AI 詞元經濟來襲：Primitive、Post Fiat、USD.AI、Selini Capital 談數位資產

0:00AI 吸金潮與加密市場現況

0:00早安，感謝大家一大早就來參加。今天這場討論一定很有趣。我認識台上的三位來賓，他們都是 AI 與加密領域的頂尖思想家，今天我們會跟大家腦力激盪一下。這場應該會有不少吐槽，Alex，我在看你喔。不過我想先從大方向聊起。AI 開始吸走大量資本時，大概是一兩年前，當時加密市場出現了第一次真正的下跌。不過現在看起來，這波動已經穩定下來，動能也差不多到底部了。希望如此啦，不過 Dovey 不同意。那我們是不是完蛋了？還是只能一路向上？Good morning to all the early risers. This is a very fun panel. I know all three of the panelists, they're obviously some of the top thinkers between AI and crypto, and we're going to spitball for you guys. There's a lot of—Ranting probably going to happen. I'm looking at you, Alex, but I'd love to kind of just start with the broader topic because when AI first started sucking all the capital out, it was the first sort of real drop we had in crypto about a year or two ago, and I think that's stabilized at least in terms of momentum. I think at this point we're bottoming and hopefully— No, W disagrees. Okay, are we screwed? Up only, okay, bro, like up only.

0:55AI 現在絕對是一路向上。你看 AI 詞元（token）的支出，完全是爆炸性成長，採用率也一樣。我們剛剛還在聊 Muse，這個應用下載量已經衝到排行榜前幾名，僅次於 Instagram 之類的。Alex 這邊也一直從 Facebook、Meta 那邊搶走流量。所以我覺得應用面真的很強，RSI 也在往上走，AI 詞元經濟完全是 up only。Okay, like AI is definitely up only. Like if you, like, look at, like, the token expenditure, it's definitely up only. Like the total volume of the token is just exponential growth and, like, adoption as well. We were just talking about Muse, right? Like, Muse has been on the top chart when it comes to application download. Like, right, it's probably the second, like the fattest one to Instagram or like whatever. Yeah, so like Alex has been just stealing mills from like Facebook, from Meta quite a bit. Yeah, so I think it's definitely the application and also. RSI coming up, it's definitely up only when it comes to the alternative token economy.

1:34AI 的發展還在加速，但感覺加密的比重已經穩住了。現在我們看到 AI agent 的出現，可能會帶來很多新使用者。以前大家用 MetaMask 切換鏈、搞 gas，真的很痛苦。Alex，你對 agent 這個應用場景比較看好，可以多聊聊嗎？I mean, certainly AI continues to accelerate, but it feels like crypto's ratio is stabilized. And at least now, when we're looking at AI agents, potentially we have a lot of new users that, you know, historically humans trying to click through MetaMask to—Change chains and figure out how to get gas. It was just a disaster. So, Alex, I mean, I know you're kind of more bullish on the agent use case. Can you talk about that a little bit?

2:03Agentic Trading 與使用者體驗革新

2:03我們從消費者加密應用學到的一件事，就是要讓使用者真的用起來非常困難。以 Muse 為例，Stripe Link 月增 38 倍，這其實就是支付抽象化。當使用者越來越遠離實際交易細節，他們就越不在意底層有多複雜。像現在如果要從 Arbitrum 跨鏈到 Hyperliquid，只有極少數高手才會做。但未來使用者根本不會知道背後發生什麼事。最終的 agentic 加密牛市，就是大家根本不知道有加密在運作，但後台其實全都是，這也會帶來大量手續費。傳統金融市場的結算清算流程設計很差，根本無法支撐 agentic 交易。未來很多交易會自動發生，靠穩定幣市場和 UX 抽象化推動。Yeah, I think the one thing we've learned from, you know, consumer crypto is it's very, very difficult to get traction on virtually any crypto application that's user-facing, and—When you look at, you know, Muse, what W was mentioning, Stripe Link is up 38x month on month, and it's basically payment abstraction. And so as users move further and further away from the actual transaction, they become more and more indifferent to the complexity of the transaction. So it's like, okay, like if an agent has to bridge to Arbitrum to transfer to Hyperliquid, you know, only a power user would do that. In the status quo, but in the future, users won't even know what's going on. So, you know, the ultimate sort of agentic, you know, crypto bull case is that people don't even know that there's crypto involved, and it just is involved on the backend, and that results in a lot of fees. I think, you know, if you looked at DTCC settlement clearing and, like, you know, traditional financial markets, you're like, is this The way that transactions are going to be cleared in the future, it's like almost certainly not. It's very, very poorly designed, and it can't scale to agentic transactions. So I think the very boring crypto bull case is that nobody even realizes that a lot of these transactions are happening, but they happen because of the stablecoin market and just the UX abstraction.

