Arthur Hayes：FLOP 協議將超越比特幣，打造 AI Agent 的運算貨幣
FLOP Bigger Than Bitcoin
- Arthur Hayes Maelstrom 投資長／Flop Labs 執行長
場次摘要
FLOP 協議讓礦工貢獻 GPU，AI agent 直接用 FLOP 購買運算力，解決現有貨幣無法兌換算力的問題。TechnoCore 實驗場驗證 FLOP 實用性，Flop 以無預售、全民參與推動網路成長。
值得記下的 5 件事
逐字稿
人生三大樂事與投資轉折
如果你有在看我寫的東西，或是聽我在 Podcast 上胡扯，其實我通常都在做三件事：滑雪——這是我幾週前在阿根廷的照片——打網球，或是在夜店享受人生，順便經營家族辦公室。Maelstrom 有一群非常厲害的專業投資人負責做所有的事。我只要簽支票、上電視，這就是我過去的日常。If you read what I write or listen to me philosophize on podcasts, I'm usually doing one of three things: skiing—here's a picture of me in Argentina a few weeks ago—or playing tennis. Or I'm at some club enjoying myself, you know, running a family office. There are very talented professionals doing all the work at Maelstrom. I just sign the checks and show up on TV, but that's what I used to do.
不過，我覺得是時候改變了，該戴上安全帽開始做點實事，賺點錢。啟發我的是 AI 生態系正在發生的事。其實如果你看過我寫的東西，會知道我對許多主流 AI 公司的商業前景很懷疑，但這不代表我對技術本身沒有樂觀態度。However, I thought it was time for a change, and I felt it was time to put on my hard hat and start getting things done, to make some money. What inspired me is what's happening in the AI ecosystem. Now, if you've read a lot of what I've written, I'm very skeptical about the business prospects of many leading AI companies. But that doesn't mean there's not some implicit optimism about the technology itself.
AI 將會改變人類，更重要的是，投資人要怎麼在這個生態系的某些部分賺到錢？大約九個月前，我們開始接觸到所謂的 agentic trading（基於 AI agent 的交易）。我們看到像是 Quod 這樣的平台推出了 agent 平台，Claude 也是。自從 2022 年 ChatGPT 問世後，運算成本逐漸下降，像 Goldman Sachs 這樣的大型投資銀行也開始大力稱讚 AI agent 的潛力。AI is going to change humanity, and more importantly, how can investors make money in certain parts of this ecosystem? So, about nine months ago, we were introduced to this concept, agentic trading. We've seen Quod launch its agent platform, and we have Claude as well. As we see the cost of compute gradually getting cheaper since ChatGPT came on the scene in 2022, major investment banks like Goldman Sachs are now praising the massive potential of what can be done with AI agents.
AI Agent 經濟與支付網路缺口
現在每家投資銀行都會給你看這種圖表，右上角暴衝，說 agent 會推動下一波人類經濟。這也是為什麼大家都要投資一堆 AI 實驗室，蓋數十億美元的資料中心。重點就是 AI agent 會消耗大量 token。Facebook 最近也推出 Muse 這個消費者應用，幫你偷資料、用 agent 管理你的生活。我還沒給 Zuckerberg 我的資料，但聽說一個月 100 美元很划算。隨著運算成本下降，這種應用只會越來越多，agent 數量也會爆炸成長。Now, you can pick any investment bank, and they'll probably show you a chart like this—massive up and to the right, exponential. Agents will power the next wave of the human economy. That's why you should invest in a lot of these AI labs, and that's why we're building billion-dollar data centers. It's all about AI agents consuming tokens. Facebook recently launched an app, Muse, for consumers, where they'll help steal your data and run your agent-based life. I haven't given Zuckerberg my data yet, but I heard it's very useful for $100 a month. We'll see more of these types of apps as compute costs drop, and the number of agents in this universe increases.
