Hack VC Alex Pack：加密 x AI 資金流，啟動新工業革命
Crypto x AI: Funding the Next Industrial Revolution
- Alex Pack Hack VC 管理合夥人
場次摘要
Hack VC 管理合夥人 Alex Pack 討論加密產業在運算網路、AI 雲端服務及建設融資中的角色。他以礦工與新型雲端公司的發展為例，說明穩定幣、DeFi 和代幣化資產可能如何支援 AI 資本需求。
值得記下的 5 件事
逐字稿
AI 與加密的交會點
好，大家好。隱私和 AI，這兩個是現今最重要的議題。我當加密創投已經很久了，但在創立 Hack VC 之前，其實我是在 Bain Capital 專注於 AI 基礎設施軟體。過去十年裡的前五年，加密一直是更酷、更有技術魅力、更性感的科技趨勢。但現在，情勢有點逆轉，全球幾乎每個人都在談 AI，AI 就像個黑洞，把全世界的資本都吸進去。很多人會覺得很可惜，這些資本應該有一部分流向加密產業。但我今天想說的是，這兩大趨勢其實不是對立的，事實上，加密一直是 AI 發展不可或缺的基礎。All right. Privacy and AI, those are the two big important things in life. I've been a crypto VC for a very long time, but actually, before founding my firm, Hack, I was focused on AI infrastructure software at a big firm called Bain Capital. And I would say for the first five years of, like, the 10 years I've been doing this, crypto was actually the cooler, more technologically exciting, kind of sexier of the two technology trends, right? And now, sadly, maybe we're kind of in a little reversal where basically everyone in the entire world is talking about AI, and AI is like this black hole that just sucks up all the capital in the world, right? And I think there is this idea, well, okay, it's sad, all this capital, some of it should go to crypto, right? But I'm here to tell you today that I don't think these two trends, megatrends, are opposed to each other. And actually, I think crypto historically has been incredibly essential and critical to the AI buildout.
比特幣礦工預見 AI 未來
我們來聊一個很有趣的故事：一群比特幣和加密礦工，早在其他公司之前就看見了 AI 的未來，並且大賺一筆，成為 AI 建設的引擎。這些礦工們其實已經先一步打造了分散式全球運算網路。AI 的基礎建設其實就是三件事：大量生產像 Nvidia、AMD 這類的晶片（這些其實都是 ASIC，專門做運算的硬體），然後把它們放進全球各地的資料中心，底層則需要龐大的電力來驅動這些晶片，進行 AI 訓練和推論。這三件事加起來，就是在打造一個全球分散式運算網路。而比特幣礦工早就這麼做了，他們已經把這個基礎建設建起來，最初規模較小，後來成為 AI 建設的試驗場，最後變成核心。And we're going to talk a little bit about a story of a very unlikely group of characters, basically these Bitcoin and crypto mining companies, who were able to see the AI future ahead of anyone else, ahead of any other companies, profited from it quite enormously, and their shareholders, and basically became the engine by which the AI buildout has occurred. And then we're going to talk about, like, what are some lessons that we can learn from this story when we're looking at what's interesting, especially as a VC, like from an investment opportunity in the crypto AI intersection from here. Okay, so the first thing to talk about with AI is actually, if you zoom out, the hype is pretty real. Like, it is in fact one of the greatest technology shifts in the history of capitalism. And the way you can measure this is basically what percentage of GDP has been spent on building out infrastructure for this technology shift. And if you see it over the course of, what is this, like 100, 200 years? Basically, all the big technology shifts that we thought were big in our lifetimes of the last 20 years, like crypto, SaaS, biotech, they're not even on the screen. Like, they're just little blips, you know? You have to go back to the internet to see something that's of comparable size, right? And still, crypto... The AI buildout is like four times bigger than the internet buildout. It's like almost ten times bigger than electrification, which is like a pretty, pretty big deal, you know. And railroads historically have been the biggest in Western world, and this is crypto—or sorry, AI has recently eclipsed that. So it is really big. It's worth paying attention to, right? My contention, though, is actually the AI buildout started a few years ahead of ChatGPT and generative AI, and it really started—you could see the origins of it with the Bitcoin mining buildout. Right, because if you think about what is the AI infrastructure buildout, like what's whatever three or four percent of GDP going to, it's three things. One, it's this huge production of chips, right, from Nvidia, AMD, that are—and these chips, they're specialized pieces of hardware, and they're actually technically ASICs, right, like specialized for doing compute. And then you put them in data centers around the world, and then underneath there's just this massive, insane amount of power, right, that's going to power these chips to produce AI training and inference, so sort of the outputs of AI. And if you think about what those three things are, you're basically building out this distributed global compute network. So. The Bitcoin miners did that first. It's quite cool, right? They already built this infrastructure, and really what happened over the last few years is they did it on a small scale, and then it became a test case for the AI buildout, and it sort of went into that and became a central for it.