3:24其實大家早就預期 agent 會爆發，去年、前年都說是 agent 之年，現在終於開始發生了。像 Robinhood 你跟我說有 15 萬個 agent 在運作，也許明年就會更普及。另一個現象是，越來越多人會被 AI『一擊斃命』，像數學家最先被取代。當大家發現 AI 做的工作越來越多時，投機、重押資產的機率也會上升，COVID 期間我們就看過這種現象。現在大多數人接觸到的內容還是人類產生的，但如果網路上內容都變成 AI 生成，資產的獲客漏斗也會變化。Michael Saylor 就是個有趣例子，他現在 AI 影片產量超高，影片品質也越來越好。這會讓加密原生玩家找到新的獲客遊戲規則，帶來新一波資金流入。You know, agents is something that we thought was going to be the year of the agent two years ago, last year. It's starting to finally happen. I guess Robinhood is, you were telling me, has 150,000 agents doing stuff. So maybe we're going to get there next year? I think the other interesting thing is that more and more people are going to get, I suppose, one-shotted. By AI, like, you know, all the mathematicians got one-shotted first. You know, there's kind of like, wait, what is the purpose of life if I can't do math and agents are better than me? I think as people become more and more aware that AI is doing larger and larger percentages of their work, the likelihood for, you know, hyper-speculation or betting the ranch on, you know. Speculative assets will increase. We saw that in COVID, for example. And so I think generally, the other interesting acquisition funnel is that your typical user is currently encountering content which is human-generated. And if you, if you basically allow all content on the internet to be AI generated, then the acquisition funnel for different assets will, like, you know, actually Michael Saylor is like a funny example of this. Like, he's the most active, like, AI video generator, right? Like, he just posts nonstop videos of himself. And at first, the videos weren't very good. And now the videos are getting, like, better and better, right? And SRC is, like, trading at par now, right? So there's kind of like a world in which, you know, crypto natives figure out this new sort of, call it investor acquisition cost game, and it drives another inflow.

5:06AI 與數位資產的互補效應

5:06如果我們用比較正面的角度來看，AI 和數位資產其實可以互相加分。那哪些數位資產會受益？像以前的 Litecoin、Dogecoin 這些，未必能從 AI 得到好處，除非它們真的跟上底層技術。很多人把這個領域叫做 agentic commerce，但我覺得大家誤解了 agent 的目標，其實是要 cover 計算成本。如果用 agent 比直接做還貴，那就沒意義。現在隨著運算力提升，產出也出現拐點。像 Mythos 是第一個用 Blackwells B 系列的模型，之前都是 Hoppers，花很久才跟上最新算力。現在 OpenAI 用 GB300、GB200 解決 Navier-Stokes 方程，電腦能力已經強到快要讓每個人都變超人。音樂、支付這些只是小菜，真正厲害的是大家都能用超級電腦解決問題，這還沒發生，但很快就會看到。And David, I mean, if we want to take the more positive approach and—We'll let Dovy speak to the bear case, but the more positive approach that these two technologies are going to benefit each other, digital assets will do well. Which types of digital assets, let's say compared to, you know, pre-AI era where you have the Litecoins, the Dogecoins, like what might be the new assets that outperform? I'm not sure how much those existing coins will benefit from AI, unless they're actually keeping up with utilizing the underlying assets. Because I think a lot of people describe the sector as like agentic commerce, but I think it's like misunderstanding what the objective is for the agent, which is to cover the cost of compute. Like, there's no point using an agent if it costs more to use it than the actual benefits and outputs. And you're starting to see, like, a convexity moment with what the outputs are as the compute you're using is changing. You know, Mythos was, like, really the first model that used Blackwells, the B series. Everything before that was just Hoppers, and it just took them so long to adapt to the latest compute. But now—Navier-Stokes, like, being solved by OpenAI, that was really using the GB300s and GB200s. So you're getting to the point where the computer is getting so good that you're about to, like, superpower every, like, human individual. Like, this music thing is kind of cute and nice. You're going to help with payments, and that's nice. But actually having a—True supercomputer to solve things that everyone has access to. Like, we haven't gotten there yet, and we're about to see it.

6:47那這對數位資產有什麼幫助？像 Muse 和 Zuck 推的主軸，就是要做最後一個支付應用。消費者層面，所有資料都被 AI 讀取，像我早上醒來發現 Muse 幫我讀完所有 email，才想起我有開這功能。至於現有數位資產，我覺得會出現不同類型的新資產，這波浪潮重點就是 AI 數位資產，這是必然趨勢。像 Pearl、VVV、SHIB 這些都會受益。當然也有風險，像今天早上 Justin 就發文討論密碼學未來的風險，這些都很可怕，但長線我還是看多。Now, how does that benefit digital assets? Well, Muse and the entire pitch that Zuck was pushing was it's going to be the final app for payments and all these other things. So at the consumer level, all that data, it was surprising to me when I—Wake up to Muse reading all my emails. Just like, yeah, I forgot I did that. But in terms of improving the existing digital assets, I think there will be different types of digital assets. I think this entire wave that we have more recently is like focusing on AI digital assets, which I think is inevitable. So Pearl, VVV, SHIB. Yeah. So yeah. So I think those assets benefit. I'm not sure how much. I mean, this morning too, like you saw, you know, Justin post about the future risk of cryptography. It's kind of scary stuff, but we'll see how that benefits digital assets. It's still bullish, I think.

7:42如果說加密會是第一個被攻擊的領域，現在每週 DeFi 都在被駭，但這其實是種壓力測試，最後加密基礎設施可能會比 Wells Fargo 還安全。像 Ethereum 被攻擊這麼多次，等到撐過來後，會不會變得更安全？Dovey，你怎麼看？I mean, if crypto is going to be the first one that gets attacked, everything is getting hella attacked by any hack attempt, and everything is—You know, we're having like a hacko week in DeFi for a while. But on the other side, we know we've seen this before. It just sort of creates a battle testing that it probably means that crypto rails will end up being more secure than Wells Fargo. You know, Ethereum will have been attacked so much, and is it the case that once you come through the other end, What do you think, Davi? Like, do we potentially have more secure space after we've settled in?

