus two. You remove counterparty risk. So that's really valuable for a trading application. But if you're booking a flight, it doesn't matter. So I think there's—I'd caution people when they think about, you know, this. Yes, stablecoins are great primitives. They're quite useful, but that doesn't mean that an agent will use—maybe the agent will reason and say there's actually a lot of smart contract risk, because the thing that, of course, has hindered crypto is, you know, North Koreans have gotten really good at hacking crypto. And, and so I think with, there's, there's, I would argue less Lindy now with AI in some of these crypto systems, because you can apply just a whole bunch of compute and intelligence to hack through some of these systems. And so I think, you know, even simple flows, like with a credit card, you know, you have the reversibility of a credit card transaction. You know, chargebacks and some of these components are important primitives. And I think some of the biggest mistakes I've done in investing is just thinking that because the technology is so advanced, that will change how humans actually interact. And that's just not true. And I think I've lost most money, like, going against human, you know what I mean? And I think that's sort of the temptation when you're—You're trying to wrap your head around the thesis of, like, how does stablecoin circulation go up like 100x from here? And I think it's certainly possible, but it will likely come from some of the stuff you and I were talking about this morning, which is mostly in, like, financial use cases. And, you know, there's interesting stuff like on the margin there, but some of the things that you in the crypto industry. Because we're a bit defensive, we have to just make these very hyperbolic statements of how this technology would just be applied to absolutely everything. And sometimes going faster is not a good thing.

20:08AI 實用性與數位同事的未來

20:08現在很多新模型像 Astra，已經很會操作電腦了。你可以叫模型幫你做事，甚至不用 API 或 MCP 連線，直接用滑鼠鍵盤操控電腦。這其實是很大的創新。我最近重啟 podcast，原本用 Autopod 軟體自動剪輯，現在直接請 Claude 或 Astra 幫我剪，完全不用連線，整集自動完成。這已經是最差的狀況，未來只會越來越好。從大方向來看，我們最期待的就是 AI 變成數位同事，真正成為團隊的一份子。這也是我們為什麼關注記憶體和 AI 資本支出的原因。資料中心一座要 500 到 700 億美元，未來還有多少新機電要上線？從預訓練（主要靠 GPU）到推論（主要靠記憶體），投資人可以找到很多新機會。只要保持好奇心，世界一直在變，機會永遠存在。Yeah, I think, I mean, even playing around with more recent models like Astra, they've gotten very good at computer use. You can tell the model to go do something on your behalf. And you don't even need an API connection or MCP connection. It can literally take over your computer and click with a mouse or kind of keyboard, so to speak, and type. And this is like a pretty profound innovation. I recently restarted my podcast, and I used to have this software called Autopod. Where it would essentially edit the podcast for me based off whichever speaker is talking. But more recently, I can have Claude or Astra go in, do computer use without any connection. And it edited the entire podcast. And you kind of think about it, this is the worst it's really going to be. It's only going to get better. And so maybe, like, high level, I think the thing that we're broadly excited about, if you kind of think about where AI is going on the extreme side of things, is just kind of digital coworkers and really kind of being part of your team. Which has generally got us excited about kind of memory and kind of the AI kind of capex that's going into the ground. You look at how much data centers cost, they're 50 to 70 billion dollars generally. How many new gigawatts are coming online? You can kind of do the back of the envelope math as you shift from pre-training to inference. Pre-training mostly GPU-bound. As you switch to inference, it's a lot more memory-bound. And there's a lot of interesting things as an investor that I think you can parse apart. And so I think as we always chat, like, you got to stay curious and Interested in learning about these things because there is a lot of opportunity and the world is always very dynamic.