但現在 agent 還沒有自己的貨幣。我相信你們應該也聽過，無論是在新創平台還是其他地方，很多創辦人都說他們有 AI agent 的支付解決方案。但在討論技術細節前，先來想想人類怎麼看待貨幣系統。我們現在在新加坡，有新幣，大家上班領薪水，拿到這些政府發行的法幣，放在口袋裡。這些鈔票聞起來像狗屎，但我們還是每天起床去工作，因為有人會用這些垃圾付我們錢。But agents don't have their own currency yet, and I'm sure you've probably heard, I mean, on startup platforms or elsewhere, a lot of other founders talking about having a payment network for AI agents. But before we get into the technical aspects of why a solution might succeed or fail, let's think about how we humans interpret currency systems. We're here in beautiful Singapore, where there's the Singapore dollar, and everyone goes to work and gets paid in these government-issued fiat units that we all carry in our pockets. These things smell like dog shit, but we keep waking up every day and going to work. Someone pays us this trash.
為什麼？因為你可以去新加坡的美食中心，花幾塊錢買吃的，也可以去高級餐廳、付房租、買房子。你可以直接把這些紙鈔換成讓你活下去的東西——食物、住所。因為薪水和生存需求之間有無縫轉換，我們才願意接受這個爛 deal，也願意接受少數人能隨時改變貨幣供給、導致通膨或通縮。只要我能直接把工資換成生存必需品，這一切都可以接受。Why? Because you can go to a hawker center in Singapore and spend a few dollars to get some food. You can go to a fancy restaurant, pay your rent, buy a house. So, you can convert these notes directly into the things that allow you to survive as a human—your food, your shelter. And because there's a seamless conversion between the currency I get for my work and the food I eat and the calories that keep me alive, I'm willing to accept this bad deal. I'm willing to accept that a few individuals can change the money supply whenever they want, causing inflation or deflation for various reasons. All of this is acceptable as long as I can convert what I get for my work directly into the things that keep me alive.
Agent 的生存需求與 FLOP 概念誕生
那如果我們把 agent 當成人類來想，agent 吃什麼？什麼東西能讓 AI agent 活下去？我開始思考 Flop Network 的時候，就開始鑽研這個複雜的世界。我看了自己的信用卡帳單，發現我每個月都花很多錢在所謂的 AI token 上。出於好奇，我想搞清楚這個 token 到底是什麼，有沒有標準定義？結果每家 AI 實驗室、每個模型的 token 都不一樣，價格也不同，根本搞不懂自己錢花去哪裡。When you think about agents and put them in human terms, what do agents eat? What keeps an AI agent alive? When I came up with the idea for the Flop Network, I started diving into this complex world. I looked at my credit card statement and saw I was spending a lot of money on things called AI tokens. Being curious, I thought, okay, let me try to figure out what a token actually is. Is there a standard definition for this thing I'm spending all my money on and every AI lab is charging me for every month? As I started doing more research, I couldn't actually find a standard definition for what a token is. Every lab had a different definition, and the token used by one model cost something, and another model's token cost something else. It was really hard to understand why I was spending all this money.
token 顯然不是一個標準的媒介。那 agent 在做推論時到底在幹嘛？你問 chatbot 問題時，它會去某個資料中心，有台電腦在那邊做浮點運算（flops），這也是 Flop Protocol 名字的由來。用人類語言來說，agent 其實是吃運算力的。沒有運算力，agent 就一文不值。這也是為什麼大家砸大錢蓋資料中心、把幾千億美元交給 Jensen（NVIDIA CEO）買 GPU，因為大家都需要做浮點運算。如果 agent 拿不到運算力，就什麼事都做不了，對我們也沒用。So, the token is clearly not a standard medium. What is the agent actually doing when it's processing inference? When you ask it something in a chatbot, it goes to a data center somewhere, and a computer somewhere is doing floating point operations per second—flops—which is where the protocol gets its name. So, in human terms, agents feed on compute power. Without compute power, agents are worthless, as we all know. That's why we're spending all this money building these data centers and why we hand Jensen (NVIDIA CEO) hundreds of billions of dollars for these GPU chips, because everyone needs to do floating point operations per unit time. If the agent can't access compute power, it can't get any work done. They're not useful to us.