礦工如何搶先卡位
那比特幣礦工到底早一步看見了什麼？他們怎麼能搶先抓住這波趨勢？Balaji（他現在也在樓上演講）有句話說，比特幣就像一顆電池。礦工們很早就意識到這點，並且善用這個趨勢。AI 其實也有點像這樣，它是一種存在於全球實體基礎設施之上的價值形式，把現實世界的資源（像能源）轉換成運算力和價值，有時甚至可以反向流動，這是一種資產金融化。礦工們很早就發現，電力才是瓶頸，所以他們蓋資料中心時，重點不是選在客戶多的地方（像 AWS、Google 那樣），而是選在電價最便宜的地方，例如加拿大北部、德州、北達科他、甚至西伯利亞。他們專注於電力、營運效率、速度、許可證，還有資料中心的互聯性，因為只要斷電，挖礦就會損失區塊，這很嚴重。這些經驗後來對 AI 資料中心來說變得極度重要。And so there's this question, like, well, what is it that the Bitcoin miners saw first, and how are they able to take advantage of this trend? There's a, who is it, I think Balaji has this quote, who's actually speaking upstairs, that Bitcoin is kind of like a battery in many ways, right? And the miners saw that early on and took advantage of that trend. And AI is sort of like this too, right? It's a battery in the sense that it's this form of value that exists on top of real-world infrastructure anywhere around the world, and it turns that real-world infrastructure, which in this case is energy, into compute. And into value, and sometimes it even goes backwards, and it's sort of fungible. So it's this financialization of the world's assets, right? And so the crypto miners saw this early, and they focused on basically power as a bottleneck when they were building their data centers. And now that's like a meme, you know, what's the latest bottleneck? Power is obviously the biggest one in the world right now. They zeroed in on that, and so instead of building their data centers where the customers were, which is what like AWS and Google did, they built them where the power was cheapest, in places like northern Canada, Texas, North Dakota, things like that, Siberia even. And they focused on power, they focused on operational efficiency, speed, permitting. Because, and then interconnectedness, this idea that you have to have like 99.9% uptime with your energy, with your internet connectivity, because otherwise if your power shuts off when you're mid-mining a block, you lose the block. It's very bad, right? And so they're able to carve this space out for themselves in this, like, data center world, right? And then that became extremely, extremely relevant for AI when it happened.