22:02明年 AI 資本支出會達到一兆美元，這已經是加密總市值的一半。你要搭上最大浪潮，現在最大浪潮就是 AI。這不代表加密沒機會，只是 AI 的規模和產業重塑力道太大。過去我覺得加密是一場社會經濟實驗，像 Ethereum 這種技術創造了最多百萬富翁，而且是開放的。AI 現在也開始討論開源 vs. 封閉模型，這跟加密的開源精神很像。你會更珍惜開源，但也不是說封閉模型就會歸零，兩者會並存。AI 的部署規模真的很驚人，這就是所謂的大浪潮。I mean, just put it this way, there's a trillion dollars of capex that are going to be deployed in AI next year. That's basically half of crypto's market cap. You always want to be riding the biggest wave. Crypto is no longer the biggest wave, and attention has shifted dramatically to AI. That doesn't mean there's not opportunity in crypto. It's just—This is the sheer magnitude of this reindustrialization. And, you know, for a while I thought crypto was like a really interesting socioeconomic experiment because, you know, a piece of technology like Ethereum created the most amount of millionaires ever across a wide variety of people. It's open. There's something we haven't talked about in AI that is becoming more and more of a topic of conversation is this idea of open source models versus closed source models. I think, again, this just goes back to the point of you've been in crypto, you've seen this evolution of, like, open source versus closed source, and generally, you know, you have a greater appreciation for open source. It's not as simple as all the closed source models are going to go to zero and everything's going to shift to open source, right? But, but yeah, I think it's, it's, it's just kind of impressive to look at the deployment of AI and from that standpoint, you always talk about riding big waves. There is no greater wave than AI.

23:14全球資本市場、代幣化與 AI 的結合

23:14我們在 Frictionless Capital 的看法是，美元其實是最早被代幣化的產品。美國一開始對穩定幣很排斥，後來發現穩定幣會買美債，才覺得這對美國有利。現在你看到鏈上股票市場開始出現，等於把美國資本市場輸出到全世界，同時也把其他國家的股票引進來。這其實是把全球不同金融系統用一個 API（區塊鏈）統一起來，市場規模非常大。即使把目前鏈上股票的資產管理規模放大 1,000 倍，也僅約美國資本市場的 6%，還很早期。至於加密和 AI 的交集，目前看起來還是很有限。AI 本身則是超級大浪潮，值得學習。不管是資本支出還是投資趨勢，像長期看好 agent、數位同事，甚至未來到機器人領域，這些都是很大的市場。And how we really think about it at Frictionless Capital is that as you kind of started with the dollar really being kind of the main product or tokenization of the dollar. Originally, the United States was not super excited about stablecoins, and then it realized that it was buying treasuries, and they're like, okay, this is actually great for us. And now I think you're starting to see that within the equities market, and you're exporting the U.S. capital markets, but you're also importing other jurisdictions around the world where you can have their equities, and so. To us, how we broadly think about it is you're taking all these kind of countries around the world that have these different financial systems and unifying them with kind of one single API, which is the blockchain. And I think if you just look at that, it is a massive market that will broadly continue. I think if you look at the amount of tokenized equities today, In terms of AUM and 1000x that, it's about 6% of just the United States capital markets. So we're still fairly early in kind of this broad tokenization trend. But then again, if you look at that intersection of crypto and AI, at least today, it seems relatively limited in the intersection. And then you look at only AI, and as you point, it's a massive wave, and I think. Learning about that, both in terms of CapEx, where that's being deployed, also understanding how to go long certain trends, whether that's going long agents, how you ultimately express that in the sense of if you can think about it in the most extreme scenario, what is the most useful thing kind of an LLM can do? is kind of act as this digital coworker until you go physical, and then once you're physical, then it's kind of robotics, which is a separate field. But these three kind of buckets, or maybe just broad two buckets, which is AI and kind of global tokenization, massive market. And then you have AI, which is kind of really rethinking a lot of what we have done in the past, and as kind of knowledge work.