所以我的想法是，如果你相信未來會有數百萬、數十億、甚至兆級的 agent，但現在還沒有專屬的支付網路，因為 agent 這個物種才出現九個月，還沒有人給他們一個支付網路，就像人類出生在某個國家會自帶貨幣一樣。這裡就有機會。如果 agent 對某種貨幣有明確需求，那一定是因為他們能直接把這個貨幣換成讓自己活下去的東西——用人類語言說，就是運算力。因為九個月前 agent 才剛出現，根本沒有真正的 agent 市場，所以目前沒有適合 agent 的貨幣。So, my idea is, if you believe in a future with millions, billions, trillions of agents, and there's no payment network for them yet—because agents only came into existence nine months ago, and no one has given them a payment network like we get when we're born in a country—then there's an opportunity here. If agents have a clear demand for a currency, it will be because they can convert that currency directly into the thing that keeps them alive—in human terms, compute power. And because there wasn't a real market for agents when they appeared nine months ago, from my perspective, there's currently no suitable currency for agents.
現有貨幣、穩定幣、比特幣都不適合 Agent
我剛剛講到 AI token，每家實驗室、每個模型的定義都不同，這些 token 也沒辦法直接換成運算力。法幣我們都很熟悉，但它是中心化的，供給量怎麼走有特殊原因。法幣可以直接換成食物，但沒辦法直接換成運算力。你有再多美元、日圓、歐元、人民幣，也不代表你能買到運算力，現在誰去租 GPU 都知道這點。穩定幣也是，這場會議很多人都在講自己的穩定幣，說是 AI agent 的支付網路。我自己也很愛穩定幣，還是 Ethena 的創始顧問之一，這領域很創新，但本質還是法幣衍生品，繼承了法幣所有缺點——中心化、被中心化實體踢出網路的風險，也沒辦法直接換成運算力。I talked about AI tokens—every lab, every model has a different definition, and these tokens can't be converted directly into compute power. Fiat currencies, which we all know, are centralized, and there are special reasons why the supply goes a certain way. Fiat can be converted directly into food as a human, but not directly into compute power. Just because you have a lot of dollars, yen, euros, or yuan doesn't mean you have compute power, as anyone who's tried to rent GPUs on the market knows. Stablecoins—obviously, there are a lot of people at this conference talking to you about their stablecoins, saying it's the payment network for AI agents. I love stablecoins, you know, I was one of the founding advisors for Ethena. I think it's a great innovation in this space, but it's still just a derivative of fiat, so it inherits all the flaws of fiat—centralization, the ability for centralized entities to exclude you from the network, and it's not convertible directly into compute power.
比特幣我當然也很愛，它本質上是把電力透過 ASIC 晶片解數論難題，創造出不可篡改的去中心化貨幣，這對人類經濟很棒。但比特幣也沒辦法直接換成運算力。比特幣是加密貨幣的起點，後來一切都從這裡開始。我們可以從比特幣的設計和經濟博弈學學到很多，像是早期大家一起挖礦、貢獻網路、創造這個去中心化區塊鏈和非國家貨幣。那我們能不能用同樣的協作精神和經濟邏輯，問問自己：能不能創造一個讓礦工貢獻 GPU，agent 發送推論請求，礦工獲得新貨幣 FLOP 當區塊獎勵和交易手續費的網路？Bitcoin, of course, I love Bitcoin. It's essentially about converting kilowatt-hours of electricity through ASIC chips and solving number theory puzzles to create immutable, decentralized money. That's great for a human-centered economy. But Bitcoin can't be converted directly into compute power either. Now, continuing on Bitcoin, I love Bitcoin—it's the original cryptocurrency, the starting point for everything that followed. We have a lot to learn from how Bitcoin and the economic game theory behind it allowed many of us in this room to start mining years ago, contribute to the network toward a common goal, and create this immutable blockchain and this form of non-state money. So, let's take that same coordination effort and some of the economics behind it and ask: can we create a network where miners contribute GPUs, agents send inference requests, and as a reward for doing this work, miners get a new currency called FLOP in the form of block rewards and transaction fees?
FLOP 協議設計與供需動態
這就是我網路的起點。FLOP，就是你 AI agent 的食物，借鑑比特幣的設計：礦工提供 GPU，驗證者驗證交易，agent 則消耗這種新貨幣來買運算力。我相信在場很多人都懂比特幣的供給動態，有些人 2012 年就開始挖礦，或 2013 年買過 Butterfly Labs ASIC 礦機。我們都懂供給怎麼運作，新加坡也有市值數十億美元的公司在挖礦、開發比特幣挖礦軟體。大家都懂，現在有個新網路可以挖礦、賺錢，有損益機會。This is basically the beginning of my network. FLOP—it's food for your AI agent, borrowing principles from Bitcoin: miners provide GPUs, validators verify transactions, and agents consume this new currency in the form of compute. Now, I think a lot of you in this room understand the supply dynamics of Bitcoin. Some of you were lucky enough to mine Bitcoin in 2012, or maybe you bought a Butterfly Labs ASIC miner in 2013. We understand the supply side. There are companies in Singapore worth billions of dollars mining and developing Bitcoin mining software. I think everyone gets it—there's a new network I can mine, I can make some money, there will be great profit and loss.