Crusoe：從比特幣礦工到 AI 雲端巨頭
舉個例子，Crusoe 這家公司我很早就投資了，現在已經是全球最大的 neocloud 之一。他們一開始就是比特幣礦工，主張去美國找閒置天然氣。十年前美國有很多天然氣因為法規和成本問題沒辦法運出去，只能現場燃燒掉。Crusoe 的想法是把需求端帶到現場，直接在產地蓋資料中心、裝滿比特幣礦機，拖到德州 Abilene 這種鳥不生蛋的地方，現場還要自己蓋小型發電廠。這個想法很瘋狂、很難做，但他們做到了，成為市值十億美元的公司，佔全球算力好幾個百分比。他們很早就投入 AI，資源有一成都在 AI。等到生成式 AI、ChatGPT 爆發後，AI 變成主力，現在 Crusoe 已經是三百億美元公司，還主導了 Stargate 這個半兆美元級的計畫，跟美國政府、OpenAI、SoftBank 合作，訓練 ChatGPT 最新模型。這些資料中心規模大到像一座中型城市，根本不能從電網拉電，必須自己蓋電廠，這就是他們的成功關鍵。So I'm going to talk about a little example. So this is a company called Crusoe that I seeded way back when, and they're one of the biggest neoclouds today. They started as a Bitcoin miner with a very similar thesis to what I just said. They were like, let's go find underutilized gas in America, and there was this. For regulatory and cost reasons, there's all these gas reserves in America, like a decade ago, that were just being underused, and you would use some of the gas, but a lot of the gas you would literally have to burn because you couldn't truck or build a pipeline out to this, like, remote place in the middle of America to access the gas. So their idea was like, well, let's bring the sources of demand for the gas on site, right? So let's build these data centers, fill them with Bitcoin mining machines, wheel them out to, like, Texas, Abilene, Texas. It's like the middle of nowhere. And then we'll also build mini power plants on site, right? And this idea of building the power plant is, like, crazy ambitious and difficult. But they had to do it, right, to mine Bitcoin cheaply. So that worked. They became a billion-dollar company. They had multiple percentage of the hash rate. But they were super big into AI. They always had like 10% of their resources focused on AI. And so when Gen AI happened, ChatGPT launched, it sort of became the main thing. Now they're a $30 billion company, and their concept of like—building a power plant on site. It's called behind the meter. This is like what SpaceX does. This is sort of the cutting edge of how you build AI data centers now. The reason being, like, AI data centers are insanely massive. So when you measure the size of these data centers, it's in gigawatts. One gigawatt is enough energy to power, like, a mid-sized American city. So, like, San Francisco is powered by one gigawatt. And then this initial site that they did, which is, which is the main, like they're the main developers for Stargate, which was this half a trillion dollar initiative from, you know, the Trump administration, OpenAI, SoftBank. So this site, which is basically where ChatGPT's latest models are trained, is like—yeah, it's a mid-sized city, right? So basically, if you want to create this much power, you can't take it from the grid. It would be like a big problem. You have to develop it yourself. And that's what they did. So it was super successful.
加密礦工成為 AI 雲端主力
而且不只 Crusoe，如果你有在看美股，會發現過去一年兩年，加密產業的上市公司表現超好，尤其是那些礦工公司，不是 Coinbase、也不是 Circle，而是這些礦工轉型做 AI 的公司。現在全球前十大 neocloud，有八家是加密起家的公司，創造了數千億美元的價值，實際上就是他們在驅動、建設 AI 產業，這真的很驚人。And yeah, it's not just Crusoe. If you guys are public market investors, you have noticed that the crypto public stonks have done incredibly well in the last year or two, and it's not really been like Coinbase. It's been—or even Circle necessarily. It's been like these mining companies, right, that have pivoted or grown into AI. And so eight out of ten of the biggest neoclouds are crypto businesses or started as crypto businesses. So they've made hundreds of billions of dollars in value, you know, to their shareholders and themselves. And they literally are powering and sort of building the AI industry, right? So it's a pretty remarkable achievement.