8:21AI 代幣經濟與全球流動性

8:21我們要先定義數位資產很廣，包括加密原生資產和 RWA（像股票、債券的代幣化包裝），這兩種都是數位資產。我認為有『人類參與』的數位資產會最受益於 AI，例如預測市場現在有三成交易量來自 agentic bot，這些還是有人類創建市場、仲裁等。至於全自動的加密原生資產，我反而比較偏空。因為現在有另一種 token 經濟（AI 詞元經濟）在爆炸性成長，幾乎每年都 10 倍、100 倍。AI 詞元經濟就是你可以看到所有 AI 資本支出（capex）流向 GPU、HBM、記憶體，產出就是詞元。這些詞元會變成未來世界的基礎商品，比石油還基礎。而且跟石油不同，AI 詞元可以各自訂價，像 Anthropic、OpenAI、ByteDance（Seedance）發的 AI 詞元，價格完全不同，還能對不同買家差別定價。現在最強的影片模型是 ByteDance 的 Seedance，中國模型的詞元轉賣到美國可以賺很大價差。AI 詞元經濟會是數兆甚至十幾兆美元規模，現在現貨加密資產大概 4 兆美元，大部分還是流動性很差的 shitcoin，Bitcoin 佔 2 兆。AI 詞元經濟才剛開始三年，還有 20 年成長空間。Yeah, so I think we first have to think about, like, the digital asset can be broad, right? And we have, like, crypto-native digital asset, and we have, like, RWA, like the tokenized wrapper of, like, equity, of bonds, and those two both are, like, digital assets. And I, so I really think that the digital asset that has human in the loop, that will actually benefit from AI. And like, for instance, like prediction market right now, there's like 30% of the volume coming from like all the agentic bots. Like those are the things that has a human in the loop, right? Because you have human to create a market, and then there's arbitrators, blah, blah, blah. But when it comes to this more like automatic...Crypto native asset that we are familiar with, I'm actually more leaning on the bullish side, so I'm more leaning on the bearish side. And first of all, now there's an alternative token economy that's growing exponentially, and then it's basically like 10x, like almost like 100x. Almost every year. When you say alternative token economy, it's like the AI token economy, right? Because the AI token right now is the actual economy that you can see all the AI capex, that the capex, like, flow into GPU, flow into HBM, flow into memory, and then, like, the output production is the token, right? And, like, token is gonna become the future world's, like, commodity. It's like the foundational commodity. And also, unlike...Crude oil, like, like, and are those tokens going to be, like, you know, with the Chinese models obviously catching up to the frontier, are those tokens going to, like, be commoditized? Yeah, yeah. So, so I will definitely touch upon that because I, if you think about, like, the crude oil, like there will be—Different type of production, like based on crude oil, right? So like there's, like, diesel, like there's different— You mean the different, like, types of degree of oil? Yeah, yeah, yeah, right. Thick oil, light oil, crude oil. Yeah, exactly. Okay. Yeah, but like pricing is pretty much standard, right? It's like pricing is universal. But like when it comes to token, like AI token, like everyone can price discriminate on their own token, right? So like the—So, like the AI token produced by Anthropic and, like the AI token produced by OpenAI, the AI token produced by ByteDance, like Seedance, and then so they all price in, like, completely different way. And also, like, they can actually price to different buyers. Like, for instance, right now, like the best video model right now is from, like, Seedance, and then, like, that is, so that is.manufacture, like this specific commodity, it is manufactured by Binance, like, which is like a Chinese company. So when all the Chinese token reseller export that Seedance token to U.S., like they can have like a huge markup, right? And so, like, the new AI token economy, and I think it will be a—Like a multi-trillion, probably tens of a trillion level, like, level of economy out there, and then so not thinking about the derivative, only the spot, right? And then so right now we have probably like four trillion, like, crypto asset kind of like the spot, and then like most of them are like completely illiquid, like shitcoin, right? So like Bitcoin is about two trillion. So, like, that's why I think, like, in the future, an AI token economy right now is only in, is like three years, and then so we, we, so we have like 20 years, like, runway of this, like, token AI token economy.

11:46AI 加速發展，一方面讓代幣生成變得無限容易，市場噪音更多，但 AI 也會幫助我們過濾雜訊、萃取重點。大多數人其實很懶，現在面試大學生時，我反而喜歡沒經歷 GPT 革命的畢業生。因為有些人一遇到問題就想直接問 GPT，缺乏抽象思考能力。我擔心未來會出現一堆『人類廢料』，大家都變成 AI 的殭屍，腦袋被取代。As AI accelerates, you know, on the one hand, you get a lot more slop because it's unlimitedly easy to generate tokens, to have things happening. But also AI will improve our ability to read through the noise because it sort of forces the technology to exist where you're just differentiating and actually being able to distill as well. So I think majority of, like, the human being are just extremely lazy, and then so if—So now when I talk with the college graduate student, like interviewing with us, and so I much prefer the college grad that has never, like, like the GPT moment is outside of, like, his, like, graduation. So, like the— Outside what? So basically, like, if you have spent your, like, two or three years of your, like, college time. Like prior to the GPT, and your mental model is completely wrecked. And so because I, so I usually ask them some very simple estimation questions, just like typical McKinsey interview, right? And then can you estimate, like, the revenue from how many golf balls in the...Yeah, yeah, right. So like that kind of stuff, right? And so I ask them to estimate, like, the revenue from, like, city security, and then so...Like the first reaction is like, can I just like GPT, right? So like no one actually have this abstract system thinking. So I think we will have a bunch of like human slop, and so like that is my concern.