25:22AI 與加密的未來：金融普惠與市場流動性

25:22講到大趨勢，無論是加密還是 AI，兩者都很值得關注。加密有開放市場，資本會流向最有效率的市場，流動性會形成正向循環。AI 如果有 agent 能幫你理財，雖然我不是建議大家這樣做，但你可以想像，全球有很大比例的人對金融完全沒概念。最大指標就是有多少錢閒置在活存帳戶沒生利息。大多數人沒有金融素養，所以如果 agent 來幫你管理資產，對大眾來說會更理性。全球有很多人無法直接參與資本市場，像美國這種金融體系。加密的代幣化股票、穩定幣讓大家能用美元進入全球金融系統，這對加密經濟某些領域非常有利。AI 則加速這個趨勢，讓更多人能用 agent 跨境投資。Great points. I mean, when I think about, like, megatrends, both crypto and AI, and I'll mention some in crypto and then AI. With crypto, I think you have these open markets, and capital wants to flow to the most efficient markets. That's just a normal gravitational pull. I think crypto stands a greater chance to become that venue of liquidity, and more liquidity begets just—it's that flywheel that we're in. As it relates to AI, I think if you have an agent that you trusted with your money, I'm not suggesting you do this, but let's just assume that you run this simulation, you get to a point where an agent There's a huge level of financial illiteracy in the world, and the number one statistic that you should look at is the percentage of money that sits in idle checking accounts that are not earning interest. It is remarkable. Most people don't have financial literacy, and so the upshot here is, in a world where the agent is administering your money, it will make more rational decisions for the vast majority of the population. Now, most of you might be a bit more sophisticated than the average population, but you have to consider, in both crypto and AI, the bull case is there's just a huge number of the world's population that Does not have access to financial markets or investing products or capital markets like the U.S. So that's tokenized stocks on the crypto side, right? You have stablecoins that allow you to on-ramp to the dollar and then have access to a dollar-denominated financial system that most of the world's population has wanted to get access to but has not had access to. That bodes extremely well for. a certain pockets of the crypto economy. And as it relates to AI, is just an accelerant to that force because of the agent is saying, look, you live in a place like, say, India or Mexico, that you can't have access to U.S. capital markets directly through, let's say, a brokerage account, or you needed to be a high net worth individual. Now you can do that with an exchange like, I don't know, Backpack, for instance. You can on-ramp with a stablecoin and you can buy. A tokenized stock of Sandisk. And that just closes the loop. And I think, from my standpoint, I think the financial services industry, I think the bold case for both crypto and to some extent AI, certain services of AI, is the financial services industry has probably been the industry that has resisted the most amount of innovation. Like your financial experience is terrible relative to anything else in your life. You do not have an Amazon-like experience with your bank, the way you move money, the way you interact, and it's difficult, right? And so the result of that is most people don't like to think about money. But in a world where that is further abstracted away and someone that can do that for you, you're going to gravitate to rails that are open, transparent. And efficient. And I think crypto stands. I'm not saying it will be. I'm just saying it stands a greater probability to be that venue. And that has a whole host of implications in traditional markets where anyone that has built the model around obfuscation and friction is going to basically go to zero over a long horizon. We like things that are frictionless. But maybe to round this out.

28:32總結一下，雖然我們在加密會議，但世界一直在變，你要保持動態學習。大家都對加密市場有興趣，但也要關注新浪潮。AI 現在還在早期拐點，加密則從野蠻西部慢慢變成傳統金融。未來世界會被代幣化，AI 也會成為生活更大的一部分，兩者都值得持續追蹤。兩者的滲透率都還不到 5%，機會還很大。今天就到這裡，謝謝 Santi。But maybe to round this out. I think more of the story, in my opinion, we're obviously at a crypto conference, but I think the world is changing and you need to continue to be dynamic and kind of wrapping it back, obviously. Everybody here is interested in the crypto markets, but you should continue to learn about the world as it's dynamic, as it is changing, and make sure that you're also very Interested in following the new waves. And obviously AI is really still early at its inflection point, and crypto has been the wild, wild west for a while. It's now kind of becoming traditional finance. And so I think as you kind of tokenize the world, keep an eye on for interesting things that you can do with AI as it becomes a greater part of our lives. Amen. Both are extremely early. And in terms of adoption, like less than 5% of the world's population is interacting with crypto and using AI. So, perfect. We can wrap it there. Thank you, Santi.