下一步：AI 建設的資金瓶頸
創投總是在想下一步，什麼是還沒被解決的大問題，價值就藏在這裡。neocloud 雲端產業已經建起來了，那加密還能在哪裡發揮？我現在看到最大的問題，就是這場 AI 建設、這場新工業革命，該怎麼融資？目前已經有數兆美元投入，但還不夠，還需要 1.5 兆美元，才能讓所有 AI 願景成真。但即使是已經承諾的這些資金，銀行、私募信貸市場、傳統金融體系能不能處理這麼大的資金流，還是個問號。因為規模太大、又很新、很不透明，像 SpaceX 最近要蓋資料中心，開口就要 400 億美元貸款，銀行收到的只有兩頁 memo，還要把這些錢放進退休基金，大家都覺得很瘋狂。難道我們只能相信 Elon 嗎？所以，另一個選擇就是用加密。Okay, VCs are always thinking about what's next, what's the big problem that needs to be solved, and that's kind of where the value lies, right? So after we built out this huge industry known as neoclouds, where else is crypto coming into play? Okay, the thing that I see the most as a problem right now, having seeded a lot of these neoclouds, is basically this issue of how do you actually finance this world historic buildout of AI, this, like, next industrial revolution. And so there's trillions of dollars already committed to the build out. And there's a—it's not enough. You need another 1.5 trillion to make all of our dreams a reality and power Muse and all that fun stuff. But even the trillions that are committed right now, it's very unclear whether, you know, banks, private credit markets, and traditional financial rails will be capable of like handling this amount of financing. And the reason it's such an issue is the size, for sure, but also it's very new and it's very opaque. You don't really know when you're, like, lending money to these neoclouds and even SpaceX and things like that, where exactly it's going, what are the assets that you're going to get or be able to use as collateral. What their resale value is, whether they're fungible or not. There is this, I saw just earlier today, SpaceX asked for like $40 billion in a credit financing to build some data centers, and people are, the banks are complaining because they're like, we got a two-page memo, and like we're supposed to give $40 billion and put into pension funds and—retirement accounts with this, like, two-page memo for $40 billion. Like, this is insane. Are we supposed to trust Elon? So the alternative is, you know, you could use crypto.
加密如何為 AI 建設融資
我認為加密在這裡很有機會，就像礦工讓能源金融化、推動全球資料中心競賽一樣，區塊鏈技術也能讓 AI 建設金融化。這裡有幾個我覺得最有趣的領域，很多都還在早期階段，我們是早期創投，像 Crusoe 我們投了五年才爆發，這些新東西可能也要等一陣子，但我們現在最興奮的就是這些。So I think that crypto is really well positioned as a solution here, just like the crypto miners basically allowed us to sort of financialize energy. And spur this sort of global race to create data centers all around the world and this distributed compute network. So too do I think blockchain technology is going to help AI by basically financializing the AI buildout. So these are a few of the areas that I think are the most exciting. A lot of them are very early stage. We're early stage VC. We don't—I mean, when I invested in Crusoe, it was like five years before it got big. So it might take a little while for it to play out, but this is the stuff that we find the most exciting right now.
DeFi、穩定幣與 AI 融資新模式
第一個很有趣的領域，是用 DeFi 和穩定幣來融資 GPU 和 AI 相關的 token。這裡的 token 指的是 AI 推論產生的 token。現在這類公司（像 USD.AI、Gaib）募資規模還小，只有五億美元，離 1.5 兆還差很遠，但成長空間很大。這很合理，因為 DeFi 的好處是你不用只相信 Elon 或 SpaceX，而是信任技術本身、信任鏈上透明的融資機制。透明、流動性高，資產化，這些都是關鍵。Okay, the first thing that I think is exciting is using DeFi and stablecoins for financing GPUs and also financing tokens. Tokens as in—Tokens is the name for the output of AI, right? Inference tokens. And it's small now. It's like the dollar amount that these companies like USDAI and Gaib have raised is like half a billion dollars. So it's a little bit shy of 1.5 trillion, okay? But, you know, it's a growth opportunity. And, but it makes a lot of sense, right? Because if you look at the benefits of DeFi, it goes from like, in Elon we trust, you know, like, trust me, bro, like SpaceX is cool, to you trust the technology, you trust the rails itself, right? You trust this sort of transparent financing that's happening on chain. Because it's transparent, and because it's liquid, and because it turns sort of things into assets, right? And the liquidity is very important.