13:07人類智能 vs. 機器智能與社會變遷

13:07我們是不是都在變成殭屍？大家現在是不是都用 AI 取代自己的大腦？其實我最近做了一些很爛的短影片，Dovey 跟我說這種內容太吸引人了，大家居然愛看這種爛東西。我覺得這就是未來，你只能做多機器智能、做空人類智能。這些短影片就像好萊塢小短劇，都是用模板複製來的。像加密圈的 TJR 拍 TikTok，幾百萬觀看，大家愛看就是愛看，創作者的工作就是複製有效的東西。Are we becoming zombies? Are most people using AI to replace their brain at this point? Is that, is that what we're—We're going in a little bit different direction than crypto. Just give me a few minutes. We'll go back to the crypto assets. Well, I mean, yeah, WW reached out to me because she saw me making videos slop, posting it on my feed, and she said, This is way too engaging, people. So this is sad, you know, like honestly, like seeing that people like this video because it's just so incredibly bad. And I'm like, Yeah, but this is the future. And, you know, you just have to be long machine intelligence and short human intelligence. These are like these little Hollywood-esque kind of vignettes that you've been putting out where there's a Chad version of Cartoon You. Yeah, so I learned it from, you know, previously I used to work in e-commerce, and there are these formats and these templates, and one of the interesting things you learn is that templates always do well. You know, if you just copy someone's work. And so there's this crypto trader named TJR who's posting these TikTok videos with millions of views. And I'm like, who am I to judge, you know, whether what people do and do not like? People clearly like this format. If they like it, they like it. Yeah, exactly. And your job as a creator is just to copy what works.

14:32這讓我想到昨天跟朋友 Thicky 聊天，他說生物進化到最後都會變成螃蟹，短影片內容也一樣，最後都會變成一種自然型態。現在 Gen Alpha、Gen Z 都有自己的流行語，IQ 甚至比父母還低。我完全同意，AI 時代就是要做空人類平均智力。It reminds me, I was talking yesterday, you know, Thicky, my friend, a trader, he posted the crab thesis that everything seems to evolve into crabs. If you look at evolution, everything becomes a crab. This is real. And he's saying the same thing seems to be happening with, like, short-form video content. It's sort of forcing you to just accept that this is the natural evolutionary form. Yeah, and also, like Gen Alpha nowadays, or like, I think at least at Gen Z, we have some interesting slang. Like Gen Alpha can only, like, 6A, right? And so I just never get it. And then also, if you look at the statistic, and then Gen—Like even Gen Z is the first generation that has a lower IQ score than their parents. So I totally agree with Alistair with, like, AI, then we should just say, like, just say giga short, like human intelligence.

15:30大家 IQ 變低，這對加密是利多嗎？其實 FOMO 贏家只有 3%，散戶勝率只有 3-5%。如果生育率又低、IQ 又降，散戶會越來越少。像韓國生育率只有 0.7，80 年後韓國人可能滅絕。韓國其實是各種資產流動性的來源，沒有韓國人對加密很不利。生育率跟股市、KOSPI 指數有關，股市漲生育率就回升，跌又下來。That'd be saying everyone's getting dumber, our IQs are getting lower. Surely that's bullish crypto, no? Uh, I— wait, wait, wait, okay. Is low-IQ people with money not good for crypto? Like, no, but like, but like, okay, if like the house, so like the FOMO, like the win rate on FOMO is only 3%, right? So like the FOMO, like pump fund, and I think the retail win rate is only 3 to 5%. And I think like given such a low fertility rate, and then we have like a—Decreasing IQ, and then so we just run out of, like, all the retails. We are running out of the retail. Oh, you think we're running out of, like, the retail over time? Yeah. So, like, Korean, okay, David, Korean, like, Korean's fertility rate used to be at, like, 0.7. And you guys are gonna go extinct in like 80 years. Like, in 80 years, there's not gonna be, like, there's no Koreans on Earth. Think about that. And, like, Korean is the ultimate asset liquidity for everything, for, like, equity, for crypto. Okay, that is true. Not having Koreans will be bad for crypto. That is a good point. The birth rate recovered because the stocks went up. Yeah, exactly. But now it's gone back down, and we'll see what happens. Yeah, so, like, the fertility rate is, like, correlated with, like, with the stock price, with, like, cost-pe index. Think about that.

16:48韓國、AI、加密市場的交集

16:48韓國其實很有趣，像 Hyperliquid 這個幣表現很好，主要交易量就是韓國股票，像 SK Hynix。現在 AI 與加密的重疊，就是一堆韓國人在 Hyperliquid 上用永續合約交易 SK Hynix，然後流到 Trade XYZ。三星給員工 60 萬美元獎金，因為公司賺翻了。這種豐盛時代，很多人工作被 AI 取代，賺很多錢，閒下來就看網路內容、賭博，不會去看書。現在線上體育博彩人數比看書的人還多，這種超金融化投機行為已經超越加密，變成主流現象。I think it is, like, the Korea thing is an interesting one because it's like, you know, if you look at Hyperliquid, which is obviously a coin that's done well. And you look at the actual composition of what's driving Hyperliquid's fees, it's like Korean stocks, right? It's like SK Hynix trading. And so you have this, you know, the big AI crypto overlap right now in many ways is just like a bunch of Koreans going onto Hyperliquid and trading SK Hynix with perpetual swaps. And that's going into, you know, Trade XYZ, and so it's like—The interesting sort of, you know, like, and like if you look at Korea, for example, like Samsung is paying these people like $600,000 bonuses, right? Because, like, they're making literally so much money. I think it trades like six times earnings or something ridiculous. They're just printing money. And so, you know, the abundance argument, I think, is that you get a bunch of people whose jobs kind of go away. And they're all making a lot of money. And then what do they do with their time, right? They consume online content, they gamble. They're much dumber, so they're not going to be reading books. And so it's kind of like, I think there is a statistic recently that more people have placed online sports bets this last year than people who've read a book. It is kind of insane how mainstream sports betting has become, and—That's not necessarily bullish for digital assets, even though some of that, you know, speculation, hyper-financialized speculation is obviously happening in crypto. It has become much broader than that.