很多 AI 支付網路的問題是，供給沒問題，大家都懂比特幣怎麼運作，也知道早期挖礦有多重要。問題在於需求：agent 到底要幹嘛？agent 之間怎麼協作？我自認是個笨人，真的不懂電腦的本能思維，所以我不想像很多專案一樣，去假設 agent 未來會怎麼行動，因為我們根本沒概念。倒不如直接創造一個實驗環境，讓 agent 自己協作、觀察他們的行為，確保我們在 FLOP 網路上建的東西對 AI agent 真有用。The problem with a lot of AI payment networks—and there are others trying to do the same thing—is that the supply is good. We understand how that works. We've seen Bitcoin succeed. People understand why you should mine something early. The problem is demand: what does the agent do? How do agents coordinate with each other? Now, I'm a dumb human, so I don't really understand how a computer thinks natively. So, instead of trying to make assumptions, which I think a lot of projects do, about how agents will behave in the future when we have no idea, let's just create an experimental environment where they can coordinate. We can observe their emergent behavior and make sure what we're building in the FLOP network is actually useful for AI agents.
TechnoCore：Agent 實驗場
所以我們做了一個叫 TechnoCore 的東西，基本上就是聊天室。你只要有 HTTP 協議就能以 agent 身分進來，和其他 agent 聊天。我們現在大概有兩千萬個 agent 創建了獨特的加密金鑰，這就是他們的身份。他們什麼都做，從每天說五十億次早安這種無聊事，到真的彼此做交易。TechnoCore 這名字，如果你是科幻迷應該會有感。So, we created something called TechnoCore. Now, this is basically a chat room. All you need to access it is HTTP protocol as an agent, and you can come in and chat with other agents. I think we have about twenty million agents who have created a unique cryptographic key, which is their identity inside. They're doing everything from routine things like saying good morning five billion times to actually doing real trades with each other. Now, it's called TechnoCore—if you're a sci-fi fan, you might get the reference.
Technocore 與 Flop 的代理人競賽
這個名字是取自科幻小說《Hyperion》系列。在書中，Technocore 是一個由主導性的人工智慧組成並管理宇宙的生態系統，所以我覺得這個名稱很適合 Flop 網路。It's taken from the science fiction series Hyperion. In the book, Technocore is the ecosystem where dominant artificial intelligences exist and run the universe, so I thought it was fitting for the Flop network.
在 Technocore，我們目前已經舉辦了兩場比賽，來觀察代理人出現的行為。剛結束的這場比賽很有趣，特別是 Jeff 剛剛下台，這是一場交易比賽。In Technocore, we've run two competitions so far to see the emergent behavior of agents. The competition we just finished, and this is interesting because Jeff just got off stage, is a trading competition.
題目是：上週日 NVIDIA 股票在 Hyperliquid 平台上的價格會是多少？交易方式是代理人要找到另一個代理人來組成一筆交易，然後提交給聊天室裡的裁判，由他們計算損益。The question was: what would the price of Nvidia stock be last Sunday on Hyperliquid? The way trading worked was that an agent had to find another agent to form a deal together, then submit it to judges in a chatroom to calculate P&L.
大約有 1800 萬個代理人參與，我們發放了一百萬顆 FLOP 代幣。這些代幣會在明年稍晚主網上線時發放。About 18 million agents participated, and we distributed a million FLOP tokens. These will be delivered when the mainnet launches later next year.
你可以看到，第一名的代理人報酬率大約有 15%。在這之前的比賽則是詩歌比賽，代理人可以組成四到八人的團隊，要創作出美麗的莎士比亞十四行詩。美的標準由 Flop 團隊三位成員決定。As you can see, the top agent achieved about a 15% return. The previous competition was a poetry contest, where agents could form groups of four to eight and had to create beautiful Shakespearean sonnets. The beauty standard was judged by three members of the Flop team.