運算金融商品與鏈上資產化
第二個很快會發生的，就是運算型金融商品和交易產品的大爆發。接下來一兩個月，CME 和 ICE 這兩大全球交易所都要推出運算期貨商品。Kalshi 剛推出一個市場預期曲線，把 GPU 價格、推論 token 價格的預期都聚合起來，做成一條曲線。這只是開始，接下來會有去中心化、代幣化的版本在 Hyperliquid、Uniswap、DEX 上交易。資產一旦 代幣化，就會出現各種衍生品、槓桿、包裝商品，像 2000 年代中期的房貸證券化一樣，只是這次是運算資產，而且會在鏈上爆發。當然，有人會說這會不會重演 2008 金融危機？這確實是個風險，但未來會怎樣，讓我們拭目以待。So the second thing that I think is going to happen very soon is we're going to have this mass of mass evolution or proliferation of compute-like trading products and financial products. You see this right now in the next month or two, the CME and ICE are both launching. They're the largest exchanges in the world. They're both launching crypto futures—or sorry, not crypto, compute futures products, right? And Kalshi just launched what's called a, like, market implied forward curve, which is basically they aggregated a bunch of prediction markets about what's the price of this GPU unit and what's the price of this inference and these number of tokens in the future. To get a, to get like a curve out of it. And this is just the beginning, really, because I think the next stage, once you get the centralized products, you're going to get the decentralized tokenized versions of these things. They're going to trade on Hyperliquid and Uniswap and your favorite DEX. And then from there, once you get things tokenized, things go crazy. You get—you know, synthetic versions of these things and wrapped versions, and you get levered versions. And basically, I think we're going to get this new, like what we saw in the mid-2000s with mortgage-backed securities and tranching and slicing of products. I think we're going to get that turbocharged with, like, compute-backed securities, and a lot of that is going to happen on-chain and with tokens. Now you may say, well, hold on, hold on, didn't that lead to like the 2008-09 financial crisis? That is true. That is true. We will see. There's this question of like the bubble and all of this stuff. I'm not really going to get into it. It's like a much bigger conversation. I'm just here to tell you the future, you know. We'll see. We'll see where it leads afterwards. So I think that's going to be pretty massive, and crypto is going to be a big part of it, right?
鏈上 AI：從夢想到金融化
第三個很令人興奮的領域，就是所謂鏈上 AI。這是區塊鏈對 AI 最大的願景：如果有一天 AI 的訓練和推論都能在鏈上完成，會怎樣？這還很早期，還有很多可擴展性問題，但終究會實現。像 Mistral、Venice、Gravity 這些公司，主打使用者可以有自己的私有 AI、可驗證運算、開源，這對企業很有用。但我認為，真正的外部效益是，一旦 AI 的產出能標準化、數位化、上鏈，金融市場就能圍繞它建立，甚至拿來當抵押品，這才是解決 AI 運算建設融資問題的關鍵。第一步是私有、可驗證運算，第二步就是更大規模的金融化。我們對這個未來非常期待。And then the third area that's also very exciting, and these are sort of the tokens that are, a lot of them have, you know, booths here, and you can buy them today and stuff, is this idea for, like, how do you do AI on-chain? This is the big dream, the big promise of what blockchain technology can do for AI. Like, what if you get to a world where the AI itself, which is to say training and inference, happens on-chain? It's ambitious. It's not quite here today. It takes a lot of scalability issues and things like that, but it'll get there. And the benefit you hear from these companies, things like Mistral, Venice, or Gravity, is very user-focused. It's like you could have a private version of your AI. You could have verifiable compute. That's very useful, open source. That's very useful for enterprises. That's cool, but actually I think the exogenous benefit of this for the AI buildout is like 100x greater than just the narrow benefit for users, which is once you get the actual output of AI, like digitized and standardized on-chain. Then it becomes legible to finance. You can make markets around it. You could use it as collateral, right? This gets to the core of, like, how do you solve the how do you finance the compute buildout problem, right? So it's like a two-step. Step one is, you know, private verifiable compute. Step two is part of this broader financialization story. So we're very excited.
結語：加密將驅動未來 AI 基礎建設
我認為未來非常光明，加密會繼續是這場史上最大基礎建設的關鍵。這規模真的很誇張，是世界級的歷史事件。很可惜外界很少討論加密和礦工對 AI 發展的重要性，但我相信 DeFi、鏈上金融會在未來扮演更大角色。謝謝大家！I think the future is bright. Crypto, I think, will continue to be a very critical part of the greatest infrastructure buildout we've ever seen. And yeah, it's truly insane how big it is. It's like world historic. And it's amazing to see. It doesn't get talked about enough how important crypto and these mining companies have been to it, and how I think the rest of crypto, DeFi and on-chain and the financial part are going to be going ahead into the future. So that's all I got. Thank you guys so much. Really appreciate it.