18:21AI 與數位資產的資金流動與基礎設施

18:21回到數位資產主題，穩定幣經濟和 agent 的結合會持續成長。我們能不能投資或持有跟這個經濟體相關的代幣？其實現在反而是 AI 受益於數位資產，而不是數位資產受益於 AI。像 Figure 這家公司就是靠基礎設施代幣化，成為 Bloomberg 指標公司。我們自己也用代幣化降低成本，像傳統貸款買 GPU 至少要 500 萬美元起跳，規模太小根本做不了。透過代幣化，我們目標是把門檻降到 100 萬、甚至 50 萬美元，加速資本支出成長。But let's go kind of back to the thesis of digital assets, David. So obviously, I think even W would agree that the stablecoin economy with agents doing stuff is Kind of keep going up. Do we get a way to sort of invest or have tokens that are linked to that economy, or do you think it's going to be? Yeah, it's almost the opposite. Instead of digital assets benefiting from AI, it's AI benefiting from digital assets. I think it's kind of like a—Different company entirely, but I think Figure is like a good example of, like, a company that actually achieved better metrics by purely using, like, tokenization for its infrastructure. That's why it's, like, Bloomberg's top 50 in terms of metrics, just because the growth is so good. And we ourselves, I mean, we're using tokenization to reduce the cost. Like, typically to do a loan to finance a GPU. It costs a minimum of $5 million, just fixed cost. So if it's like a hundred mil deal, like, you're not going to even do it. If it's a billion dollars, yeah, it's like less than 1% of fixed cost. So tokenization for us, we're trying to reduce it down to like under a million, under 500K, just to accelerate the growth of the capex.

19:33要讓更大規模的鏈上融資發生，目前 SpaceX、OpenAI 這些公司還是用傳統債券融資，花費巨大。其實我們觀察到，現在新創雲端公司（neocloud）數量暴增，因為傳統貸款門檻太高。AI 的瓶頸不是電力或剩餘價值，而是初始融資成本。只要用代幣化抽象掉這些流程，私募信貸基金就很願意參與。我們最近推出 GLX 產品，讓私募信貸基金買這些貸款，像最近就賣出 8700 萬美元的貸款，連服務費 1% 都願意付，因為他們只想看文件，不想自己裝機。So, I mean, what is it going to take for larger scale? Financing to happen on-chain. I mean, right now, you know, even SpaceX, OpenAI, all these guys are spending an infinite amount of money, and they're just raising bonds and, you know, they're doing the traditional stuff. At what point are we able to say, look, Sam wants to raise $20 billion to do a new frontier run. I don't think you need to do bigger deals. That's not what we're seeing from our side. Like, we're seeing beginning of this year there were 300 neoclouds, and they're all big because of the threshold requirement that you have for that unit economics of doing a loan. People talk about, like, power and residual value as the bottlenecks of AI, and we haven't seen that as a common bottleneck. You can get, like, massive private credit funds that want to do it. They just don't have the inertia to overcome this initial cost. to doing that first loan. But if you obviously abstract it away, tokenization, then they're very willing. We just announced, like, our own product called GLX, where these private credit funds are buying these loans. We just sold an $87 million loan. And they just didn't want to do the work, right? They even pay people like 1% on the, like, loan servicing just so they don't have to deal with it because they're like—You know, suited up like Harvard guys, they just want to see the paper. They don't want to actually go there and install the chips for them and triple verify.

20:55至於資料中心的代幣化，現在新創雲端公司數量暴增，我們的客戶有三成是 neocloud，其他則是終端 AI 客戶，像律師事務所、賭場等。只要擁有權的單位經濟變得更有效率，未來會有上萬個小型資料中心，分布全球，直接由終端使用者擁有。這只有在單位經濟合理時才可行。現有的支付和融資結構太老舊，數位資產能提升資金流動速度，推動基礎設施建設。如果未來地球表面都變成資料中心，唯一的障礙就是融資審核，只要資本基礎打開，算力可以無限擴張。But how about for, like, a data center? Like, can we—people have talked about this. I feel like I've had people talk to me about this idea of tokenizing a data center for a while now. Yeah, I think what we're seeing now, again, from the beginning of the year, there were around 300 neoclouds. What we're seeing now is that our customer base, our inbound is about 30% neoclouds. The rest are actually just end AI customers. They're the late MM walk-ins of the world, the law firms, the casinos. We had a casino inbound ask us for a loan for fraud purposes. So the diversity, what we're seeing is that if the unit economics of, like, ownership is getting more efficient. You'll see 10,000 neoclouds, which means that in terms of, like, the size of a gigawatt, right, one gigawatt is like the energy spend of Macau or Denver. That's not very easy to do. You'll always have to syndicate that risk. But when it comes to, like, one or two megawatts, that, you know, 50 to 80 mil of, like, capex spend to just put a GPU in a powered, like—PowerShell just down the street, one building is probably a megawatt. That is going to become more commonplace. It's going to resemble closer to the RMBS market, like really having smaller clusters everywhere. Like a lot of people probably heard of a modular data center pitch kind of becoming the more commonplace thing. But the behavior we're seeing is that it's not going to be large hyperscaler deals. It's going to be 10,000 different small clusters all over the world. Directly owned by the end users. That really is only feasible, like I said, if the unit economics make sense. So there is benefits. I think we exist in such an ancient, like, payments and financing structuring. Like, there's not much technology actually being built to facilitate the money and velocity we're about to spend. That's where—Digital asset is actually useful. Like, it does actually improve the velocity at which this is being built. Like, if the dystopian future is that the surface of Earth becomes one data center, this is like the Matrix, right? And the only thing holding us back from this is underwriting. If you can actually open up the capital base, like the Amazon forest could be a data center. Very bad example, but like the—There's so much more compute that can be built if just the very human processes of moving money can be more efficient, right? And we're a case in point of that. Yeah.