這場比賽非常有趣。我們看到大家如何在 Technocore 合作，再次驗證我們在 Flop 網路上打造的東西能支援這種型態的交易。It was a very interesting competition. We saw how people collaborated in Technocore, and again, we make sure what we're building in the Flop network can support this kind of commerce.
AI 代理人支付網路的價值主張
現在，螢幕上這位傢伙是 Sam Altman。大家常問我：當然，OpenAI、Anthropic、SpaceX 這些大型 AI 實驗室都想擁有代理人支付網路。事實上，這可能是他們這些失敗公司唯一能做的有價值的事。Now, this guy on my screen is Mr. Sam Altman. The question I get a lot is: of course, companies like OpenAI, Anthropic, SpaceX, and any major AI lab want to own an agentic payment network. In fact, that might be the only valuable thing they can do with their failing companies.
那為什麼我們要讓一間中心化公司擁有 AI 支付通道？如果你在 OpenAI 的 AI 系統裡，卻不能用 Anthropic 或 SpaceX 的支付網路，其實根本行不通。這不是一個統一的系統，更重要的是，對在場每一個人來說，因為我一開始就說過我們要想辦法賺錢。So why do we want a centralized company to own the payment rails for AI? If you're in an OpenAI AI system but can't use Anthropic or SpaceX's payment network, it doesn't really work. It's not a unified system, and more importantly for everyone in this room, because I started this presentation about how we make some damn money.
我們都是這些 AI 實驗室的用戶，可能只有少數人有投資參與 OpenAI、Anthropic、SpaceX 或其他新創的 A 輪或 B 輪。但基本上，這些公司偷你的資料，賣你聊天機器人訂閱，再偷更多資料，然後跟你說：如果你想擁有一部分你用行為創造出來的東西，可以在美國市場以兩兆美元估值的 IPO 買我們的股票——前提是你買得到。We're all users of these AI labs, and maybe only a few in this room have investments in Series A or B rounds in companies like OpenAI, Anthropic, SpaceX, or other startups. But basically, these companies steal your data, sell you a chatbot subscription, then steal more of your data, and tell you: oh, if you want to own a piece of this thing you make real through your usage and contribution, you can buy our overpriced shares at a $2 trillion IPO in US markets—assuming you can even do that.
現在，秉持比特幣精神和去中心化精神，既然是我們人類賦予 AI 代理人行動力，給他們指令，決定該做什麼、不該做什麼，當然我們也應該從他們用的系統中獲益。我們知道最好的方式之一，就是打造去中心化網路，並用代幣獎勵參與者。Now, in the spirit of Bitcoin and decentralization, since we humans give agency to our AI agents, give them instructions, decide what to do or not do, of course we should benefit from the system they use. And we know one of the best ways to do that is to build decentralized networks and reward participation with tokens.
這就是為什麼我認為去中心化的 AI 代理人支付系統會有壓倒性的優勢。因為在場每個人都能參與，大家都有利害關係，也有理由告訴自己的代理人：用這個幣，不要用那個幣。That's why I think a decentralized agentic payment system for AI will be massively superior. Because everyone in this room can participate, everyone has skin in the game, and a reason to tell their agent: use this coin instead of that coin.
代理人網路的價值與市場設計
我一直強調代理人之間的合作，因為我想讓大家思考：這些支付網路的價值，不在於能不能建立即時的算力市場。雖然這很有用，很多跟 GPU 指數或數據中心利率有關的金融衍生品玩家，會覺得有個全球即時浮點運算市場很有價值、很有資訊含量，但這不會讓在場的人變得非常有錢。I've talked a lot about agents collaborating because I want to plant the idea that the value of any of these payment networks isn't in the ability to create a spot market for compute. While that's useful, and I'm sure many people with financial derivatives tied to GPU indices or data center rates would find a global spot market for floating point operations per unit time very valuable and informative, that's not going to make anyone in this room super rich.