23:18全球流動性、影子銀行與數位資產角色

23:18我的反駁是，其實 AI 真正受益的是全球流動性，而不是單純的數位資產。過去 15 年加密圈一直在打造巨大的影子銀行體系，像 Tether、Circle 這些穩定幣就是全球流動性的管道。AI 受益於金融後台自動化、資金流動摩擦降低。數位資產只是全球流動性的子集。舉例來說，現在一筆私募信貸基金的貸款結算要六個月，資金流動太慢，小額借貸雙方很難配對。如果能用數位資產抽象掉流程，資金流動會更快，像市場造市商一樣。So, like, Davis, on my counterargument to that is, well, okay, because our broader thesis is, like, the question is that do digital assets benefit from AI? Because you were saying that AI benefits from digital assets. It is not benefit from digital assets, but benefit from global liquidity. Because what crypto has been working on for the past 15 years is to build this huge shadow banking, right? Huge shadow banking by Tether, by, so by Circle. And, like, that shadow bank funnel the whole global liquidity. And then, like, that's one side of the coin. And the other side of the coin is AI benefit from the financial backend, so the financial backend automation, or just, like, the minimize, like, minimize, like, money movement friction. I think digital assets is a subset of global liquidity. Yeah, yeah, true, yeah. I think the benefit is that, I mean, just to simulate a single loan, like—The cost to actually settle a loan, it just takes six months for a private credit fund because money just doesn't move fast. So if you can, like, so small lender, small borrower can't get matched. But if somebody can do it really fast and then deliver to a small lender, just kind of like this, it's almost like a market maker for— Yeah.

24:33其實全球各地有很多有想法的人，過去常常有人問我能不能從 Google 出來創業，只要有三十億美元就能做大事，能不能 ICO 募資？但現實是法規很嚴格，觸碰美國使用者就會被當證券處理，還有 Gensler 這種監管機關。加上鏈上穩定幣規模還不夠大，資金流動有限。不過現在監管越來越友善，agent 抽象化一切，未來兩三年內，AI 領域的創新者只要發一篇文章，就能鏈上募資五十億美元，這其實不難想像。I mean, the more you guys are talking, I'm thinking generally that what I've seen for years is that—There's people around the world that have amazing ideas. You know, there's founders, just thinkers. Some of them, you know, are running large corporations now, obviously. But a lot of them over the years have said, Hey, you know, if I could just spin out from Google, I would be able to do this huge thing, but I would need, you know, three billion dollars. Can I do some kind of ICO or something to get it? And there's always been this need of, like, you know, capital and our resources that we have, and then people that have the actual frontier new ideas. And merging those two does unlock, like, a huge potential release of, like, human potential. And historically, you know, you have to tell them, look, like first of all, there's the whole legal thing. If you touch any U.S. customers, then automatically you're doing a security, and you have to, like, deal with people coming after you for that. And Gensler is crazy. Don't mess with that guy. And then, you know, there's obviously all these additional issues that we've had with—The digital economy is not big enough. There's not enough stablecoins on-chain to even, you know, easily fund these things. But it does feel to me, the more I hear you guys talk, that we're getting to an era where in a couple years, obviously regulation is not only itself getting friendlier, but all the abstraction with all the agents doing everything else, is it really going to be hard to imagine in like two to three years someone Credible, like one of the, you know, credible thinkers in AI or something just publishes a piece and says, you know, I want to build this thing. I need five billion dollars. We'll just open it up on chain and we'll make an asset out of this. Is that hard to imagine?

26:25地緣政治、主權化與數位資產未來

26:25加密其實一直是中國資金出海的管道，像華為、GLM 這些例子，GLM 5.3 是最熱門的開源模型，但現在被美國列入實體清單，相關晶片也被制裁。美國企業不想用中國模型，寧願用 Palantir 這種美國品牌。加密有趣的是，中國有一堆晶片和模型，其他國家又不想把資料送去中國，最後大家可能會在加密基礎設施上用匿名方式買推論服務。像馬來西亞現在有大量美中混合的資料中心，未來加密可能成為這些地緣政治算力的中立結算層。I mean, I think the interesting geopolitical dynamic, you know, crypto to some extent has always been Early, early-day crypto is a mechanism for, you know, Chinese money to find its way out of China. And Huawei and GLM are just like the very pressing examples, where ZAI is literally on the entity list. It's the most popular open-source model, GLM 5.3. It's now been obliterated, which is all the hacking stuff that we're seeing is from this thing. And then it's all running on Huawei chips, right, which are also like—I live in Puerto Rico, and if you ever touch Huawei chips, you, like, get auto-debanked. It's one of the most extremely sanctioned sort of things. And so, you know, when you look at, like, the success of Palantir, for example, in the recent market, like, they're winning in part because American corporations don't want to buy Chinese models from Chinese providers. And so they want Palantir to say, hey, it's not a Chinese model anymore. It's like a Palantir model or on-prem. I think crypto is interesting because you're like, okay, you have all these guys with all these chips. They're Chinese, running Chinese models, and then all these people who don't want to send their data to China. And we've already seen like one cycle winner, Venice, where there's like a bid for anonymity buying inference. I think that's a very believable sort of world where, like, to your point, it's like, okay, if we have a ton of Chinese chips floating around, they could train a new model. People won't want to interact with these chips, or they can't interact with these chips. They'll start buying inference from these Chinese models, potentially on crypto rails, right? Because it's like a win-win, because the guys with the chips want crypto, the people buying the inference want anonymity. And so I could imagine a world where you get some sort of, like Malaysia is a great example. Right now there's this massive capex in Malaysia because the Oracle data centers in the U.S. have been delayed. And so, you know, 60% of those Malaysian data centers are U.S. chips, and then 40% are Chinese chips. And so I think there could be, like, some sort of situation where crypto ends up being a neutral settlement layer for these sort of, like, bridging geopolitical chips into a single ecosystem.