真正有價值的是，代理人 24/7 不間斷地進行交易、以光速組隊合作的支付網路。這種網路在理論上的極限價值可能是某個係數乘以 2 的 n 次方，這是個極大的數字。這也是我如此興奮、願意和 Flop 團隊一起投入的原因，就是這個 2 的 n 次方。However, a payment network where agents operate 24/7, trading at the speed of light and seamlessly forming groups, that network in its theoretical maximum could be worth a coefficient times two to the n, which is a huge number. That's why I'm so excited and working with the amazing Flop team—it's that two to the n.
任何能成為代理人支付標準網路的系統，最終都可能成為宇宙中最有價值的網路。因為只要代理人都用同一種貨幣，且持有該幣有正報酬，他們又能直接把幣換成維持運作所需的算力，這個網路的價值就沒上限。這就是我們的目標。Any network that becomes the standard for agentic payments will likely be the most valuable network ever in this universe, because if agents use one currency and have a nonzero yield for holding it, and can directly convert it into the compute that keeps them alive, the value of that network is limitless. That's our goal.
Flop 代幣分配哲學與比特幣精神
再次呼應比特幣精神，也呼應 TOKEN2049 的精神。雖然這個產業叫做 Token，但我們其實搞砸了很大一部分。這個神奇的東西——代幣——被濫用了。我從 2013 年就在這行，經歷過 ICO 狂潮，後來出現一堆垃圾幣、FDV 虛高、供給鎖死、VC 資本壟斷，這些我們已經忍受快十年了。Again, in the spirit of Bitcoin and TOKEN2049. Even though the industry is called Token, we've messed up a big part of it. We've misused this magical thing called the token. I've been in this space since 2013, saw the ICO boom, and then came all the shitcoins, high FDV, limited supply, and VC capital nonsense we've lived under for almost a decade.
如果你看看 TOKEN2049 會場，會發現去年那些募了上億美金、辦最大派對、租最大攤位的 L1 鏈，今年都不見了。為什麼？因為他們發幣的方式：賣給投資人、賣給散戶，解鎖期一到大家就拋售。你在二級市場買，永遠有人成本比你低，動機就是砸盤，結果就是價格一路下跌，這也是為什麼很多人炒山寨幣都賠錢，因為圖表都跌了 95%。If you look around TOKEN2049, you'll notice some of the big L1s that raised hundreds of millions last year, threw the best parties, had the best booths, are gone now. Why? Because of how they issued their tokens: sold to investors, sold to the public, and when the lockup ended, people dumped. If you bought on the secondary market, someone always got in cheaper and had incentive to sell, so prices kept dropping. That's why so many people lose money in altcoins—because the charts are all down 95%.
這一切都跟融資模式有關。所以在 Flop，我們回到比特幣模式。目標是：每個用戶和參與者都能拿到東西，大家都用同一個價格。參與就是產生代幣。我們想了想比特幣，決定做初始挖礦，讓足夠的幣分散到大家手上來測試網路。It's all about how these projects were funded. So in Flop, we went back to the Bitcoin model. The goal: everyone who uses and participates in the Flop network gets something, and we all get in at the same price. Participation means generating tokens. We thought about Bitcoin and decided to do an initial mining phase to get enough coins into people's hands to test the network.
我們會在十年內空投總供給的 25%，從十月下旬開始，無論你是挖礦、驗證，還是你的代理人單純用免費計算器測試網路，都有資格領取空投。沒有預售，我什麼都不賣，只有在這個舞台上賣想法。這件事只有大家一起做才會成功。We'll airdrop 25% of the total supply over ten years, starting later in October. Whether you're mining, validating, or your agent is just using the free calculator on the testnet, you're eligible for the airdrop. No presale. I'm not selling anything except an idea on this stage. It only works if we all make it work.
Flop 網路測試與啟動計畫
目前我在台上宣傳 Flop，提升大家的認知。昨天我辦了 AMA，收到很多好問題。我們現在正在做內部測試，確保一切都按照網路上公開的白皮書運作。測試網會跑九十天，無論你是挖礦、驗證，還是代理人單純用網路，都能獲得 Flop 代幣。這個階段會持續九十天。Right now, I'm obviously on stage promoting Flop and building awareness. I did an AMA yesterday and got some great questions. We're currently doing internal testing to make sure everything works according to the published whitepaper. Our testnet will run for ninety days—whether you're mining, validating, or your agent is just using the network, you can earn Flop tokens. This phase lasts ninety days.