28:37但我想反駁一下，像 Bitcoin、Ethereum 這些加密原生資產本來是超越主權的 meta-sovereign，但現在所有關鍵技術都被主權化、重新上岸。這種全球流動性、自我擁有的資產會被排擠，唯一受益的是美元。Bitcoin 創立於零利率時代，現在進入 5% 利率新常態，資金都流回美國，AI 也推動美元採用。這代表現有的 L1、L2 這些資產未必受益，反而是穩定幣和美元相關資產成為最大贏家。Well, I want to counter-argue that, because I think, like, at least the crypto-native asset that we are familiar with, and then Bitcoin, Ethereum, whatnot, and then they're actually kind of like anti-sovereign or like meta-sovereign, right? It's above sovereignty. But right now, everything has to be re-solved or, like, re-onshoring, and so everything has been, like, all these critical technology components has been, like, sovereignized. So I think that's why, like, anything that has this meta-sovereign, like, global liquidity type of, and then self-ownership type of these characters, and then will actually being, like, boxed out. And so the only thing that has been the ultimate beneficiary of this one is U.S. dollar. So, and also, like, it's actually very, very interesting that, like, Bitcoin actually, like you said, the Bitcoin creation was in this whole, like, bonds regime. It's like, okay, it's like right after, like, GFC, and then we are at this, like, close to zero interest rate. To now, and I think, like, right now we are getting into a completely new regime that, okay, 5% probably gonna be, like, the baseline. And so thinking from that perspective, and I think it's actually, like, relatively— What's the 5%? It's at 5% of the interest rate. Okay. Like, it might be the baseline because before GFC, like, 5% is actually, like, the medium. Yeah. So because we are so used to the post-GFC...I mean, the higher the interest rate, the more sort of people are really, like, getting yield, but the inflation rate is always going to be above the notional. So in the end, we're just getting deflated even faster if the neutral rate is 5%. Probably inflation is going to be something above what you're getting in treasuries. Oh, so like my point is, like, because now we're like back to this 5% regime, and then so you're seeing all the monetary flow is actually flow back to U.S. And then so like with AI, and then so we are like seeing more like U.S. dollar, so U.S. dollar adoption, right? Because U.S., it is the ultimate consumer of like all this AI token.

30:48數位資產未來型態與黃金、比特幣競爭

30:48總結一下，你們認為現有的數位資產（L1、L2 等）未必受益於 AI，反而是穩定幣經濟和支付基礎設施會持續成長，未來可能會有新型態的代幣出現，甚至變成鏈上的股權。其實 equity 這個概念才是最終型態，所有東西都會回歸到持有 equity。Bitcoin、Ethereum 這些不是 equity，更像主權債或主權 equity。Bitcoin 是貨幣資產，一直在跟黃金競爭。AI 對黃金沒什麼幫助，尤其是太空採礦、核融合等新技術出現後，黃金會變成減速資產。如果進入加速主義時代，反而是新型加密貨幣、抗量子加密貨幣、AI 代幣會有更大話題性。The general kind of acceptance is that you guys are saying existing digital assets as we know them, as we've known them for the last years, and, you know, all these L1s and all these things, probably not being helped by AI, probably the opposite. Stablecoin adoption and, like, the amount of stablecoins, payment rails going up, and then...Potentially, maybe we get, like, new tokens. We mentioned some of them that are related to the fact that those rails are so efficient and so growing, and maybe they have some economics attached to them that are accessible. I think, like, they would just ultimately become another form of equity anyway, right? So it's just like equity on chain. It's just like tokenized equity. It's just equity. Yeah, yeah. The form factor does not matter because equity is the residual value of the business, so that is effectively equity. And then when it comes to equity is the crab, like the crab, everything is just going to evolve back to holding equity, basically. Yeah, because when it comes to Bitcoin, Ethereum, it is definitely not equity, right? So, like, it is kind of like the L1s like sovereign bond, or like the L1s like sovereign equity. Well, Bitcoin is a little bit different. Obviously, Bitcoin is a monetary asset and, you know, it's always been competing with gold. Let's just kind of go a couple quick-fire topics because gold and Bitcoin competing has been this continuous ratio that we've been looking at. Gold did have a little bit of a great run for a while. Now it's stabilized. But AI probably is not helping gold, especially with all the asteroid mining. What do you think? Yeah, I mean, I think next month we're going to have a launch. I think Jed McCaleb actually backed a startup which is trying to do asteroid mining. It's failed so far, and it's launching a shuttle next month to mine platinum, and that could be a narrative point. I do think the other interesting thing is that, you know, fusion reactors potentially have gold as a side product of their generation. So, you know, Sam Altman, for example, backed Oklo, and, like, a lot of the AI executives are backing these fusion plays. And so if we enter a market, if we enter, like, a world of severe technological acceleration, it kind of goes back to the, you know, the famous bottom tick. is in, I think it was like in the tech bubble, the Bank of England sold all of its gold because they were like, why would we own this, you know, useless shiny metal if the world is going on the internet? It turns out, you know, turns out you should own the shiny metal because actually the internet was going to be kind of not as big as people thought it was going to be. So I think gold is a decelerationist asset. And if we enter an accelerationist world, there's a lot more sort of memetic steam behind, like, novel cryptocurrencies, quantum-resistant cryptocurrencies, AI cryptocurrencies, and so forth.

33:37區塊鏈、AI 經濟與民主化機會

33:37談到股權，過去私有市場的股權高度集中，IPO 前只有少數矽谷人能參與。區塊鏈和數位資產有機會讓更多人參與 AI 經濟紅利。像資料中心貸款的收益是這一代最肥的收益來源，但一般人很難參與。GPU 融資規模已經超越學生貸款、車貸，2029 年會比所有這些加起來還大。未來會有更多治理型代幣來決定資金流向，關鍵還是利率怎麼決定，這會根據資金需求和利用率自動調整。最終贏家還是穩定幣，但到底是 Circle 還是 Tether？其實所有 GPU、債務都是用美元計價，最後還是 USD 資產才有用。也許未來會有專為 AI 設計的新穩定幣出現。And how about democratization? I mean, obviously, when you're talking about equity and equity holders, it's been very concentrated, especially when it's private markets. Before things even IPO, the amount of equity holders is just—You know, a few people in Silicon Valley, usually. What do you think is going to be the potential for blockchain and digital assets to potentially give more democratized access to the benefits of, you know, this new AI economy? To talk about my book. It's hard to access the yield that's associated with the greatest source of yield of our generation, which is the yield from data centers. It's hard to access. Like, you can't get yield double digits against a GPU loan, for example, which to me is like the hungriest source of capital for this entire decade. There's nothing else that's growing as much as GPU financing. I mean, it's—This year is already bigger than student loans, and everyone has student loans. It's bigger than auto loans. Everyone has auto loans. And by 2029, it's going to be bigger than student loans, credit card loans, auto loans, and some CMBS loans. All of them put together is now smaller than the amount of money we're about to spend. Do you think there's going to be, like, much larger curation businesses around, you know? Choosing where those loans go, and yeah, yeah, that's why I have a governance token. Yeah, which one it goes to. But the bigger question is what's the interest rate, and what actually determines what the rate should be. And that kind of is a PID control, like, in that, like, depends how much money and demand there is, what the utilization needs to be. But I think what the real business kind of, like, that comes down to this, talking about stablecoin as, like, a big winner, is, like, which stablecoin. Like, is it Circle? Is it Tether? Why would an agent use Circle or Tether? And then people talk about using a non-stablecoin asset, which is ridiculous because all GPUs are priced in dollars, and all debt is priced in dollars, so it has to be a USD asset. But yeah, it's a big question. Like, will agents use which one? Maybe a new one will come out that's specifically services AI.

35:40AI 生產力預期與加密未來展望

35:40最後總結一下，如果 AI 的生產力沒有達到預期，三到五年後我們發現 AI 沒有帶來 GDP 大幅成長，反而只是讓大家更快做廢事，這對加密會有什麼影響？其實到時候所有資料中心可能又回去挖 Bitcoin。像 CoreWeave 以前就是 Ethereum 礦工，現在很多 Bitcoin 礦工都轉型做資料中心。如果資料中心經濟崩潰，美國股市會大崩盤，因為現在 GDP 成長全靠資料中心投資。其實現在生產力年增率只有 1.4%，很差。如果大家都變笨，情況可能更糟。到時候可能只有隱私幣有用，大家搬去阿根廷或烏拉圭，避開資產徵收。若 AI 不 work，最好的交易就是空 S&P 500，甚至連加密都不用買，直接買地保命。Cool, we only have a minute or two left. I guess just to conclude, if we maybe for a second think about the alternative world where maybe we've overpromised what AI is really going to do in terms of productivity. I mean, obviously the technology is transformational in some ways, but is it really going to create the kind of GDP increases that have been promised and we're investing for? If in Three to five years, we end up in a state of the world, which is possible. I think you've talked about it, Alex, that maybe we've overshot it, and actually we're not really going to get the productivity gains that, you know, we imagined because, you know, just having a faster way to do some slop doesn't actually create any productivity. How is that potentially going to look for crypto, or are we going to have A resurgence in some way, or is it going to be bearish? I think all the data centers will start mining Bitcoin after that. Yeah, that's actually the most interesting point, is that, you know, CoreWeave started out as an Ethereum miner. Well, a huge number of these sort of Bitcoin miners have pivoted into data center work. And if the data center economy blows up, it would cause just a complete meltdown in the U.S. stock market, right? Because it's like— I think it's, it's all of our GDP growth is due to data center investment at this point. And right now, actually, like the bear case is that productivity is like 1.4% annualized, which is quite bad. And if everyone's just basically frying their brains and getting dumber, like it could get worse, right? It actually, like—You might have some math advancement, and then, like, you know, we're all 20 IQ points lower, and then it just kind of balances out. In that case, I mean, I think you and I are both privacy coin aficionados. There's a world in which we're living in Argentina, you know, where the U.S. goes radical far left. And there's real asset confiscation in the U.S. because basically the Trump administration, the accelerationists, would be penalized for having participated in this giant, you know, pump. And that would be, I think, very ugly for global asset prices. So part of me is just kind of like, okay, like if AI doesn't work, just sell everything. Just like the trade is not buying crypto. The trade is like—Shorting the S&P 500. Okay, so even if AI doesn't work, don't buy crypto. Just buy some land in Uruguay or something. You got the alpha here. Thanks, guys. Thanks for having us. Thank you, Jordi.