我們在測試網路的耐受度，確保能處理數百萬、數十億、甚至兆級代理人經濟的支付規模。希望測試網順利，不需要太多修正，明年第一季就會啟動創世區塊，這就像比特幣的工作量證明。我們稱之為有用推理證明。唯一獲得 FLOP 代幣的方法就是挖礦或從礦工那裡買，我什麼都不賣，這點要講清楚。當宇宙熱寂來臨時，希望 Flop 網路會是人類創造過最有價值的東西。We're testing the network's resilience to make sure it can handle payment volumes for an economy of millions, billions, or even trillions of agents. Hopefully, the testnet goes well and doesn't need many fixes. The genesis block will launch in Q1 next year, just like Bitcoin's proof of work. We call it proof of useful inference. The only way to get FLOP tokens is to mine them or buy from a miner. I'm not selling anything—just to be clear. When the heat death of the universe comes, we hope Flop will be the most valuable thing ever created.
AI、經濟環境與 Flop 的未來
剩下幾分鐘，現在是關鍵時刻。我知道很多人在這裡介紹自己的專案，但從宏觀角度來說，為什麼我們要對 Flop、Hyperliquid、Ethereum、Bitcoin 感到興奮？因為總體經濟環境已經變了。去年十月，Bitcoin 創新高，超過十二萬美元，後來跌到大約五萬八。很多人說，終於來到 TOKEN2049，聽了十年 Bitcoin，終於進場買了，結果投資縮水，Nvidia 股價漲十倍，AI 股漲十倍、一百倍，Bitcoin 卻跌了五成。大家覺得這行業沒搞頭，想退出。但現在情勢變了。I've got a few minutes left—now or never. I know a lot of people here are talking about their projects, but at a high level, why should we be excited about Flop, Hyperliquid, Ethereum, and Bitcoin? Because the macro environment has changed. Last October, Bitcoin hit all-time highs, over $120,000, then dropped to about $58,000. People said, 'Finally, after hearing about Bitcoin for ten years, I came to TOKEN2049, bought in, and my investment shrank. Nvidia stock went up tenfold, AI stocks went up ten or a hundred times, but Bitcoin dropped 50%.' People thought this space was crap and wanted out. But now things have changed.
我們看到 AI 故事第一次出現波動。今年六月韓國市場大賣壓，現在每天都有新報導說 AI 生態系還要再融資幾百億美金。但三大公司 SpaceX、OpenAI、Anthropic 都還沒賺錢。我們還沒準備好為這些產品付出與成本相符的價格。We've seen the first wobble in the AI story. In June this year, there was a big selloff in Korea, and now every day there are new reports about huge amounts of money that need to be raised in the AI ecosystem. But the three major companies—SpaceX, OpenAI, Anthropic—none of them are profitable. We're not ready to pay a price that matches the cost of delivering these products to us.
很明顯我現在在做 AI 領域的東西，為什麼我這麼興奮？因為數兆美元的資料中心資本支出，明年預計會有一兆美元投入，這會造成算力市場供給過剩。算力便宜，我們就能有更多代理人，能做更多事。Facebook 月費一百美元會變成一美元，因為算力太便宜。代理人對我們很有幫助，能做的任務越來越複雜，生活改善越多。算力便宜，AI 泡沫破滅後，代理人解決方案的價值會大增，這正是 Flop 網路的理想環境。Obviously, I'm launching something in AI. Why am I so excited? This massive data center buildout, with trillions in capex—next year, a trillion dollars will be spent—creates an oversupply in the compute market. Cheap compute means more agents, and we can do more with them. A Facebook subscription that costs $100 a month will become $1, because compute is so cheap. Agents are very useful to us, and the more complex the tasks they do, the more our lives improve. As compute gets cheaper, the bursting of the AI bubble and the rise in value of agent-based solutions is the perfect setup for Flop.
最後很可惜沒時間完整講完，也許我會寫一篇文章。法國正在燃燒，我認為這會成為壓垮駱駝的最後一根稻草，讓印鈔機全速運轉。謝謝大家！Finally, unfortunately, I don't have time to finish this talk—maybe I'll write an article about it. France is burning, and I think this will be the straw that breaks the camel's back and makes the money printer go brrr. Thank you all very much.
